Machine Learning for Promoting Environmental Sustainability in Ports

Meead Mansoursamaei; Mahmoud Moradi; Rosa G. González-Ramírez; Eduardo Lalla-Ruiz

doi:10.1155/2023/2144733

Machine Learning for Promoting Environmental Sustainability in Ports

Mansoursamaei, Meead;Moradi, Mahmoud;González-Ramírez, Rosa G.;Lalla-Ruiz, Eduardo 2023-03-03 00:00:00 Hindawi Journal of Advanced Transportation Volume 2023, Article ID 2144733, 17 pages https://doi.org/10.1155/2023/2144733 Research Article Machine Learning for Promoting Environmental Sustainability in Ports 1,2 1 3 Meead Mansoursamaei , Mahmoud Moradi , Rosa G. Gonza ´ lez-Ram´ırez , and Eduardo Lalla-Ruiz University of Guilan, Rasht, Iran University of Twente, Enschede, Netherlands Facultad de Ingenier´ıa y Ciencias Aplicadas, Universidad de los Andes, Santiago, Chile Correspondence should be addressed to Meead Mansoursamaei; m.mansoursamaei@utwente.nl Received 19 May 2022; Revised 12 September 2022; Accepted 10 February 2023; Published 3 March 2023 Academic Editor: Dongjoo Park Copyright © 2023 Meead Mansoursamaei et al. Tis is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Maritime transportation is one of the essential drivers of the global economy as it enables both lower transportation costs and intermodal operations across multiple forms of transportation. Maritime ports are essential interfaces that support cargo handling between sea and hinterland transportation. Besides, in this area, environmental protection is becoming extremely important. Global warming, air pollution, and greenhouse gas emissions are all having a detrimental infuence on the environment and will most likely continue to do so for future generations. Hence, there is a growing need to promote environmental sustainability in maritime-based transportation. Te application of machine learning (ML), as one of the main subdomains of artifcial intelligence (AI), can be considered a component within the process of digital transformation to advance green activities in maritime port logistics. Tus, this article presents the results of a systematic literature review of the recent literature on machine learning for promoting environmentally sustainable maritime ports. It collects and analyses the articles whose contributions lie in the interplay between three main dimensions, i.e., machine learning, port-related operations, and environmental sustainability. Troughout a review protocol, this research is constituted on the major focuses of impact, problems, and techniques to discern the current state of the art as well as research directions. Te research fndings indicate that the articles using polynomial regression models are dominant in the literature, and the recurrent neural network (RNN) and long short-term memory (LSTM) are the most recent approaches. Moreover, in terms of environmental sustainability, emissions and energy consumption are the most studied problems. mAccording to the research gaps observed in the review, two broad directions for future research are identifed: (i) altering attention on a greater diversity of machine learning approaches for promoting environmental sustainability in ports and (ii) leveraging new outlooks to perform more green practical works on port-related operations. growing demands of the logistics and transportation sectors 1. Introduction [3]. Expanding maritime transportation activities has enabled Maritime freight transportation is one of the vital drivers of the urban economies to prosper to some extent; however, they also global economy as it enables both lower transportation costs have caused resource waste and environmental pollution. To and faster intermodal operations across multiple forms of achieve the sustainable growth of ports and cities, energy- transportation [1, 2]. Indeed, maritime ports are the essential saving and emission-reduction methods have to be used [4, 5]. interfaces that support cargo handling between sea and hin- Te numerous sources and types of port-related emissions, terland. Besides, environmental sustainability has become one such as those from maritime vessels, trucks, and cargo- of the important foundations on the agenda of many maritime handling machines, have a considerable impact on the envi- ports due to the challenges of climate change as well as the ronment [6]. Moreover, a signifcant proportion of emissions 2 Journal of Advanced Transportation reviews. However, port sustainability in terms of environ- in ports are due to interterminal transportation and container drayage operations [7]. Terefore, the research on environ- mental impacts is currently experiencing new challenges; hence, the literature needs to be updated. In addition to the mental enhancements in maritime ports as main trans- portation nodes has grown in importance as they allow for previous reviews, the studies that have used machine improvements in a variety of areas of environmental sus- learning approaches as solutions to cut down on port-related tainability that governments and entrepreneurs are attempting pollution have been considered in this research. Concerns to address [3]. among stakeholders, including port ofcials, policymakers, Artifcial intelligence (AI) approaches and subdisciplines users, and citizens have grown over the environmental can be divided into two main categories: (i) reasoning and sustainability of ports and machine learning techniques as technological solutions. Terefore, to direct the future decision making and (ii) learning [8]. While the frst set is focused on decision making concerning planning, solution growth of this topic, a timely literature review is essential. Te article is organized into four sections. Section 2 searching, and optimization, the second set relates to learning, communication, and perception. In this sense, presents the research methodology by describing the data collection and analysis methods along with initial statistics learning refers to the ability of systems to automatically learn, determine, predict, adapt, and react to changes regarding recent publication trends in ML applications in without being explicitly programmed [9]. Te techniques green-port operations. Section 3 presents the initial bib- related to learning include neural networks, deep learning, liometric analysis. Ten, relying on the bibliometric analysis, reinforcement learning, and decision trees, among others the evolution of ML in port logistics is studied. First, the [8, 10], and the three main subdomains are supervised, evolving process of ML applied to aid sustainable maritime unsupervised, and reinforcement learning [11]. port operations based on the growing volume of publications over time is examined. Second, the impact and citation In this research, the application of ML as one of the main subdomains of AI to advance green activities in port logistics patterns to characterize the selected articles are presented. Ten, the most cited articles and researchers with a higher is studied. A supervised learning model as a subdomain of ML can learn and check the plausibility of schedules and number of contributions within the scope of the review up to th predict the energy consumption of battery-electricauto- October 2021 are presented. Finally, the evolution and guided vehicles (AGVs) in horizontal transportation areas interaction of topics, techniques, and problems are scruti- [12]. Neural networks are being used in the ports of nized. Te article concludes with a discussion on research Hamburg, Rotterdam, and Singapore. For instance, Ham- limitations and potential research directions. burger Hafen und Logistik AG (HHLA) has become one of the frst ports in the world to implement ML solutions for its 2. Research Methodology Hamburg container terminals to predict the dwell time of a container at the yard of the terminal [13]. Regarding Tere are several reporting guidelines to perform a sys- environmental concerns and emissions from maritime tematic literature review [16–18]. Such reporting guidelines, trafcs at the port of Rotterdam, machine learning tech- e.g., SRQR (standards for reporting qualitative research), niques have been developed to predict the estimated time of ENTEQ (enhancing transparency in reporting the synthesis arrival (ETA) for containerships to reduce waiting time and of qualitative research), eMERGe (meta-ethnography thus emissions [14]. Te SAFER project of the maritime and reporting guidelines), and RAMESES (realist and meta- port authority (MPA) of Singapore and IBM have imple- narrative evidence syntheses: evolving standards) are sug- mented seven ML-based modules within Singapore port gested for qualitative reviews [19]. Likewise, the PRISMA waters in order to predict arrival times of vessels and po- (preferred reporting items for systematic reviews and meta- tential congestion areas as well as to detect ofending ships to analyses) proposes a standardized method for ensuring environmental regulations [15]. Terefore, an ever- transparent and thorough systematic review reporting [20]. increasing demand for ML-based technologies to support PRISMA is a systematic review protocol that includes a 27- green ports is arising. Besides, several relevant works have item checklist and a four-phase fow diagram [21] (see been published recently in journals and proceedings. Hence, Figure 1). In this study, PRISMA was selected as the a systematic literature review (SLR) that analyses the articles methodology for the systematic review over other existing investigating green activities in maritime ports using ML protocols because of its comprehensiveness, its use in applications becomes necessary. a variety of disciplines, and its potential to improve con- Tis study with the purpose of investigating the appli- sistency across reviews. Tis protocol involves (i) the def- cations of machine learning techniques to stimulate envi- nition of research questions and (ii) the identifcation of the ronmental actions in maritime ports seeks to explore the search string and source selection. state-of-the-art research within the interplay of machine learning, environmental sustainability, and maritime ports. Tus, the current research contributes to the literature by 2.1. Defnition of Research Questions. Te frst step of this identifying research categories based on a systematic liter- review is the defnition of the research questions (RQs) ature review (SLR), i.e., impacts, problems, and techniques. related to the main research question of this work, i.e., how Most recently, the sustainability of ports is receiving machine learning has been used to promote environmentally more attention. Te number of research on the topic has sustainable port operations. Tis way, the following research increasingly expanded and been published in some literature questions are addressed: Journal of Advanced Transportation 3 Systematic Search (Databases): Full-Text Scopus (482) Screening + Web of Science (231) (62 papers) (i) Title (i) Full-text Context Final Result (ii) Abstract (ii) Focus on research of Search: (iii) Keywords questions (29 papers) Abstract Screening Snowball Technique (713 papers) (28 papers) (i) Abstract (i) Exploring forward and (ii) Focus on machine learning backward references of and environmental the selected papers sustainability in maritime (ii) Internal loop port logistics (iii) Removing duplicates Figure 1: Te overall procedure for selecting and fltering the articles [21]. (i) RQ1. What is the environmental impact of ML snowballing technique was applied to all the selected articles approaches on ports operations? (Impact related) to minimize the risk of missing relevant studies [22]. Tis resulted in addition of one more article leading to a total of (ii) RQ2. Which environment-related port operations 29 articles that were analyzed in this SLR. Te procedure was have ML techniques being used for? (Problem carried out by two of the authors of this manuscript who related) independently screened the selected articles. Finally, there (iii) RQ3. How ML techniques have been used to pro- was a consensus among all authors about the articles that mote environmentally sustainable port operations? were excluded [23]. Figure 1 indicates a fow diagram of the (Technique related) search procedure and the results obtained at each stage. 2.2. Identifcation of Search String and Source Selection. 3. Discussion and Results To achieve the fnal result of the search, several steps in- cluding the systematic search, abstract screening, full-text In order to answer the research questions raised in Section screening, and snowball technique were applied. Te Scopus 2.1, three subsections, each corresponding to a research and Web of Science (WoS) databases were utilized to collect question, are provided. First, the impact of selected articles is and select related references systematically. Te query used discussed, and the most cited works and authors are listed. in this work is composed of a list of keywords distributed Second, based on the physical area of port-related opera- into three main pillars (i.e., machine learning, maritime tions, the environmental problems of the selected articles are ports, and environmental sustainability) as shown in Fig- identifed and discussed. Tird, the machine learning-based ure 2. Te AND operator delimited each group, while the OR techniques that have been proposed in the literature to operator bundled all keywords within each group. Te AND address the environmental problems are outlined. NOToperator eliminated irrelevant subject areas. After that, the query was used in the abovementioned databases to fnd those articles considering the selected keywords within their 3.1. Impact Analysis (RQ1). With the aim of answering RQ1 titles, abstracts, or keywords from 2017 to 2021. By the initial involving impact, the number of works published so far, the search, 482 articles from Scopus and 231 articles from WoS number of citations, and data collection methods are were discovered from the databases; however, English ar- investigated. ticles were included as a limitation and irrelevant subject Te number of articles published annually from 2017 to areas such as agriculture, medicine, and chemical engi- 2021 by the source type is shown in Figure 3. Based on the neering were excluded. Moreover, since this review is fo- data, it is observed that the majority of works (25 articles) cused on ports, maritime shipping-related keywords were were published in journals, while the other 4 articles were th excluded. Te last search was run on 15 October, 2021. published in conference proceedings. Te Journal of Cleaner Production with four articles and Transportation Research Information of articles including the title, abstract, publication year, and sources were screened to determine Part D with three articles are the journals where more of the whether a article is included or not. Based on the three pillars selected articles were published, while the other works (76%) of our scope, there was one reason to discard articles, are from various sources. th namely, the articles whose contribution did not lie within the To illustrate the citation rate up to 15 October, 2021, interplay between machine learning, maritime port opera- Figure 4 displays a histogram. It can be observed that 25 of tions, and sustainability were removed. After the abstract the 29 collected articles have been cited so far. Te most cited and full-text screening, a total of 28 articles were collected. article, [24] with 44 citations was published in the Transport During the selection process, the forward and backward Policy. In this article, the performance of 17 ports in China 4 Journal of Advanced Transportation Maritime ports logistics (i) ("maritime" AND ("logistic*" OR "transport*" OR "port")) OR (("container" OR "cargo" OR "roro") AND ("terminal" OR "port" OR "harbor" OR "yard" OR "staking" OR "dispatching" OR "drayage" OR "railhead" OR "truck" OR "interchange area" OR "freight station")) OR ("yard" AND ("terminal" OR "operations" OR "storage" OR "truck" OR "management" OR "area")) OR "inter-terminal transport" OR "intermodal drayage" OR "seaport" OR "sea port" OR "bulk port" OR "landside" OR "seaside" OR "berth*" OR "crane" OR "stowage planning" OR "hinterland" OR "internal vehicles" OR "straddle carrier" OR "gantry carrier" OR "terminal gate" OR "gate area" OR "quay" OR "transport area" OR "horizontal transport system" OR "truck area" OR "train area" OR "transport area" OR "gate operations" OR "container storage" Machine learning (i) "machine learning" OR "active learning" OR "feature extraction" OR "adaptive learning" OR "generative adversarial network" OR "adversarial machine learning" OR "generative model" OR "adversarial network" OR "multi-task learning" OR "anomaly detection" OR "neural network" OR "artifcial neural network" OR "pattern recognition" OR "automated machine learning" OR "probabilistic learning" OR "automatic classifcation" OR "probabilistic model" OR "automatic recognition" OR "recommender system" OR "bagging" OR "recurrent neural network" OR "bayesian modelling" OR "recursive neural network" OR "gaussian process" OR "reinforcement learning" OR "classifcation" OR "semi-supervised learning" OR "clustering" OR "statistical learning" OR "collaborative fltering" OR "statistical relational learning" OR "content-based fltering" OR "supervised learning" OR "unsupervised learning" OR "convolutional neural network" OR "support vector machine" OR "data mining" OR (data?driven) OR "data pre-processing" OR "transfer learning" OR "deep learning" OR "unstructured data" OR "deep neural network" OR "ensemble method" OR "regression" OR "dimen* reduc*" OR "K-Means" OR "principal component analysis" OR "kernel" OR "decision tree" OR "K-Nearest" OR "random forest" OR "Q-Learning" OR "proximal policy" Environmental sustainability (i) "green" OR "emission*" OR "sustainab*" OR "environment*" OR "noise" OR "waste" OR "energy*" OR "pollution" OR "biofuel*" OR "biomass" OR "battery" OR "decarboni?ation" OR "greenhouse" OR "renewable" OR "solar" OR "recycle" OR "carbon" OR "footprint" OR "contamination" OR "ecolog*" OR "spill" OR "electr*" OR "smart grid*" AND NOT (i) "fsher*" OR "algae" OR "trout" OR "salmon" OR "mussel" OR "molecular" OR "herring" OR "larva" OR "oyster" OR ("sea AND bass'') OR "aerosol" OR "atmospheric chemistry" OR "atmospheric deposition" OR "troposphere" OR "sewage" OR "ozone" OR "microb*" OR "biology" OR "organism" OR "animal" OR "plants" OR "electron" OR "species" OR "eutrophication" OR "odor" OR "house" OR "building" OR "retinal" OR "steel" OR "fshing" OR "ore" Figure 2: Te used keywords based on the three pillars: maritime ports, machine learning, and environmental sustainability. 2017 2018 2019 2020 2021 Journal Conference Figure 3: Te number of articles published annually from 2017 to 2021 by source. under environmental concerns using regression models was observations, measurements, and experiments as primary evaluated. data sources. Nonetheless, three of the articles (10%) utilized In terms of data collection methods and data sources, 21 interviews and surveys solely or as a part of their data out of 29 articles (72%) used secondary data sources from collection method. earlier works, research institutes, or governments. However, Te environmental indicators used in port logistics in 7 of the articles (24%), researchers gathered data by applications include emissions, water pollution, noise Frequency Journal of Advanced Transportation 5 7 77 33 3 2 2 2 th Figure 4: Citations of the selected articles (the last update was on 15 October, 2021). pollution, solid waste, energy-saving, and renewable energy evaluation and energy-saving expenditures. Moreover, the were reported in the collected articles. Terefore, among authors of reference [27] evaluated the environmental ef- them, 25 articles (86%) raised emissions, 13 articles (45%) fciency of the Kaohsiung container port in Taiwan, con- considered energy-saving, and 6 articles (21%) used water sidering the same problems. In the same way, the authors of reference [28] benchmarked the top 10 ports in the US and pollution as the environmental indicator of their work. Nonetheless, noise pollution, renewable energy, and solid considered water pollution and emissions within port areas. Furthermore, the authors of reference [29] benchmarked waste, each with only one article (3%), are the smallest used environmental indicators. Figure 5 presents the contribution operating practices of 20 ports in the US for enhancing of the selected articles based on environmental sustainability environmental efciency in terms of greenhouse gas emis- indicators. sions, oil spill prevention, and energy efciency of port operations. Similarly, the authors of reference [30] bench- marked 24 container ports in Europe regarding emissions 3.2. Port-Related Problems (RQ2). To answer RQ2, the works and energy-saving problems. were categorized in terms of the port-related operations and To evaluate ports environmental efciency in emission their application areas in the port (i.e., seaside, yard, control areas (ECAs) (emission control areas (ECAs), as landside, and overall port areas). Te specifc machine outlined by Annex VI of the 1997 MARPOL protocol, are sea learning techniques of those works are later discussed in areas where regulations have been implemented to prevent Section 3.3. Terefore, based on the area within ports, the emissions from ships), 23 ports in the Baltic and the North articles have been distributed according to the location of the Sea and 25 non-ECA ports in Europe were investigated by addressed problems, which resulted in seaside operations [31] to examine the impact of ECA regulations on reducing with 28%, yard operations with 12%, and landside opera- emissions in European ports. Furthermore, using an In- tions with 16%. Nonetheless, 44% of the articles have tergovernmental Panel on Climate Change (IPCC) method, considered the environmental sustainability indicators for the authors of reference [32] measured CO emissions from benchmarking, performance evaluation, and air quality the port container distribution (PCD) to evaluate the sus- prediction not being specifcally focused on an area but the tainable development ability of 30 ports in China. Likewise, overall port. Hence, these contributions are categorized as the authors of reference [33] predicted air quality, fne the “overall port” in Figure 6. particulate composition, and mass in the area of Long Beach port in California. In addition, the authors of reference 3.2.1. Overall Port Areas. Several articles benchmarked and [34, 35] predicted air quality and emissions in 4 ports in Turkey (Ambarli, Izmir, Mersin, and Kocaeli ports) and evaluated ports in terms of environmental efciency and did Busan Port in Korea, respectively. Moreover, the authors of not consider a specifc area within port (i.e., seaside, yard, reference [36] simulated the indoor air quality of roll-on/ and landside), the reason for which they are classifed as roll-of (RORO) ships and predicted pollution emitted from “overall port areas.” In this regard, the authors of reference cars in maritime ports. Furthermore, the authors of refer- [24] evaluated 17 Chinese ports in terms of NOx emissions ence [37] developed a container terminal logistics general- and energy savings. Similarly, the authors of reference [25] ized computing architecture (CTL-GCA) for planning, investigated those environmental problems in 18 ports of scheduling and decision making to establish a better con- China. Using a diferent machine learning technique, the nection among liners and rubber-tired gantry cranes authors of reference [26] benchmarked 15 seaports in China (RTGCs) and block community to reduce carbon emissions. in terms of wasted water treatment as well as air quality Frequency Sun et al., 2017 Heilig et al., 2017 Cheon et al., 2017 Hill & Böse, 2017 Chang et al., 2018 Alasali et al., 2018 Jahangiri et al., 2018 Wang et al., 2018 Park et al., 2019 Goldsworthy et al., 2019 Song et al., 2019 Peng et al., 2020 Wang et al., 2020 Wang et al., 2020 Kuo & Lin, 2020 Nastasi et al., 2020 Caballini et al., 2020 Quintano et al., 2020 Agamy et al., 2020 Cammin et al., 2020 Eatough et al., 2020 Holly et al., 2020 Zhoa et al., 2020 Wen et al., 2021 Wang et al., 2021 6 Journal of Advanced Transportation 86% 45% 21% 3% 3% 3% Emissions Energy saving Water pollution Noise pollution Renewable energy Solid waste Environmental Indicators Figure 5: ML-based articles in terms of environmental sustainability in maritime port logistics. Landside 16% Yard 12% Overall port 44% Seaside 28% Overall port Seaside Yard Landside Figure 6: Distribution of selected articles based on application areas within the port. 3.2.2. Seaside Area. Te seaside operations that received renewable energy. For instance, the authors of reference [12] more attention are those regarding berth allocation plan- presented a study on how to use battery-electricauto-guided ning. For instance, to manage real-time data and air vehicles (AGVs) in the yard for handling containers in the emissions reduction in maritime ports due to the berth port of Hamburg (Germany). Tey utilized a synthetic case operations, [38] developed a predictive system for vessel by generating data for checking the plausibility of schedules arrivals, considering ship features and expanding estimated and predicting energy consumption. Container cranes are time of arrival (ETA) features to date, time, and weekday, also one of the main sources of energy consumption and based on the previous model presented by the authors of pollution in the yard. In this regard, the authors of references reference [39]. Te authors of reference [40] simulated berth [46, 47], and [48] considered environmental problems (i.e., planning problems and predicted the arrival time of vessels energy consumption and emissions) of the rubber-tyred using machine learning techniques. In the same way, the gantry (RTG) in the port of Felixstowe (UK) and port of authors of references [41–43] and [44] used case studies to Casablanca (Morocco) and a synthetic case, respectively. solve berth planning problems at diferent ports that are presented in detail in Table 1. From the other perspective, 3.2.4. Landside Area. Trucks are the main source of emis- noise emissions by ships around the port areas are one of the sions in the landside area. Hence, several studies have paid important issues for port cities. Tis has been studied in [45] attention to the environmental problems caused by trucks in where the authors with machine learning techniques the area. For instance, the authors of reference [49] proposed identifed the afecting parameters of noise emitted by ships a forecasting engine for truck arrivals to logistics nodes, i.e., in the industrial port of Livorno, Italy. empty container depots, packing facilities, or terminals, to mitigate greenhouse gas emissions from truck congestions 3.2.3. Yard Area. Several contributions related to container beyond the gates of an empty container depot in northern and cargo handling in the yard area to mitigate greenhouse Germany. Based on the proposed model, companies can gas emissions, energy-saving, and promoting the use of adjust their route planning to minimize truck waiting times. Percentage of use (%) Journal of Advanced Transportation 7 Table 1: Problems and ML techniques for promoting environmental sustainability in maritime port logistics. Areas Environmental problems Machine learning Data Research Case Articles collection Overall Water Noise Energy Renewable Solid Techniques scope studies Seaside Yard Landside Emission Input Output methods port pollution pollution saving energy waste (tools) Port performance Port assets, berth quantity, and Net proft, cargo throughput, and Secondary [24] ● ● PR 17 Chinese ports evaluation geographical location NOx emissions data sources A smoothed graph for the Port performance Facility, vessel and other pollution Secondary [28] ● ● ● KDE distribution of pollution incidents 10 American ports evaluation incidents data sources probability density Historical data, truck arrival time, administrative waiting start and end time; intermediate waiting start and Waiting time, arrival rates that An empty container end time; node-specifc forecasting Secondary [49] ● Truck scheduling ● NN (BP) translates into a reduction of trafc depot in Northern parameters, e.g., dispatching modes data sources congestion and air pollution Germany and storage policies; and external forecasting parameters, weather and trafc information Number of clusters and the archive Secondary Port of Hamburg [50] ● Truck scheduling ● K-means Cluster centroids containing n solutions data sources (Germany) Berth length, the number of cranes, terminal area for the efciency Port performance TEUs handled and the impact of Secondary [31] ● ● estimation. City gross domestic PR 48 ports in Europe evaluation emissions control regulations data sources product, variance infation factors, and emissions control regulations Te average of the previous day load, Yard crane the average of the previous week load, NN (BP) and Primary data Port of Felixstowe in [46] ● ● RTG crane demand of one hour demand the same hour load for previous day, SVM sources the UK and the previous hour load Maximum continuous rate (MCR) Primary data Two ocean-going [40] ● Berthing ● PR Emissions (NO , SO , CO , and CO) x x 2 measured by megawatt, shaft speed sources vessels in Australia Trolley position, trolley speed, loading Secondary [48] ● ● Antiswing crane ● ● NN (ANFIS) Te driving force of the trolley — angle, and angular velocity data sources Number of quay crane, acres, berth Port performance and depth, undesirable output (CO ), Clusters of decision-making units Secondary 20 American [29] ● ● ● ● SOM evaluation and desirable outputs (calls, (DMUs) data sources container ports throughput, and deadweight tonnage) Ship identifcation, position, speed, Ports of Newcastle, Emissions (NO , SO , PM , VOC, Primary data x 2 2.5 [41] ● Berthing ● course, heading and navigational PR Jackson, Botany, and CO, NH , CO, N O, and CH ) sources 3 2 4 status, and timestamp Kembla in Australia Te background with low-rank Primary data [44] ● Berthing ● ● Observed video PCA property and the foreground with Unknown sources sparse property RTGC number, block number, handling container specifcation, A container terminal Port performance Resource allocation for container Secondary [37] ● ● ● stevedoring full or empty category, K-means on the east coast of evaluation terminals data sources handling volume for a task, and the China number of clusters Te net tonnage, deadweight tonnage, GBoost, RF, NN Secondary [42] ● Berthing ● actual handling volume, and efciency (BP), PR, and Energy consumption Jingtang port (China) data sources of facilities KNN CO emission driver factors of the city where the port is located are gross domestic product, total resident population, the number of port berth, total imports, total exports, the frst Clusters of similar ports in terms of Port performance industrial value, the secondary environmental sustainability (LISA Secondary 30 Chinese container [25] ● ● ● ● Spatial clustering evaluation industrial value, the primary industrial cluster maps of PCD carbon data sources ports value, gross industrial production, emissions) fxed assets investment in the tertiary industry, per capita income, railway freight volume, highway freight volume, and waterway freight volume 8 Journal of Advanced Transportation Table 1: Continued. Areas Environmental problems Machine learning Data Research Case Articles collection Overall Water Noise Energy Renewable Solid Techniques scope studies Seaside Yard Landside Emission Input Output methods port pollution pollution saving energy waste (tools) Number of berths, the length of the Cargo throughput, NO emissions, Port performance Secondary [32] ● ● ● ● terminal, the number of staf, and the PR SO emissions, and solid waste 18 Chinese ports evaluation data sources total fxed assets containers Determinant factors of the survey are lean management, green operational Highly correlated input variables PCA practices, green behavior (green participation and green compliance), Port performance and green climate Kaohsiung container [27] ● ● ● ● Survey evaluation Determinant factors of the survey are port (Taiwan) lean management, green operational Green performance (fnancial and practices, green behavior (green PR nonfnancial) participation, green compliance), and green climate Noise of moving Draught, speed, and Primary data Industrial port of [45] ● ● PR Sound emitted ships in port areas ship-to-microphone distance sources Livorno (Italy) Container features are cycle, type, weight, special (e.g., hazard shipping), agreement (between stakeholders), Secondary Port of Altamira Hierarchical [51] ● Truck scheduling ● vessel departure time, distance (of two Container groups data sources, (Mexico) and Port of clustering containers in the yard), customs survey Genoa (Italy) clearance, dwell time, and fnal destination Total gross weight of goods, air Port performance Energy consumption and number of Hierarchical Secondary 24 European [30] ● ● ● pollutant emissions, and the rank of evaluation employees clustering, PR data sources container ports ports in terms of eco-efciency Indoor air quality A liner between CO concentration and load (number of Te reference fow rate of the Secondary [36] ● prediction ● ● NN (BP) Egypt and Saudi cars) ventilation system data sources (RORO) Arabia ports ETA features (date, time, and Secondary [38] ● Berthing ● weekday) and ship features (ship type SVM Arrival time of vessels — data sources and length) Air quality Fine particulate mass and fne Primary data [33] ● ● PR Air quality Long Beach (US) prediction particulate composition sources Scheduled arrival, departure, and load/ unload start time, planned berthing Primary data Hamburg container [12] ● AGV ● ● place, planned position of front and NN (BP) Availability of AGVs sources terminal (Germany) rear of the ship, and number of containers to load and unload Principal components (container Container truck Highly correlated data of trafc and Secondary Waigaoqiao port [52] ● ● PCA truck volume, other vehicles volume, emissions particle number concentrations (PNC) data sources (China) and PNC data) Ports of Ambarlı, Air quality Type of pollutant, the operating mode, Emissions (SO , NO , CO , VOC, Secondary 2 x 2 [34] ● ● PR Izmir, Mersin, and prediction and gross tonnage of ships PM, and CO) data sources Kocaeli (Turkey) Energy consumption of hoist, gantry, Secondary Casablanca port [47] ● ● RTG crane ● ● PR General energy consumption of RTG and trolley data sources (Morocco) Air quality Meteorological data, air quality data, Emissions (PM , PM , SO , O , Secondary 2.5 10 2 3 [35] ● ● RNN and LSTM Busan port (Korea) prediction and shipping activity data NO , CO) data sources Hourly data of energy (electricity) LSTM, NN (BP), Secondary A navigation route in [43] ● Berthing ● ● Day-ahead prices of energy prices and load demands Elman, RBF data sources Australia Air quality, rate of treatment of wastewater, standard-reaching rate of nearshore water, green coverage Highly correlated input variables PCA rate in developed areas, and expenditure on energy-saving Secondary Port performance [26] ● ● ● ● investments per capita data sources, 15 Chinese seaports evaluation Air quality, rate of treatment of survey wastewater, standard-reaching rate of Hierarchical Te rank of ports based on nearshore water, green coverage rate in clustering environmental sustainability features developed areas, and expenditure on energy-saving investments per capita Journal of Advanced Transportation 9 Figure 7. As can be observed in the fgure, supervised Furthermore, considering truck emissions in the port of Hamburg, the authors of reference [50] developed a multi- learning with 70% is the most used type of technique while unsupervised learning is the second option with 30% of objective model for interterminal truck routing problems and utilized a machine learning technique as part of the collected articles. Particularly, polynomial regression (PR) decision support system. Moreover, using two real container with 30% and neural networks (NN) with 27.5% are the most terminals, i.e., the port of Altamira (Mexico) and the port of used tools. Terefore, according to the main categorization Genoa (Italy) as case studies, the authors of reference [51] of ML techniques discussed in Section 3.3, there is no article proposed a methodological framework to reduce empty using classifcation nor reinforcement learning among the truck trips to minimize the deviation from their preferred collected articles. time slots and turnaround times in container terminals and reduce emissions. Te authors of reference [52] studied the 3.3.1. Supervised Learning. Supervised learning, commonly relationship between trafc volume and the particle number called predictive learning, is used for labelled datasets in concentrations (PNC) caused by emissions of container which the response of a scenario or example is known [53]. It trucks in the port of Waigaoqiao (China). For this, they enables several regression tools (e.g., polynomial regression, combined a machine learning technique with statistical neural networks, and k-nearest neighbour) for predicting methods to characterize the variation of particles in the the behaviour of a dataset. Moreover, some classifcation port area. tools (e.g., Bayesian network, logistic regression, and de- cision tree) are other applications of supervised learning 3.3. ML Techniques to Promote Green Port Operations (RQ3). when the output is categorical [54]. As seen in Table 1, only Researchers or practitioners who seek to apply ML in regression-related algorithms have been developed in the scope of this review. maritime port operations should possess the fundamental competency of selecting an algorithm that is appropriate for Regression is a supervised learning technique that aims to identify the correlation between variables and predict the a given task or problem. However, conceptualizing a way toward using ML to improve the performances of port continuous output variables based on one or more predictor variables. In this regard, to evaluate port efciency in terms operations is challenging in the absence of expertise or prior research of a similar type, especially when taking into ac- of environmental problems, [24] used polynomial regression count the numerous algorithms that have been ofered in the for predicting the amount of NOx emission based on port technical literature. assets, berth quantity, and geographical location of 17 port ML is mainly classifed into three diferent types, i.e., enterprises in China. Te authors showed a beneft of the supervised learning, unsupervised learning, and re- polynomial regression model for ports performance eval- inforcement learning [53]. Given that division, a systematic uation and found that the medium-sized and large-scale ports should focus on emissions reduction compared to literature review by [54] illustrated the machine learning techniques used in industrial applications so far which are small-sized ports that should focus on improving the service level and full resource utilization. Similarly, [25] considered organized in Table 2. Te diferent tools related to (i) su- pervised learning with classifcation and regression algo- the number of berths, the length of the terminal, the number rithms, (ii) unsupervised learning with clustering and of staf, and the total fxed assets of 18 ports in China as the dimensionality reduction algorithms, and (iii) re- input variables of their regression model to predict NOx and inforcement learning, are presented in the table. In order to SOx emissions as well as solid waste and energy con- highlight current and emerging trends and, more impor- sumption in the selected ports. Based on the results of tantly, to guide researchers or practitioners in the selection a regression model, the authors found that economic de- velopment positively impacts green efciency. Reference of ML techniques, the table demonstrating the sub- classifcation of ML algorithms is used in this review to map [31] used berth length, the number of cranes, terminal area, and amount of cargo handled as the independent variables of techniques when analyzing the collected articles. For further information on the tools, see the reference of the table. their model for predicting emissions as well as port per- formance evaluation. Te authors found, by applying a re- To provide a better vision of the problems (RQ2) and the techniques (RQ3) discussed in this SLR, Table 1 is presented. gression model, that although ECA regulation reduces Table 1 summarizes the main characteristics of the reviewed emissions, it signifcantly harms port productivity due to articles based on the following categories: the application area losing cargoes. within the port (i.e., seaside, yard, landside, and overall port), For evaluating the port of Kaohsiung in Taiwan in terms the research scope, environmental problems (i.e., emissions, of environmental sustainability, [27] used several input water pollution, noise pollution, energy saving, renewable variables including lean management, green operational practices, green behaviour (green participation and green energy, and solid waste), the machine learning technique, involved factors (input and output), the data collection compliance), and green climate to predict green perfor- mance (fnancial and nonfnancial). Using the results of method, and the case study. Te information provided in this table is discussed in Sections 3.1, 3.2, and 3.3. a regression model, the authors concluded that lean man- agement positively impacted green operations and green Considering the ML techniques used in the collected articles, a hierarchical categorization including the ML behaviour. Green operational practices had a positive in- techniques, the subdomains, and the tools is shown in fuence on both green behaviour and green performance. 10 Journal of Advanced Transportation Table 2: Categorization and used machine learning techniques in industrial applications [54]. ML domain ML subdomains Algorithms Tools (i) K-means (i) Spatial cluster (SC) (i) Local outlier factor (LOF) Clustering (ii) K-median (ii) Gaussian mixture model (ii) Neighbour-based clustering (NBC) (iii) Hierarchical clustering (HC) (GMM) (iii) Parzen windows (PW) Unsupervised (i) t-distributed stochastic neighbour (i) Principal component analysis (i) Kernel principal component learning embedding (t-SNE) Dimensionality (PCA) analysis (K-PCA) (ii) Uniform manifold approx. and projection reduction (ii) Linear discriminant analysis (LDA) (ii) Singular value decomposition (UMAP) (iii) Kernel density estimator (KDE) (SVD) (iii) Self-organizing maps (SOM) (i) Neural networks (NN) (i) NN, multilayer perception (ii) NN, back propagation (BP) (MLP) (iii) NN, convolutional neural network (ii) NN, radial basis function (RBF) (i) Locally weighted regression (LWR) (CNN) (iii) NN, recurrent neural network (ii) Support vector machine (SVM)- regressor (iv) NN, extreme learning machine (RNN) (iii) Gradient boosting (GBoost) Regression (ELM) (iv) Linear regression (LR) (iv) Random forest (RF)- regressor (v) NN, long-short term memory (v) Polynomial regression (PR) (v) K-nearest neighbor (KNN)-regressor Machine (LSTM) (vi) Fuzzy regression (FR) (vi) Gaussian process regression (GPR) learning (vi) NN, deep learning (DL) (vii) Bayesian regression (BR) Supervised learning (vii) NN, adaptive neuro-fuzzy (viii) Lasso regression (LASSO) inference system (ANFIS) (i) Adaptive support vector (i) Decision tree (DT) machine (ASVM) (i) K-nearest neighbor (KNN) (ii) Gradient boosting (GBoost) (ii) Learning vector quantization (ii) Quadratic discriminant analysis (QDA) (iii) Naive bayes (NB) (LVQ) (iii) Random forest (RF) Classifcation (iv) Bayesian network (BN) (iii) Linear discriminant analysis (iv) Logistic regression (LogR) (v) Kernel method (KM) (LDA) (v) Pattern recognition (PattR) (vi) Multi-layer perception (MLP) (vi) Support vector machine (SVM) (iv) Stochastic gradient descent (SGD) (i) Approximate dynamic (i) Adaptive heuristic critic (AHC) programming (ADP) (i) State-action-reward-state-action (SARSA) Reinforcement (ii) Deep deterministic policy gradient (ii) Proximal policy optimization (ii) Temporal diference learning (TD) learning (DDPG) (PPO) (iii) Trust region policy optimization (TRPO) (iii) Q-learning (QL) (iii) Deep Q- learning (DQL) Journal of Advanced Transportation 11 2 2 2 1111 111 1 11 K- PR NN SVM GBoost RF KNN HC means SC PCA KDE SOM (30%) (27.5%) (5%) (2.5%) (2.5%) (2.5%) (7.5%) (5%) (2.5%) (10%) (2.5%) (2.5%) Regression Classifcation Clustering Dimensionality (70%) (0%) (15%) Reduction (15%) Supervised Unsupervised Reinforcement (70%) (30%) (0%) Machine Learning Techniques Figure 7: Categorization of ML techniques (subdomains, algorithms, and tools) for promoting environmental sustainability in port logistics. Reference [30] used the total energy consumption of ports [49] developed a forecasting engine for truck arrivals to and the number of employees as the input variables of their logistics nodes, i.e., empty container depots and packing model to predict the total gross weight of goods, air pollutant facilities or terminals that mitigate greenhouse gas emissions emissions, and the eco-efciency rank of ports. With a re- from truck congestions in the landside. In doing so, they proposed a neural network model by taking historical data of gression model, they revealed that the energy consumption variable had a signifcant diverse correlation with the eco- truck arrival time, administrative wait time, intermediate wait time, node-specifc forecasting parameters (e.g., dis- efciency of ports. Moreover, [33] used the fne particulate mass and the fne particulate composition as the input patching modes and storage policies), and external fore- variables of their model to predict air quality. Te authors casting parameters (e.g., weather information and trafc concluded that polynomial regression models provided information) as the inputs. Te authors showed the beneft useful analysis for air quality management. of neural networks in the smoothed peak workloads at the Te authors of reference [34] considered the type of nodes due to adaptive truck routing and reduced pollutant, the operating mode, and the gross tonnage of waiting times. ships to predict the amount of emission. Based on the re- Intending to manage the energy consumption of RTG gression analysis, they found that innovative methods cranes, the authors of reference [46] utilized neural networks and a support vector machine and considered the average of proposed by the International Maritime Organization (IMO) such as carbon capture and storage systems, in- the previous day load, the average of the previous week load, creasing energy efciency, and emissions converting tech- the same hour load for the previous day, and the previous nologies had a signifcant impact on emissions reduction. hour load as the input variables of their model to predict Te authors of reference [35], by using long short-term RTG crane demand of one hour. Tey revealed that the memory (LSTM) and the recurrent neural network (RNN), efectiveness of the neural networks model was signifcantly used meteorological data, air quality data, and shipping high when the estimation of the number of crane moves and activity data as the input variables to predict emissions in container gross weight was accurate. Furthermore, to predict ports. Te authors indicated that besides meteorological data the general energy consumption of RTGs, the authors of reference [47] proposed a regression model based on the and air quality data, ship activities, as one of the main sources of emissions in port areas, should be considered in energy consumption of hoist RTG, gantry RTG, and trolley RTG. Te authors showed huge air pollution decrease and the prediction model to enhance the performance. Using neural networks, the authors of reference [36] developed cost-saving on energy by the forecasting model. In other a predictive model for controlling the CO concentration in work, through an adaptive neuro-fuzzy inference system RORO ships indoors. Tey considered CO concentration, (ANFIS), the authors of reference [48] developed a model to load (number of cars), and the reference fow rate of the minimize swings of RTG during loading/unloading of ventilation system as the input variables of their model. Te containers and cargo in the yard and seaside area which leads authors concluded that neural networks models combined to prevent emission of hazardous materials into the air and with other methodologies such as fuzzy controlling and water. Tey considered trolley position, trolley speed, particle swamp optimization signifcantly guarantee the loading angle, angular velocity, and the driving force of the trolley as the input variables of their neural network model. robustness of the indoors CO concentration reduction in RORO ships to an allowable limit. Te authors of reference Te authors showed that the ANFIS control method was BP (12.5%) LSTM (5%) Elman (2.5%) RBF (2.5%) RNN (2.5%) ANFIS (2.5%) 12 Journal of Advanced Transportation sources of noise from ships [55]. For this sake, the authors of robust and quick-response under diferent rope lengths and working conditions, but not reliable enough when the noise reference [45] proposed a regression model based on draught, distance from a recording microphone, and speed to evaluate was strong. In the same area, to predict the demand for battery-electric AGVs, the authors of reference [12] took the the correlations among the variables and predict sound scheduled arrival, departure and work started, planned emitted from moving ships in port areas. Te authors in- berthing place, the planned position of front and rear of the dicated that ship draught was not an infuencing parameter for ship, and the number of containers to load and unload as the noise emissions. Also, the authors concluded that for the noise input variables of their neural networks model. Te authors assessment in port areas, the right placement of the noise reported a beneft of the neural networks model for checking source that provides precise input data plays an essential role in improving the output of an acoustic model. the availability of AGVs in the horizontal transportation area of ports. In the seaside area, the authors of reference [40] pre- dicted the engine exhaust emissions using the polynomial 3.3.2. Unsupervised Learning. Unsupervised learning or descriptive learning is used for unlabeled datasets that the regression based on the maximum power output and shaft speed of ships during berthing operations. Tey proposed response to a scenario or example is unknown [53]. It en- a forecasting model that was signifcantly accurate for dif- ables several clustering tools (e.g., hierarchical clustering, k- ferent engine types at berth, manoeuvring, and sea. More- means, and fuzzy c-means) for recognizing the behaviour of over, to develop the berth allocation planning to manage a dataset that there is no historical output. Moreover, some real-time data and air emissions reduction, [38] used ETA dimensionality reduction tools (e.g., principal component features (date, time, and weekday) and ship features (ship analysis and self-organizing maps) are known as other applications of unsupervised learning that minimize the type and length) as the inputs of their support vector ma- chine model to build a predicting system for vessel arrivals. volume of datasets to an efcient computation process. Te authors concluded that the use of additional features (e.g., weekday) and discarding irrelevant inputs (e.g., the (1) Clustering. Clustering is an unsupervised learning technique that considers a set of selected features to group shipping line) have a positive infuence on the performance of the SVM model. With the same purpose, using a re- objects with similar attributes. Te purpose of the clustering gression model, [41] utilized ship identifcation, position, technique is to construct clusters where data objects within speed, course, timestamp, heading, and navigational status the same cluster are similar to one anther but diferent from of ships. Tey reported the beneft of regression analysis to the objects in other clusters. In this regard, for bench- model the spatial extent (the active area) of the emissions at marking ports in terms of environmental sustainability, diferent temporal resolutions (hourly and daily). In a sim- hierarchical clustering is used to create dendrograms or cluster trees. For instance, the authors of references [26, 30] ilar work, using the net tonnage, deadweight tonnage, actual handling volume, time of ships arrival, and efciency of evaluated and benchmarked several ports in Europe and China considering energy consumption, rate of wastewater facilities as the input variables, [42] used fve regression tools of the machine learning technique including gradient boost treatment, standard-reaching rate of nearshore water, the (i.e., Gboost), backpropagation neural network, linear re- green coverage rate in developed areas, and expenditure on gression, k-nearest neighbour, and random forest to predict energy-saving investments per capita. Based on the revealed energy consumption and emissions from ships during clusters, they both concluded that the ports in the same berthing operation. Te authors found that the time of ships cluster with the best performance in terms of technical ef- arrival without infuencing the performance of the model fciency showed a better eco-efciency performance than could be eliminated to reduce the difculty of data collec- other clusters. tion. Tey also concluded that when the efciency of fa- Spatial clustering by splitting spatial data into a series of meaningful subclasses aims to consider the selected features cilities was doubled, the energy consumption of ships was reduced by 34.17% at berth and 8.41% in overall port areas. to group spatial objects in the same cluster that are similar to each other and dissimilar to those in diferent clusters. In this To manage the energy consumption of an all-electric ship (AES), the authors of reference [43] proposed several regard, to investigate the spatial characteristics of emissions neural network-based models including the Elman, back- from the port container distribution (PCD), the authors of propagation (BP), the radial basis function (RBF), and long reference [32] used this type of clustering and considered short-term memory (LSTM). Tey considered hourly data of several parameters such as CO emission driver factors of the electricity prices and load demands as the input variables. city where the port is located, gross domestic product, total Te authors revealed that the LSTM method could predict resident population, the number of port berths, total im- the hourly price of electricity onshore accurately. As a result, ports, total exports, the frst industrial value, the secondary the method combined with an optimization model resulted industrial value, the primary industrial value, gross in- in the minimum cost and emission of the AES. dustrial production, fxed assets investment in the tertiary Ship sources, one of the main sources of noise emissions in industry, per capita income, railway freight volume, highway the port area (i.e., roads, railways, ships, port activities, and freight volume, and waterway freight volume. With the industrial plants), are from all the activities related to the spatial clustering technique, they reported that ports with movement and stationing of ships. Engines, funnels, and similar geographical locations showed a similar pattern of ventilation, as well as transit in port regimes, are the main PCD carbon emissions. Journal of Advanced Transportation 13 In the yard area, to manage trucks operations in con- climate to evaluate ports in terms of green performance tainer terminals and reduce empty truck trips, it is important (fnancial and nonfnancial). Tey showed the beneft of PCA as a data preprocessing tool for fnding the relationship to identify container features. In doing so, the authors of reference [51] performed a dendrogram and clustered between principal components of an equation in polynomial containers based on several input variables including the regression models. Moreover, for ranking ports based on cycle of import/export, the ISO type of the container, the several environmental sustainability factors, [26] utilized weight of the container, special (e.g., hazard shipping), the PCA to identify the independent variables. Tey proposed agreement between stakeholders, vessel departure time, the the rate of treatment of wastewater, the standard-reaching distance of two containers in the yard, customs clearance for rate of nearshore water, the green coverage rate in developed import/export, the dwell time of the container, and the fnal areas, and expenditure on energy-saving investments per destination of container in the hinterland. Te authors capita as the input variables. Te work concluded that PCA showed the beneft of container clustering in terms of re- helped in reducing the dimension of indicators when ducing the number of trucks for moving the same number of combined with a hierarchical clustering model. Te authors of reference [44] developed a computer vision-based model containers. In the same area, the authors of reference [37] developed a k-means model based on the number of RTGs, to detect ships entering into the imaging area at the seaside blocks, handling container specifcations, stevedoring full or and help them with automatic berthing. Tey considered the empty category, handling volume for a task, and the number observed video as the input variable of the PCA and sep- of clusters to manage the relationship between RTG crane arated the foreground object (ship) from the background teams and the given block sets. Te authors concluded that scene of each video frame as the outputs. Te authors re- the k-means model was an efcient tool for clustering block ported the beneft of PCA for reducing the dimensions of communities and dispatching RTG cranes in the yard area. image features. Moreover, to identify the relationship be- Furthermore, the authors of reference [50] utilized a k- tween the trafc and particle number concentrations (PNC) means as part of a decision support system to provide data from container truck emissions in the yard, the authors representative solutions for a multiobjective interterminal of reference [52] applied the PCA and proposed container truck volume, other vehicles’ volume, and PNC data as truck routing problem. Tey used the number of clusters and a solution archive as the inputs and the cluster centroids as uncorrelated variables for characterizing the variation of the output of the model. Te authors showed that using k- particles. Tey found that the method had a high perfor- means inside their multiobjective algorithm was a suitable mance in dimensionality reduction when combined with clustering approach for reducing the set of solutions and, a Pearson correlation analysis. Tey also concluded that thus, making the decision process more manageable. dimensionality reduction signifcantly reduced the com- putation cost and data collection difculties. (2) Dimensionality Reduction. Dimensionality reduction in machine learning is a data preprocessing technique that 3.4. Advantages and Disadvantages. Regarding Table 1 that refers to reducing the number of input variables in a dataset summarized the main characteristics of the reviewed articles to minimize computational costs and increase speed [53]. It based on the following categories: the application area enables several tools (e.g., principal component analysis within the port (i.e., seaside, yard, landside, and overall (PCA), kernel density estimator (KDE), and self-organizing port), the research scope, environmental problems (i.e., map (SOM)) for dataset volume reduction [54]. emissions, water pollution, noise pollution, energy saving, Regarding benchmarking and evaluating ports, [29] used renewable energy, and solid waste), the machine learning SOM combined with a data envelopment analysis (DEA) to technique, involved factors (input and output), the data propose similar groups of ports in terms of environmental collection method, and the case study, Table 3 reports the efciency. In doing so, they took the number of quay cranes, advantages and disadvantages of the used ML techniques in acres, berth, depth, the number of calls, throughput, the collected articles. deadweight tonnage, and CO emissions as well as the inputs of the SOM. Te authors concluded that the SOM was a suitable tool for reducing the dimensions of the features to 3.5. Open Issues and Future Works. Tis subsection discusses some important future research directions and open view- a simple visualized map. For the same purpose, the authors of reference [28] implemented the KDE tool to measure points toward the application of ML techniques in the environmental sustainability of maritime ports. Future re- pollution incidents intensity among the ports. Te KDE selected a common distribution and estimated the param- search may concentrate on analyzing the efectiveness of eters for the density (intensity) function from the data various machine learning (ML) algorithms for reducing the sample. Te authors indicated that KDE was an accurate tool impact of environmental concerns related to solid wastes or for generating a smooth surface variation and mapping the noise pollution near port cities, as there are numerous spatial distribution of ports pollution probability density of pollutants but few solutions for those issues as of yet. Te use a geographic area. Reference [27] applied PCA for reducing of ML approaches to help renewable energy technology is the dimension of highly correlated input variables of another important study area that can be tackled by future a survey to a few independent factors including lean academics. Te use of optimization techniques like meta- heuristics, mathematical programming, and heuristic ap- management, green operational practices, green behaviour (green participation and green compliance), and green proaches to aid decision making for planning and carrying 14 Journal of Advanced Transportation Table 3: Advantages and disadvantages of the used ML techniques for promoting environmental sustainability in maritime port logistics. ML techniques Advantages Disadvantages Related articles (i) Works on any size of the dataset (i) Te correct polynomial degree should be chosen for Polynomial regression (PR) (ii) Gives information about the relevance of [24, 25, 27, 30, 31, 33, 34, 40–42, 45, 47] a good bias features (i) Hardware dependent (i) Efciency (ii) Complex algorithms (ii) Continuous learning Neural networks (NN) (iii) Black-box nature [12, 35, 36, 42, 43, 46, 48, 49] (iii) Data retrieval (iv) Approximate results (iv) Multitasking (v) Data-dependency (i) Works very well on nonlinear problems Support vector machine (ii) Easily adaptable (i) Requires feature scaling [38, 46] (SVM) (iii) Ignores outliers (i) Performs very well on medium and small datasets (ii) Easy to interpret (i) Sensitive to outliers Gradient boost (GBoost) (iii) Prevents overftting (ii) Hardly scalable [42] (iv) A great approach for enhancing classifcation (iii) Poor results on unstructured data and regression solutions (i) Accurate (i) Te number of trees should be chosen Random forest (RF) (ii) Powerful [42] (ii) No interpretability (iii) Works very well on linear/nonlinear problems (i) Fast (i) Need feature scaling K-nearest neighbor (KNN) [42] (ii) Easy to implement (ii) Data should be cleaned (i) Easy to implement Hierarchical clustering (HC) (i) Long runtime [26, 30, 51] (ii) Does not require the number of clusters (i) Requires the number of clusters (i) Easy to implement (ii) Requires initial values (ii) Large datasets K-means (iii) Requires dimensionality reduction tools if the [37, 50] (iii) Convergence number of dimensions is high (iv) Diferent shapes and sizes with generalization (iv) Does not ignore outliers (i) Does not require the number of clusters (i) Poor results for datasets with various densities Spatial clustering (SC) (ii) Diferent shaped clusters (ii) Poor results for unstructured data [32] (iii) Ignores outliers (iii) Not deterministic (i) Eliminates correlated features (i) Less interpretable Principal component (ii) Enhances algorithm performance (ii) Data has to be uniformed [26, 27, 44, 52] analysis (PCA) (iii) Reduces overftting (iii) Loss of information (iv) Enhances visualization Kernel density estimator (i) Smooth visualization (i) Biased at the boundaries [28] (KDE) (ii) Works with various shapes and sizes (ii) Information loss by oversmoothing (i) Interpretable Self-organizing maps (SOM) (i) Requires initial weights [29] (ii) Applicable for large datasets Journal of Advanced Transportation 15 implementation of ML techniques to promote environ- out port operations that incorporate environmental aspects is also something we propose as an extension given that the mentally sustainable practices at ports, we propose as an extension the consideration of optimization techniques such focus of this SLR was on the implementation of ML tech- niques to promote environmentally sustainable practices as metaheuristics, mathematical programming, and heu- in ports. ristic approaches to aid decision making for planning and executing port operations incorporating environmental aspects. Finally, a complementary SLR on green maritime 4. Conclusions shipping can be a relevant research direction to provide Tis article presented the results of a systematic literature insights and analyse the current contributions of ML in that review of the recent literature on machine learning for application domain. promoting environmental sustainability in ports. Te review explored contributions of fve years (2017–2021) with the Data Availability aim of capturing the most recent approaches to foster en- No data were used to support this study. vironmental sustainable port operations. It categorized the identifed articles based on their application area within the port as well as the application of the ML approach. Using the Conflicts of Interest PRISMA protocol and bibliometric tools, the research Te authors declare that they have no conficts of interest. framework was constituted on the major considerations of impacts, techniques, and problems. In general, the chal- lenges and barriers of machine learning to aid decision Acknowledgments making and promote more sustainable green practices at Te third author of the manuscript acknowledges the ports were discussed. By analyzing the 29 identifed articles support of the National Agency of Research and Develop- with their methodological approaches, this review sum- ment (ANID), Chile, through the grant no. FONDECYT marized the academic contributions considering the three N.1210735, the program STIC-AMSUD-22-STIC-09, and main dimensions including machine learning, port logistics, the grant no. FOVI220133 “Digital Ports of Next and environmental sustainability. Te articles that used Generation”. regression models were dominant in the literature, while LSTM and RNN were the most recent approaches. Also, in terms of environmental indicators, investigations on References emissions and energy consumption were predominant [1] E. Lalla-Ruiz, L. Heilig, and S. Voß, “Environmental sus- among collected articles. tainability in ports,” Sustainable Transportation and Smart Concerning the type of approaches, supervised learning Logistics: Decision-Making Models and Solutions, Vol. 65–89, was the most used type of technique while unsupervised Elsevier, , Amsterdam, Netherlands, 2018. learning was the second option with about 30% of collected [2] A. Vega-Muñoz, G. Salazar-Sepulveda, J. F. Espinosa-Cristia, articles. Dimensionality reduction tools as data pre- and J. Sanhueza-Vergara, “How to measure environmental processing techniques reduced computation costs as they performance in ports,” Sustainability, vol. 13, no. 7, p. 4035, minimized the volume of data in databases. Clustering techniques were benefcial to constructing clusters where [3] H. Davarzani, B. Fahimnia, M. Bell, and J. Sarkis, “Greening ports and maritime logistics: a review,” Transportation Re- data objects within the same cluster were similar to one search Part D: Transport and Environment, vol. 48, pp. 473– anther but diferent from the objects in other clusters. 487, 2016. Moreover, the information on ML-based prediction algo- [4] M. Ghobaei-Arani, M. Shamsi, and A. A. Rahmanian, “An rithms such as polynomial regression and neural networks efcient approach for improving virtual machine placement showed a high contribution to promoting environmental in cloud computing environment,” Journal of Experimental indicators including greenhouse gas emissions, energy and Teoretical Artifcial Intelligence, vol. 29, no. 6, consumption, and water pollution. pp. 1149–1171, 2017. Te outcomes of this SLR can be used as a starting point [5] Y. Zheng, J. Zhao, and G. Shao, “Port city sustainability: for investigations regarding the application of machine a review of its research trends,” Sustainability, vol. 12, no. 20, learning techniques for enhancing environmental sustain- p. 8355, 2020. ability in ports and provide a guideline for researchers to [6] H. Geerlings and T. Vellinga, “Sustainability,” in Ports and Networks, pp. 296–314, Routledge, Oxfordshire, UK, 1 edi- fnd research gaps in the literature. In the light of this review, tion, 2018. future studies may focus on examining the performance of [7] F. Schulte, E. Lalla-Ruiz, R. G. Gonzalez-Ram ´ ´ırez, and S. Voß, diferent ML algorithms for mitigating the impact of envi- “Reducing port-related empty truck emissions: a mathemati- ronmental concerns related to solid wastes or noise pollution cal approach for truck appointments with collaboration,” near port cities, as there are several pollutants, but there are Transportation Research Part E: Logistics and Transportation only a few contributions for solving those problems yet. Review, vol. 105, pp. 195–212, 2017. Another relevant research direction is the one related to [8] HLEG, “Independent High-Level Expert Group on Artifcial supporting renewable energy technologies based on ML Intelligence Set up by the European Commission,” 2019, techniques that can be addressed by future researchers. https://www.aepd.es/sites/default/fles/2019-12/ai-defnition. Moreover, given that the scope of this SLR was on the pdf. 16 Journal of Advanced Transportation [9] A. Shakarami, A. Shahidinejad, and M. Ghobaei-Arani, “An [24] J. Sun, Y. Yuan, R. Yang, X. Ji, and J. Wu, “Performance autonomous computation ofoading strategy in Mobile Edge evaluation of Chinese port enterprises under signifcant en- vironmental concerns: an extended DEA-based analysis,” Computing: a deep learning-based hybrid approach,” Journal Transport Policy, vol. 60, pp. 75–86, 2017. of Network and Computer Applications, vol. 178, Article ID [25] L. Wang, C. Peng, W. Shi, and M. Zhu, “Carbon dioxide 102974, 2021. emissions from port container distribution: spatial charac- [10] F. Jazayeri, A. Shahidinejad, and M. Ghobaei-Arani, “Au- teristics and driving factors,” Transportation Research Part D: tonomous computation ofoading and auto-scaling the in the Transport and Environment, vol. 82, Article ID 102318, 2020. mobile fog computing: a deep reinforcement learning-based [26] C. Wang, H. Haralambides, and L. Zhang, “Sustainable port approach,” Journal of Ambient Intelligence and Humanized development: the role of Chinese seaports in the 21st century Computing, vol. 12, no. 8, pp. 8265–8284, 2021. Maritime Silk Road,” International Journal of Shipping and [11] S. Samoili, M. Lopez ´ Cobo, E. Gomez, ´ G. De Prato, Transport Logistics, vol. 13, no. 1/2, p. 205, 2021. F. Mart´ınez-Plumed, and B. Delipetrev, “AI Watch - defning [27] S. Y. Kuo and P. C. Lin, “Determinants of green performance Artifcial Intelligence. Towards an operational defnition and in container terminal operations: a lean management,” taxonomy of artifcial intelligence,” in Joint Research Cen- Journal of Cleaner Production, vol. 275, Article ID 123105, treEuropean Commission, Brussels, Belgium, 2020. [12] S. Holly, A. Nieße, M. Troschel ¨ et al., “Flexibility management [28] S. Cheon, A. Maltz, and K. Dooley, “Te link between eco- and provision of balancing services with battery-electric au- nomic and environmental performance of the top 10 U.S. tomated guided vehicles in the Hamburg container terminal ports,” Maritime Policy and Management, vol. 44, no. 2, Altenwerder,” Energy Inform, vol. 3, no. 1, p. 26, 2020. pp. 227–247, 2017. [13] N. H. Prevljak, “HHLA container terminals use machine [29] Y. S. Park, N. M. A. Mohamed Abdul Ghani, F. Gebremikael, learning technology to boost productivity,” Ofshore Energy, and G. Egilmez, “Benchmarking environmental efciency of Broussard, Louisiana, 2020. ports using data mining and RDEA: the case of a U.S. con- [14] I. de la Peña Zarzuelo, M. J. Freire Soeane, and B. Lopez ´ tainer ports,” International Journal of Logistics Research and Bermudez, ´ “Industry 4.0 in the port and maritime industry: Applications, vol. 22, no. 2, pp. 172–187, 2019. a literature review,” Journal of Industrial Information In- [30] C. Quintano, P. Mazzocchi, and A. Rocca, “Examining eco- tegration, vol. 20, Article ID 100173, 2020. efciency in the port sector via non-radial data envelopment [15] G. Yeo, S. H. Lim, L. Wynter, and H. Hassan, “MPA-IBM analysis and the response based procedure for detecting unit project SAFER: sense-making analytics for maritime event segments,” Journal of Cleaner Production, vol. 259, Article ID recognition,” INFORMS Journal on Applied Analytics, vol. 49, 120979, 2020. no. 4, pp. 269–280, 2019. [31] Y. T. Chang, H. K. Park, S. Lee, and E. Kim, “Have emission [16] E. F. France, N. Ring, J. Noyes et al., “Protocol- control areas (ECAs) harmed port efciency in Europe?” developingmeta-ethnography reporting guidelines Transportation Research Part D: Transport and Environment, (eMERGe),” BMC Medical Research Methodology, vol. 15, vol. 58, pp. 39–53, 2018. no. 1, pp. 103–114, 2015. [32] L. Wang, Z. Zhou, Y. Yang, and J. Wu, “Green efciency [17] A. Tong, K. Flemming, E. McInnes, S. Oliver, and J. Craig, evaluation and improvement of Chinese ports: a cross- “Enhancing transparency in reporting the synthesis of efciency model,” Transportation Research Part D: Trans- qualitative research: entreq,” BMC Medical Research Meth- port and Environment, vol. 88, Article ID 102590, 2020. odology, vol. 12, no. 1, pp. 181–188, 2012. [33] D. J. Eatough, P. Cropper, W. Keeton et al., “Apportionment [18] B. C. O’Brien, I. B. Harris, T. J. Beckman, D. A. Reed, and of PM2.5 adjacent to the I-710 harbor freeway in long Beach, D. A. Cook, “Standards for reporting qualitative research: CA,” Journal of the Air and Waste Management Association, a synthesis of recommendations,” Academic Medicine, vol. 89, vol. 70, no. 3, pp. 260–282, 2020. no. 9, pp. 1245–1251, 2014. [34] A. Ekmekçioglu, ˘ K. Unlugenço ¨ glu, ˘ and U. B. Çelebi, “Con- [19] K. Flemming, A. Booth, K. Hannes, M. Cargo, and J. Noyes, tainer ship emission estimation model for the concept of “Cochrane Qualitative and Implementation Methods Group green port in Turkey,” Journal of Engineering for the Maritime guidance series—article 6: reporting guidelines for qualitative, Environment, vol. 236, no. 2, pp. 504–518, 2021. implementation, and process evaluation evidence syntheses,” [35] H. Hong, H. Jeon, C. Youn, and H. Kim, “Incorporation of Journal of Clinical Epidemiology, vol. 97, pp. 79–85, 2018. shipping activity data in recurrent neural networks and long [20] A. Liberati, D. G. Altman, J. Tetzlaf et al., “Te PRISMA short-term memory models to improve air quality predictions statement for reporting systematic reviews and meta-analyses around busan port,” Atmosphere, vol. 12, no. 9, p. 1172, 2021. of studies that evaluate healthcare interventions: explanation [36] H. Agamy, M. Abdelgeliel, M. Mosleh, K. Elserafy, and and elaboration,” BMJ, vol. 339, 2009. N. A. N. Mohamed, “Neural fuzzy control of the indoor air [21] D. Moher, A. Liberati, J. Tetzlaf et al., “Preferred reporting quality onboard a RO–RO ship garage,” International Journal items for systematic reviews and meta-analyses: the PRISMA of Fuzzy Systems, vol. 22, no. 3, pp. 1020–1035, 2020. statement,” PLoS Medicine, vol. 6, no. 7, Article ID e1000097, [37] G. Song, B. Li, and Y. He, “Container terminal logistics generalized computing architecture and green initiative [22] V. Garousi, K. Petersen, and B. Ozkan, “Challenges and best computational pattern performance evaluation,” Information, practices in industry-academia collaborations in software vol. 10, no. 12, pp. 383–424, 2019. engineering: a systematic literature review,” Information and [38] P. Cammin, M. Sarhani, L. Heilig, and S. Voß, “Applications Software Technology, vol. 79, pp. 106–127, 2016. of real-time data to reduce air emissions in maritime ports,” in [23] S. Pahlevan-Sharif, P. Mura, and S. N. R. Wijesinghe, “A Design, User Experience, and Usability. Case Studies in Public systematic review of systematic reviews in tourism,” Journal of and Personal Interactive Systems. HCII 2020, A. Marcus and Hospitality and Tourism Management, vol. 39, pp. 158–165, E. Rosenzweig, Eds., Vol. 12202, Springer Cham, New York, 2019. NY, USA, 2020. Journal of Advanced Transportation 17 [39] J. Yu, G. Tang, X. Song et al., “Ship arrival prediction and its [55] L. Fredianelli, M. Bolognese, F. Fidecaro, and G. Licitra, value on daily container terminal operation,” Ocean Engi- “Classifcation of noise sources for port area noise mapping,” Environments, vol. 8, no. 2, p. 12, 2021. neering, vol. 157, pp. 73–86, 2018. [40] S. Jahangiri, N. Nikolova, and K. Tenekedjiev, “An improved [56] H. Du and Y. Jiang, “Backup or reliability improvement strategy for a manufacturer facing heterogeneous consumers emission inventory method for estimating engine exhaust in a dynamic supply chain,” IEEE Access, vol. 7, Article ID emissions from ships,” Sustainable Environment Research, 50419, 2019. vol. 28, no. 6, pp. 374–381, 2018. [57] A. Kara and I. Dogan, “Reinforcement learning approaches [41] B. Goldsworthy, H. Enshaei, and S. Jayasinghe, “Comparison for specifying ordering policies of perishable inventory sys- of large-scale ship exhaust emissions across multiple reso- tems,” Expert Systems with Applications, vol. 91, pp. 150–158, lutions: from annual to hourly data,” Atmospheric Environ- ment, vol. 214, Article ID 116829, 2019. [58] A. Kuhnle, J. Jakubik, and G. Lanza, “Reinforcement learning [42] Y. Peng, H. Liu, X. Li, J. Huang, and W. Wang, “Machine for opportunistic maintenance optimization,” Production learning method for energy consumption prediction of ships Engineering, vol. 13, no. 1, pp. 33–41, 2018. in port considering green ports,” Journal of Cleaner Pro- [59] M. F. Monaco, L. Moccia, and M. Sammarra, “Operations duction, vol. 264, Article ID 121564, 2020. Research for the management of a transhipment container [43] S. Wen, T. Zhao, Y. Tang et al., “Coordinated optimal energy terminal: the Gioia Tauro case,” Maritime Economics and management and voyage scheduling for all-electric ships Logistics, vol. 11, no. 1, pp. 7–35, 2009. based on predicted shore-side electricity price,” IEEE [60] Q. Tan, Y. Tong, S. Wu, and D. Li, “Modeling, planning, and Transactions on Industry Applications, vol. 57, no. 1, scheduling of shop-foor assembly process with dynamic pp. 139–148, 2021. cyber-physical interactions: a case study for CPS-based smart [44] Y. Zhan and T. Liu, “Weighted RPCA based background industrial robot production,” International Journal of Ad- subtraction for automatic berthing,” in Proceedings of the vanced Manufacturing Technology, vol. 105, no. 9, pp. 3979– Chinese Control Conference (CCC), pp. 3519–3524, Guangz- 3989, 2019. hou, China, July 2019. [61] G. Wong, T. Greenhalgh, G. Westhorp, J. Buckingham, and [45] M. Nastasi, L. Fredianelli, M. Bernardini, L. Teti, F. Fidecaro, R. Pawson, “RAMESES publication standards: meta-narrative and G. Licitra, “Parameters afecting noise emitted by ships reviews,” Journal of Advanced Nursing, vol. 69, no. 5, moving in port areas,” Sustainability, vol. 12, no. 20, p. 8742, pp. 987–1004, 2013. [62] Z. Zhu, N. Anwer, Q. Huang, and L. Mathieu, “Machine [46] F. Alasali, S. Haben, V. Becerra, and W. Holderbaum, “Day- learning in tolerancing for additive manufacturing,” CIRP ahead industrial load forecasting for electric RTG cranes,” Annals, vol. 67, no. 1, pp. 157–160, 2018. Journal of Modern Power Systems and Clean Energy, vol. 6, no. 2, pp. 223–234, 2018. [47] S. Fahdi, M. Elkhechaf, and H. Hachimi, “Machine learning for cleaner production in port of Casablanca,” Journal of Cleaner Production, vol. 294, Article ID 126269, 2021. [48] Z. Wang, Y. Shi, and S. Li, “Research on anti-swaying of crane based on t-s type adaptive neural fuzzy control,” in Pro- ceedings of the Chinese Control and Decision Conference (CCDC), pp. 5503–5508, Shenyang, China, June 2018. [49] A. Hill and J. W. Bose, ¨ “A decision support system for im- proved resource planning and truck routing at logistic nodes,” Information Technology and Management, vol. 18, no. 3, pp. 241–251, 2017. [50] L. Heilig, E. Lalla-Ruiz, and S. Voß, “Multi-objectiveinter- terminal truck routing,” Transportation Research Part E: Logistics and Transportation Review, vol. 106, pp. 178–202, [51] C. Caballini, M. D. Gracia, J. Mar-Ortiz, and S. Sacone, “A combined data mining – optimization approach to manage trucks operations in container terminals with the use of a TAS: application to an Italian and a Mexican port,” Transportation Research Part E: Logistics and Transportation Review, vol. 142, Article ID 102054, 2020. [52] H. Zhao, H. He, W. Lu, and Y. Hao, “Characterizing the variation of particles in varied sizes from a container truck in a port area,” Environmental Monitoring and Assessment, vol. 192, no. 12, p. 787, 2020. [53] K. Murphy, Machine Learning: A Probabilistic PerspectiveTe MIT Press, Cambridge, MA, USA, 2012. [54] M. Bertolini, D. Mezzogori, M. Neroni, and F. Zammori, “Machine Learning for industrial applications: a compre- hensive literature review,” Expert Systems with Applications, vol. 175, Article ID 114820, 2021. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Advanced Transportation Hindawi Publishing Corporation http://www.deepdyve.com/lp/hindawi-publishing-corporation/machine-learning-for-promoting-environmental-sustainability-in-ports-xab8m6KHRm

Loading next page...

References (53)

Claudio Quintano, Paolo Mazzocchi, A. Rocca (2020)
Examining eco-efficiency in the port sector via non-radial data envelopment analysis and the response based procedure for detecting unit segments
Journal of Cleaner Production
Alejandro Vega-Muñoz, Guido Salazar-Sepúlveda, J. Espinosa-Cristia, Jonathan Sanhueza-Vergara (2021)
How to Measure Environmental Performance in Ports
Sustainability, 13
Andreas Kuhnle, Johannes Jakubik, G. Lanza (2018)
Reinforcement learning for opportunistic maintenance optimization
Production Engineering, 13
A. Kara, I. Doğan (2018)
Reinforcement learning approaches for specifying ordering policies of perishable inventory systems
Expert Syst. Appl., 91
V. Garousi, K. Petersen, B. Özkan (2016)
Challenges and best practices in industry-academia collaborations in software engineering: A systematic literature review
Inf. Softw. Technol., 79
Ying Zheng, Jing-zhu Zhao, G. Shao (2020)
Port City Sustainability: A Review of Its Research Trends
Sustainability
S. Kuo, Pei-Chun Lin (2020)
Determinants of green performance in container terminal operations: A lean management
Journal of Cleaner Production, 275
Eduardo Lalla-Ruiz, Leonard Heilig, S. Voß (2019)
Environmental Sustainability in Ports
Sustainable Transportation and Smart Logistics
Soukaina Fahdi, Mariam Elkhechafi, H. Hachimi (2021)
Machine learning for cleaner production in port of Casablanca
Journal of Cleaner Production, 294
Marco Nastasi, Luca Fredianelli, M. Bernardini, Luca Teti, F. Fidecaro, G. Licitra (2020)
Parameters Affecting Noise Emitted by Ships Moving in Port Areas
Sustainability
H. Hong, H. Jeon, Cheong Youn, Hyeonsoo Kim (2021)
Incorporation of Shipping Activity Data in Recurrent Neural Networks and Long Short-Term Memory Models to Improve Air Quality Predictions around Busan Port
Atmosphere
M. Monaco, Luigi Moccia, Marcello Sammarra (2009)
Operations Research for the management of a transhipment container terminal: The Gioia Tauro case
Maritime Economics & Logistics, 11
Qingmeng Tan, Yifei Tong, Shaofeng Wu, Dongbo Li (2019)
Modeling, planning, and scheduling of shop-floor assembly process with dynamic cyber-physical interactions: a case study for CPS-based smart industrial robot production
The International Journal of Advanced Manufacturing Technology, 105
H. Davarzani, B. Fahimnia, Michael Bell, Joseph Sarkis (2016)
Greening ports and maritime logistics: A review
Transportation Research Part D-transport and Environment, 48
Lei Wang, Zhixiang Zhou, Yang Yang, Jie Wu (2020)
Green efficiency evaluation and improvement of Chinese ports: A cross-efficiency model
Transportation Research Part D-transport and Environment, 88
Alessandro Hill, Jürgen Böse (2017)
A decision support system for improved resource planning and truck routing at logistic nodes
Information Technology and Management, 18
Zuowei Zhu, N. Anwer, Qiang Huang, L. Mathieu (2018)
Machine learning in tolerancing for additive manufacturing
CIRP Annals, 67
A. Liberati, D. Altman, J. Tetzlaff, C. Mulrow, P. Gøtzsche, J. Ioannidis, M. Clarke, P. Devereaux, J. Kleijnen, D. Moher (2009)
The PRISMA Statement for Reporting Systematic Reviews and Meta-Analyses of Studies That Evaluate Health Care Interventions: Explanation and Elaboration
PLoS Medicine, 6
Likun Wang, Chen Peng, Wenming Shi, Mo Zhu (2020)
Carbon dioxide emissions from port container distribution: Spatial characteristics and driving factors
Transportation Research Part D-transport and Environment, 82
武彦福島 (2006)
持続可能性(Sustainability)の要件
Environmental Sciences, 19
Feras Alasali, S. Haben, V. Becerra, W. Holderbaum (2018)
Day-ahead industrial load forecasting for electric RTG cranes
Journal of Modern Power Systems and Clean Energy, 6
Hossam Agamy, M. Abdelgeliel, M. Mosleh, Kamel Elserafy, N. Mohamed (2020)
Neural Fuzzy Control of the Indoor Air Quality Onboard a RO–RO Ship Garage
International Journal of Fuzzy Systems, 22
Saeed Pahlevan-Sharif, P. Mura, Sarah Wijesinghe (2019)
A systematic review of systematic reviews in tourism
Journal of Hospitality and Tourism Management
Jingjing Yu, G. Tang, Xiangqun Song, X. Yu, Yue Qi, Da Li, Yong Zhang (2018)
Ship arrival prediction and its value on daily container terminal operation
Ocean Engineering
Sofia Samoili, M. Cobo, Emilia Gómez, G. Prato, Fernando Martínez-Plumed, Blagoj Delipetrev (2020)
AI Watch. Defining Artificial Intelligence. Towards an operational definition and taxonomy of artificial intelligence
Philip Cammin, Malek Sarhani, Leonard Heilig, S. Voß (2020)
Applications of Real-Time Data to Reduce Air Emissions in Maritime Ports
Y. Park, N. Ghani, Fesseha Gebremikael, Gokhan Egilmez (2018)
Benchmarking environmental efficiency of ports using data mining and RDEA: the case of a U.S. container ports
International Journal of Logistics Research and Applications, 22
Frederik Schulte, Eduardo Lalla-Ruiz, R. González-Ramírez, S. Voß (2017)
Reducing port-related empty truck emissions : A mathematical approach for truck appointments with collaboration
Transportation Research Part E-logistics and Transportation Review, 105
E. France, N. Ring, J. Noyes, M. Maxwell, R. Jepson, E. Duncan, R. Turley, D. Jones, I. Uny (2015)
Protocol-developing meta-ethnography reporting guidelines (eMERGe)
BMC Medical Research Methodology, 15
K. Flemming, A. Booth, K. Hannes, M. Cargo, J. Noyes (2017)
Cochrane Qualitative and Implementation Methods Group guidance series-paper 6: reporting guidelines for qualitative, implementation, and process evaluation evidence syntheses.
Journal of clinical epidemiology, 97
A. Tong, K. Flemming, E. McInnes, S. Oliver, J. Craig (2012)
Enhancing transparency in reporting the synthesis of qualitative research: ENTREQ
BMC Medical Research Methodology, 12
(2020)
HHLA container terminals use machine learning technology to boost productivity
S. Jahangiri, N. Nikolova, K. Tenekedjiev (2018)
An improved emission inventory method for estimating engine exhaust emissions from ships
Sustainable Environment Research
(2019)
Independent High-Level Expert Group on Artificial Intelligence Set up by the European Commission
Young-Tae Chang, Hyosoo Park, Suhyung Lee, Eun-soo Kim (2018)
Have Emission Control Areas (ECAs) harmed port efficiency in Europe
Transportation Research Part D-transport and Environment, 58
D. Eatough, Paul Cropper, William Keeton, E. Burrell, J. Hansen, R. Farber, J. Zack, R. Cary, P. Hopke (2020)
Apportionment of PM2.5 adjacent to the I-710 Harbor Freeway in Long Beach, CA
Journal of the Air & Waste Management Association, 70
D. Moher, A. Liberati, J. Tetzlaff, D. Altman (2009)
Preferred Reporting Items for Systematic Reviews and Meta-Analyses: The PRISMA Statement
PLoS Medicine, 6
B. Goldsworthy, H. Enshaei, S. Jayasinghe (2019)
Comparison of large-scale ship exhaust emissions across multiple resolutions: From annual to hourly data
Atmospheric Environment
S. Holly, Astrid Nieße, Martin Tröschel, Lasse Hammer, Christoph Franzius, V. Dmitriyev, J. Dorfner, Eric Veith, Christine Harnischmacher, Maike Greve, Kristin Masuch, L. Kolbe, Boris Wulff, Julian Kretz (2020)
Flexibility management and provision of balancing services with battery-electric automated guided vehicles in the Hamburg container terminal Altenwerder
Energy Informatics, 3
Jiasen Sun, Yang Yuan, Rui Yang, Xiang Ji, Jie Wu (2017)
Performance evaluation of Chinese port enterprises under significant environmental concerns: An extended DEA-based analysis
Transport Policy, 60
Sanghyun Cheon, A. Maltz, K. Dooley (2017)
The link between economic and environmental performance of the top 10 U.S. ports
Maritime Policy & Management, 44
Leonard Heilig, Eduardo Lalla-Ruiz, S. Voß (2017)
Multi-objective inter-terminal truck routing
Transportation Research Part E-logistics and Transportation Review, 106
Alireza Shakarami, Ali Shahidinejad, Mostafa Ghobaei-Arani (2021)
An autonomous computation offloading strategy in Mobile Edge Computing: A deep learning-based hybrid approach
J. Netw. Comput. Appl., 178
Yun Peng, Huakun Liu, Xiangda Li, Jian Huang, Wenyuan Wang (2020)
Machine learning method for energy consumption prediction of ships in port considering green ports
Journal of Cleaner Production, 264
Guanggang Song, Bin Li, Yuqing He (2019)
Container Terminal Logistics Generalized Computing Architecture and Green Initiative Computational Pattern Performance Evaluation
Inf., 10
Hong-Mei Zhao, H. He, Weizhen Lu, Yang-yang Hao (2020)
Characterizing the variation of particles in varied sizes from a container truck in a port area
Environmental Monitoring and Assessment, 192
G. Wong, T. Greenhalgh, G. Westhorp, J. Buckingham, R. Pawson (2013)
RAMESES publication standards: meta-narrative reviews.
Journal of advanced nursing, 69 5
Luca Fredianelli, Matteo Bolognese, F. Fidecaro, G. Licitra (2021)
Classification of Noise Sources for Port Area Noise Mapping
Environments
Ignacio Zarzuelo, María Soeane, Beatriz Bermúdez (2020)
Industry 4.0 in the port and maritime industry: A literature review
J. Ind. Inf. Integr., 20
Fatemeh Jazayeri, A. Shahidinejad, Mostafa Ghobaei-Arani (2020)
Autonomous computation offloading and auto-scaling the in the mobile fog computing: a deep reinforcement learning-based approach
Journal of Ambient Intelligence and Humanized Computing, 12
C. Caballini, M. Gracia, J. Mar-Ortiz, S. Sacone (2020)
A combined data mining – optimization approach to manage trucks operations in container terminals with the use of a TAS: Application to an Italian and a Mexican port
Transportation Research Part E-logistics and Transportation Review, 142
A. Ekmekçioğlu, K. Ünlügençoğlu, U. Çelebi (2021)
Container ship emission estimation model for the concept of green port in Turkey
Proceedings of the Institution of Mechanical Engineers, Part M: Journal of Engineering for the Maritime Environment, 236
G. Yeo, Shiau Lim, L. Wynter, Hifaz Hassan (2019)
MPA-IBM Project SAFER: Sense-Making Analytics for Maritime Event Recognition
INFORMS J. Appl. Anal., 49

Publisher: Hindawi Publishing Corporation
ISSN: 0197-6729
eISSN: 2042-3195
DOI: 10.1155/2023/2144733
Publisher site: See Article on Publisher Site

Abstract

Hindawi Journal of Advanced Transportation Volume 2023, Article ID 2144733, 17 pages https://doi.org/10.1155/2023/2144733 Research Article Machine Learning for Promoting Environmental Sustainability in Ports 1,2 1 3 Meead Mansoursamaei , Mahmoud Moradi , Rosa G. Gonza ´ lez-Ram´ırez , and Eduardo Lalla-Ruiz University of Guilan, Rasht, Iran University of Twente, Enschede, Netherlands Facultad de Ingenier´ıa y Ciencias Aplicadas, Universidad de los Andes, Santiago, Chile Correspondence should be addressed to Meead Mansoursamaei; m.mansoursamaei@utwente.nl Received 19 May 2022; Revised 12 September 2022; Accepted 10 February 2023; Published 3 March 2023 Academic Editor: Dongjoo Park Copyright © 2023 Meead Mansoursamaei et al. Tis is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Maritime transportation is one of the essential drivers of the global economy as it enables both lower transportation costs and intermodal operations across multiple forms of transportation. Maritime ports are essential interfaces that support cargo handling between sea and hinterland transportation. Besides, in this area, environmental protection is becoming extremely important. Global warming, air pollution, and greenhouse gas emissions are all having a detrimental infuence on the environment and will most likely continue to do so for future generations. Hence, there is a growing need to promote environmental sustainability in maritime-based transportation. Te application of machine learning (ML), as one of the main subdomains of artifcial intelligence (AI), can be considered a component within the process of digital transformation to advance green activities in maritime port logistics. Tus, this article presents the results of a systematic literature review of the recent literature on machine learning for promoting environmentally sustainable maritime ports. It collects and analyses the articles whose contributions lie in the interplay between three main dimensions, i.e., machine learning, port-related operations, and environmental sustainability. Troughout a review protocol, this research is constituted on the major focuses of impact, problems, and techniques to discern the current state of the art as well as research directions. Te research fndings indicate that the articles using polynomial regression models are dominant in the literature, and the recurrent neural network (RNN) and long short-term memory (LSTM) are the most recent approaches. Moreover, in terms of environmental sustainability, emissions and energy consumption are the most studied problems. mAccording to the research gaps observed in the review, two broad directions for future research are identifed: (i) altering attention on a greater diversity of machine learning approaches for promoting environmental sustainability in ports and (ii) leveraging new outlooks to perform more green practical works on port-related operations. growing demands of the logistics and transportation sectors 1. Introduction [3]. Expanding maritime transportation activities has enabled Maritime freight transportation is one of the vital drivers of the urban economies to prosper to some extent; however, they also global economy as it enables both lower transportation costs have caused resource waste and environmental pollution. To and faster intermodal operations across multiple forms of achieve the sustainable growth of ports and cities, energy- transportation [1, 2]. Indeed, maritime ports are the essential saving and emission-reduction methods have to be used [4, 5]. interfaces that support cargo handling between sea and hin- Te numerous sources and types of port-related emissions, terland. Besides, environmental sustainability has become one such as those from maritime vessels, trucks, and cargo- of the important foundations on the agenda of many maritime handling machines, have a considerable impact on the envi- ports due to the challenges of climate change as well as the ronment [6]. Moreover, a signifcant proportion of emissions 2 Journal of Advanced Transportation reviews. However, port sustainability in terms of environ- in ports are due to interterminal transportation and container drayage operations [7]. Terefore, the research on environ- mental impacts is currently experiencing new challenges; hence, the literature needs to be updated. In addition to the mental enhancements in maritime ports as main trans- portation nodes has grown in importance as they allow for previous reviews, the studies that have used machine improvements in a variety of areas of environmental sus- learning approaches as solutions to cut down on port-related tainability that governments and entrepreneurs are attempting pollution have been considered in this research. Concerns to address [3]. among stakeholders, including port ofcials, policymakers, Artifcial intelligence (AI) approaches and subdisciplines users, and citizens have grown over the environmental can be divided into two main categories: (i) reasoning and sustainability of ports and machine learning techniques as technological solutions. Terefore, to direct the future decision making and (ii) learning [8]. While the frst set is focused on decision making concerning planning, solution growth of this topic, a timely literature review is essential. Te article is organized into four sections. Section 2 searching, and optimization, the second set relates to learning, communication, and perception. In this sense, presents the research methodology by describing the data collection and analysis methods along with initial statistics learning refers to the ability of systems to automatically learn, determine, predict, adapt, and react to changes regarding recent publication trends in ML applications in without being explicitly programmed [9]. Te techniques green-port operations. Section 3 presents the initial bib- related to learning include neural networks, deep learning, liometric analysis. Ten, relying on the bibliometric analysis, reinforcement learning, and decision trees, among others the evolution of ML in port logistics is studied. First, the [8, 10], and the three main subdomains are supervised, evolving process of ML applied to aid sustainable maritime unsupervised, and reinforcement learning [11]. port operations based on the growing volume of publications over time is examined. Second, the impact and citation In this research, the application of ML as one of the main subdomains of AI to advance green activities in port logistics patterns to characterize the selected articles are presented. Ten, the most cited articles and researchers with a higher is studied. A supervised learning model as a subdomain of ML can learn and check the plausibility of schedules and number of contributions within the scope of the review up to th predict the energy consumption of battery-electricauto- October 2021 are presented. Finally, the evolution and guided vehicles (AGVs) in horizontal transportation areas interaction of topics, techniques, and problems are scruti- [12]. Neural networks are being used in the ports of nized. Te article concludes with a discussion on research Hamburg, Rotterdam, and Singapore. For instance, Ham- limitations and potential research directions. burger Hafen und Logistik AG (HHLA) has become one of the frst ports in the world to implement ML solutions for its 2. Research Methodology Hamburg container terminals to predict the dwell time of a container at the yard of the terminal [13]. Regarding Tere are several reporting guidelines to perform a sys- environmental concerns and emissions from maritime tematic literature review [16–18]. Such reporting guidelines, trafcs at the port of Rotterdam, machine learning tech- e.g., SRQR (standards for reporting qualitative research), niques have been developed to predict the estimated time of ENTEQ (enhancing transparency in reporting the synthesis arrival (ETA) for containerships to reduce waiting time and of qualitative research), eMERGe (meta-ethnography thus emissions [14]. Te SAFER project of the maritime and reporting guidelines), and RAMESES (realist and meta- port authority (MPA) of Singapore and IBM have imple- narrative evidence syntheses: evolving standards) are sug- mented seven ML-based modules within Singapore port gested for qualitative reviews [19]. Likewise, the PRISMA waters in order to predict arrival times of vessels and po- (preferred reporting items for systematic reviews and meta- tential congestion areas as well as to detect ofending ships to analyses) proposes a standardized method for ensuring environmental regulations [15]. Terefore, an ever- transparent and thorough systematic review reporting [20]. increasing demand for ML-based technologies to support PRISMA is a systematic review protocol that includes a 27- green ports is arising. Besides, several relevant works have item checklist and a four-phase fow diagram [21] (see been published recently in journals and proceedings. Hence, Figure 1). In this study, PRISMA was selected as the a systematic literature review (SLR) that analyses the articles methodology for the systematic review over other existing investigating green activities in maritime ports using ML protocols because of its comprehensiveness, its use in applications becomes necessary. a variety of disciplines, and its potential to improve con- Tis study with the purpose of investigating the appli- sistency across reviews. Tis protocol involves (i) the def- cations of machine learning techniques to stimulate envi- nition of research questions and (ii) the identifcation of the ronmental actions in maritime ports seeks to explore the search string and source selection. state-of-the-art research within the interplay of machine learning, environmental sustainability, and maritime ports. Tus, the current research contributes to the literature by 2.1. Defnition of Research Questions. Te frst step of this identifying research categories based on a systematic liter- review is the defnition of the research questions (RQs) ature review (SLR), i.e., impacts, problems, and techniques. related to the main research question of this work, i.e., how Most recently, the sustainability of ports is receiving machine learning has been used to promote environmentally more attention. Te number of research on the topic has sustainable port operations. Tis way, the following research increasingly expanded and been published in some literature questions are addressed: Journal of Advanced Transportation 3 Systematic Search (Databases): Full-Text Scopus (482) Screening + Web of Science (231) (62 papers) (i) Title (i) Full-text Context Final Result (ii) Abstract (ii) Focus on research of Search: (iii) Keywords questions (29 papers) Abstract Screening Snowball Technique (713 papers) (28 papers) (i) Abstract (i) Exploring forward and (ii) Focus on machine learning backward references of and environmental the selected papers sustainability in maritime (ii) Internal loop port logistics (iii) Removing duplicates Figure 1: Te overall procedure for selecting and fltering the articles [21]. (i) RQ1. What is the environmental impact of ML snowballing technique was applied to all the selected articles approaches on ports operations? (Impact related) to minimize the risk of missing relevant studies [22]. Tis resulted in addition of one more article leading to a total of (ii) RQ2. Which environment-related port operations 29 articles that were analyzed in this SLR. Te procedure was have ML techniques being used for? (Problem carried out by two of the authors of this manuscript who related) independently screened the selected articles. Finally, there (iii) RQ3. How ML techniques have been used to pro- was a consensus among all authors about the articles that mote environmentally sustainable port operations? were excluded [23]. Figure 1 indicates a fow diagram of the (Technique related) search procedure and the results obtained at each stage. 2.2. Identifcation of Search String and Source Selection. 3. Discussion and Results To achieve the fnal result of the search, several steps in- cluding the systematic search, abstract screening, full-text In order to answer the research questions raised in Section screening, and snowball technique were applied. Te Scopus 2.1, three subsections, each corresponding to a research and Web of Science (WoS) databases were utilized to collect question, are provided. First, the impact of selected articles is and select related references systematically. Te query used discussed, and the most cited works and authors are listed. in this work is composed of a list of keywords distributed Second, based on the physical area of port-related opera- into three main pillars (i.e., machine learning, maritime tions, the environmental problems of the selected articles are ports, and environmental sustainability) as shown in Fig- identifed and discussed. Tird, the machine learning-based ure 2. Te AND operator delimited each group, while the OR techniques that have been proposed in the literature to operator bundled all keywords within each group. Te AND address the environmental problems are outlined. NOToperator eliminated irrelevant subject areas. After that, the query was used in the abovementioned databases to fnd those articles considering the selected keywords within their 3.1. Impact Analysis (RQ1). With the aim of answering RQ1 titles, abstracts, or keywords from 2017 to 2021. By the initial involving impact, the number of works published so far, the search, 482 articles from Scopus and 231 articles from WoS number of citations, and data collection methods are were discovered from the databases; however, English ar- investigated. ticles were included as a limitation and irrelevant subject Te number of articles published annually from 2017 to areas such as agriculture, medicine, and chemical engi- 2021 by the source type is shown in Figure 3. Based on the neering were excluded. Moreover, since this review is fo- data, it is observed that the majority of works (25 articles) cused on ports, maritime shipping-related keywords were were published in journals, while the other 4 articles were th excluded. Te last search was run on 15 October, 2021. published in conference proceedings. Te Journal of Cleaner Production with four articles and Transportation Research Information of articles including the title, abstract, publication year, and sources were screened to determine Part D with three articles are the journals where more of the whether a article is included or not. Based on the three pillars selected articles were published, while the other works (76%) of our scope, there was one reason to discard articles, are from various sources. th namely, the articles whose contribution did not lie within the To illustrate the citation rate up to 15 October, 2021, interplay between machine learning, maritime port opera- Figure 4 displays a histogram. It can be observed that 25 of tions, and sustainability were removed. After the abstract the 29 collected articles have been cited so far. Te most cited and full-text screening, a total of 28 articles were collected. article, [24] with 44 citations was published in the Transport During the selection process, the forward and backward Policy. In this article, the performance of 17 ports in China 4 Journal of Advanced Transportation Maritime ports logistics (i) ("maritime" AND ("logistic*" OR "transport*" OR "port")) OR (("container" OR "cargo" OR "roro") AND ("terminal" OR "port" OR "harbor" OR "yard" OR "staking" OR "dispatching" OR "drayage" OR "railhead" OR "truck" OR "interchange area" OR "freight station")) OR ("yard" AND ("terminal" OR "operations" OR "storage" OR "truck" OR "management" OR "area")) OR "inter-terminal transport" OR "intermodal drayage" OR "seaport" OR "sea port" OR "bulk port" OR "landside" OR "seaside" OR "berth*" OR "crane" OR "stowage planning" OR "hinterland" OR "internal vehicles" OR "straddle carrier" OR "gantry carrier" OR "terminal gate" OR "gate area" OR "quay" OR "transport area" OR "horizontal transport system" OR "truck area" OR "train area" OR "transport area" OR "gate operations" OR "container storage" Machine learning (i) "machine learning" OR "active learning" OR "feature extraction" OR "adaptive learning" OR "generative adversarial network" OR "adversarial machine learning" OR "generative model" OR "adversarial network" OR "multi-task learning" OR "anomaly detection" OR "neural network" OR "artifcial neural network" OR "pattern recognition" OR "automated machine learning" OR "probabilistic learning" OR "automatic classifcation" OR "probabilistic model" OR "automatic recognition" OR "recommender system" OR "bagging" OR "recurrent neural network" OR "bayesian modelling" OR "recursive neural network" OR "gaussian process" OR "reinforcement learning" OR "classifcation" OR "semi-supervised learning" OR "clustering" OR "statistical learning" OR "collaborative fltering" OR "statistical relational learning" OR "content-based fltering" OR "supervised learning" OR "unsupervised learning" OR "convolutional neural network" OR "support vector machine" OR "data mining" OR (data?driven) OR "data pre-processing" OR "transfer learning" OR "deep learning" OR "unstructured data" OR "deep neural network" OR "ensemble method" OR "regression" OR "dimen* reduc*" OR "K-Means" OR "principal component analysis" OR "kernel" OR "decision tree" OR "K-Nearest" OR "random forest" OR "Q-Learning" OR "proximal policy" Environmental sustainability (i) "green" OR "emission*" OR "sustainab*" OR "environment*" OR "noise" OR "waste" OR "energy*" OR "pollution" OR "biofuel*" OR "biomass" OR "battery" OR "decarboni?ation" OR "greenhouse" OR "renewable" OR "solar" OR "recycle" OR "carbon" OR "footprint" OR "contamination" OR "ecolog*" OR "spill" OR "electr*" OR "smart grid*" AND NOT (i) "fsher*" OR "algae" OR "trout" OR "salmon" OR "mussel" OR "molecular" OR "herring" OR "larva" OR "oyster" OR ("sea AND bass'') OR "aerosol" OR "atmospheric chemistry" OR "atmospheric deposition" OR "troposphere" OR "sewage" OR "ozone" OR "microb*" OR "biology" OR "organism" OR "animal" OR "plants" OR "electron" OR "species" OR "eutrophication" OR "odor" OR "house" OR "building" OR "retinal" OR "steel" OR "fshing" OR "ore" Figure 2: Te used keywords based on the three pillars: maritime ports, machine learning, and environmental sustainability. 2017 2018 2019 2020 2021 Journal Conference Figure 3: Te number of articles published annually from 2017 to 2021 by source. under environmental concerns using regression models was observations, measurements, and experiments as primary evaluated. data sources. Nonetheless, three of the articles (10%) utilized In terms of data collection methods and data sources, 21 interviews and surveys solely or as a part of their data out of 29 articles (72%) used secondary data sources from collection method. earlier works, research institutes, or governments. However, Te environmental indicators used in port logistics in 7 of the articles (24%), researchers gathered data by applications include emissions, water pollution, noise Frequency Journal of Advanced Transportation 5 7 77 33 3 2 2 2 th Figure 4: Citations of the selected articles (the last update was on 15 October, 2021). pollution, solid waste, energy-saving, and renewable energy evaluation and energy-saving expenditures. Moreover, the were reported in the collected articles. Terefore, among authors of reference [27] evaluated the environmental ef- them, 25 articles (86%) raised emissions, 13 articles (45%) fciency of the Kaohsiung container port in Taiwan, con- considered energy-saving, and 6 articles (21%) used water sidering the same problems. In the same way, the authors of reference [28] benchmarked the top 10 ports in the US and pollution as the environmental indicator of their work. Nonetheless, noise pollution, renewable energy, and solid considered water pollution and emissions within port areas. Furthermore, the authors of reference [29] benchmarked waste, each with only one article (3%), are the smallest used environmental indicators. Figure 5 presents the contribution operating practices of 20 ports in the US for enhancing of the selected articles based on environmental sustainability environmental efciency in terms of greenhouse gas emis- indicators. sions, oil spill prevention, and energy efciency of port operations. Similarly, the authors of reference [30] bench- marked 24 container ports in Europe regarding emissions 3.2. Port-Related Problems (RQ2). To answer RQ2, the works and energy-saving problems. were categorized in terms of the port-related operations and To evaluate ports environmental efciency in emission their application areas in the port (i.e., seaside, yard, control areas (ECAs) (emission control areas (ECAs), as landside, and overall port areas). Te specifc machine outlined by Annex VI of the 1997 MARPOL protocol, are sea learning techniques of those works are later discussed in areas where regulations have been implemented to prevent Section 3.3. Terefore, based on the area within ports, the emissions from ships), 23 ports in the Baltic and the North articles have been distributed according to the location of the Sea and 25 non-ECA ports in Europe were investigated by addressed problems, which resulted in seaside operations [31] to examine the impact of ECA regulations on reducing with 28%, yard operations with 12%, and landside opera- emissions in European ports. Furthermore, using an In- tions with 16%. Nonetheless, 44% of the articles have tergovernmental Panel on Climate Change (IPCC) method, considered the environmental sustainability indicators for the authors of reference [32] measured CO emissions from benchmarking, performance evaluation, and air quality the port container distribution (PCD) to evaluate the sus- prediction not being specifcally focused on an area but the tainable development ability of 30 ports in China. Likewise, overall port. Hence, these contributions are categorized as the authors of reference [33] predicted air quality, fne the “overall port” in Figure 6. particulate composition, and mass in the area of Long Beach port in California. In addition, the authors of reference 3.2.1. Overall Port Areas. Several articles benchmarked and [34, 35] predicted air quality and emissions in 4 ports in Turkey (Ambarli, Izmir, Mersin, and Kocaeli ports) and evaluated ports in terms of environmental efciency and did Busan Port in Korea, respectively. Moreover, the authors of not consider a specifc area within port (i.e., seaside, yard, reference [36] simulated the indoor air quality of roll-on/ and landside), the reason for which they are classifed as roll-of (RORO) ships and predicted pollution emitted from “overall port areas.” In this regard, the authors of reference cars in maritime ports. Furthermore, the authors of refer- [24] evaluated 17 Chinese ports in terms of NOx emissions ence [37] developed a container terminal logistics general- and energy savings. Similarly, the authors of reference [25] ized computing architecture (CTL-GCA) for planning, investigated those environmental problems in 18 ports of scheduling and decision making to establish a better con- China. Using a diferent machine learning technique, the nection among liners and rubber-tired gantry cranes authors of reference [26] benchmarked 15 seaports in China (RTGCs) and block community to reduce carbon emissions. in terms of wasted water treatment as well as air quality Frequency Sun et al., 2017 Heilig et al., 2017 Cheon et al., 2017 Hill & Böse, 2017 Chang et al., 2018 Alasali et al., 2018 Jahangiri et al., 2018 Wang et al., 2018 Park et al., 2019 Goldsworthy et al., 2019 Song et al., 2019 Peng et al., 2020 Wang et al., 2020 Wang et al., 2020 Kuo & Lin, 2020 Nastasi et al., 2020 Caballini et al., 2020 Quintano et al., 2020 Agamy et al., 2020 Cammin et al., 2020 Eatough et al., 2020 Holly et al., 2020 Zhoa et al., 2020 Wen et al., 2021 Wang et al., 2021 6 Journal of Advanced Transportation 86% 45% 21% 3% 3% 3% Emissions Energy saving Water pollution Noise pollution Renewable energy Solid waste Environmental Indicators Figure 5: ML-based articles in terms of environmental sustainability in maritime port logistics. Landside 16% Yard 12% Overall port 44% Seaside 28% Overall port Seaside Yard Landside Figure 6: Distribution of selected articles based on application areas within the port. 3.2.2. Seaside Area. Te seaside operations that received renewable energy. For instance, the authors of reference [12] more attention are those regarding berth allocation plan- presented a study on how to use battery-electricauto-guided ning. For instance, to manage real-time data and air vehicles (AGVs) in the yard for handling containers in the emissions reduction in maritime ports due to the berth port of Hamburg (Germany). Tey utilized a synthetic case operations, [38] developed a predictive system for vessel by generating data for checking the plausibility of schedules arrivals, considering ship features and expanding estimated and predicting energy consumption. Container cranes are time of arrival (ETA) features to date, time, and weekday, also one of the main sources of energy consumption and based on the previous model presented by the authors of pollution in the yard. In this regard, the authors of references reference [39]. Te authors of reference [40] simulated berth [46, 47], and [48] considered environmental problems (i.e., planning problems and predicted the arrival time of vessels energy consumption and emissions) of the rubber-tyred using machine learning techniques. In the same way, the gantry (RTG) in the port of Felixstowe (UK) and port of authors of references [41–43] and [44] used case studies to Casablanca (Morocco) and a synthetic case, respectively. solve berth planning problems at diferent ports that are presented in detail in Table 1. From the other perspective, 3.2.4. Landside Area. Trucks are the main source of emis- noise emissions by ships around the port areas are one of the sions in the landside area. Hence, several studies have paid important issues for port cities. Tis has been studied in [45] attention to the environmental problems caused by trucks in where the authors with machine learning techniques the area. For instance, the authors of reference [49] proposed identifed the afecting parameters of noise emitted by ships a forecasting engine for truck arrivals to logistics nodes, i.e., in the industrial port of Livorno, Italy. empty container depots, packing facilities, or terminals, to mitigate greenhouse gas emissions from truck congestions 3.2.3. Yard Area. Several contributions related to container beyond the gates of an empty container depot in northern and cargo handling in the yard area to mitigate greenhouse Germany. Based on the proposed model, companies can gas emissions, energy-saving, and promoting the use of adjust their route planning to minimize truck waiting times. Percentage of use (%) Journal of Advanced Transportation 7 Table 1: Problems and ML techniques for promoting environmental sustainability in maritime port logistics. Areas Environmental problems Machine learning Data Research Case Articles collection Overall Water Noise Energy Renewable Solid Techniques scope studies Seaside Yard Landside Emission Input Output methods port pollution pollution saving energy waste (tools) Port performance Port assets, berth quantity, and Net proft, cargo throughput, and Secondary [24] ● ● PR 17 Chinese ports evaluation geographical location NOx emissions data sources A smoothed graph for the Port performance Facility, vessel and other pollution Secondary [28] ● ● ● KDE distribution of pollution incidents 10 American ports evaluation incidents data sources probability density Historical data, truck arrival time, administrative waiting start and end time; intermediate waiting start and Waiting time, arrival rates that An empty container end time; node-specifc forecasting Secondary [49] ● Truck scheduling ● NN (BP) translates into a reduction of trafc depot in Northern parameters, e.g., dispatching modes data sources congestion and air pollution Germany and storage policies; and external forecasting parameters, weather and trafc information Number of clusters and the archive Secondary Port of Hamburg [50] ● Truck scheduling ● K-means Cluster centroids containing n solutions data sources (Germany) Berth length, the number of cranes, terminal area for the efciency Port performance TEUs handled and the impact of Secondary [31] ● ● estimation. City gross domestic PR 48 ports in Europe evaluation emissions control regulations data sources product, variance infation factors, and emissions control regulations Te average of the previous day load, Yard crane the average of the previous week load, NN (BP) and Primary data Port of Felixstowe in [46] ● ● RTG crane demand of one hour demand the same hour load for previous day, SVM sources the UK and the previous hour load Maximum continuous rate (MCR) Primary data Two ocean-going [40] ● Berthing ● PR Emissions (NO , SO , CO , and CO) x x 2 measured by megawatt, shaft speed sources vessels in Australia Trolley position, trolley speed, loading Secondary [48] ● ● Antiswing crane ● ● NN (ANFIS) Te driving force of the trolley — angle, and angular velocity data sources Number of quay crane, acres, berth Port performance and depth, undesirable output (CO ), Clusters of decision-making units Secondary 20 American [29] ● ● ● ● SOM evaluation and desirable outputs (calls, (DMUs) data sources container ports throughput, and deadweight tonnage) Ship identifcation, position, speed, Ports of Newcastle, Emissions (NO , SO , PM , VOC, Primary data x 2 2.5 [41] ● Berthing ● course, heading and navigational PR Jackson, Botany, and CO, NH , CO, N O, and CH ) sources 3 2 4 status, and timestamp Kembla in Australia Te background with low-rank Primary data [44] ● Berthing ● ● Observed video PCA property and the foreground with Unknown sources sparse property RTGC number, block number, handling container specifcation, A container terminal Port performance Resource allocation for container Secondary [37] ● ● ● stevedoring full or empty category, K-means on the east coast of evaluation terminals data sources handling volume for a task, and the China number of clusters Te net tonnage, deadweight tonnage, GBoost, RF, NN Secondary [42] ● Berthing ● actual handling volume, and efciency (BP), PR, and Energy consumption Jingtang port (China) data sources of facilities KNN CO emission driver factors of the city where the port is located are gross domestic product, total resident population, the number of port berth, total imports, total exports, the frst Clusters of similar ports in terms of Port performance industrial value, the secondary environmental sustainability (LISA Secondary 30 Chinese container [25] ● ● ● ● Spatial clustering evaluation industrial value, the primary industrial cluster maps of PCD carbon data sources ports value, gross industrial production, emissions) fxed assets investment in the tertiary industry, per capita income, railway freight volume, highway freight volume, and waterway freight volume 8 Journal of Advanced Transportation Table 1: Continued. Areas Environmental problems Machine learning Data Research Case Articles collection Overall Water Noise Energy Renewable Solid Techniques scope studies Seaside Yard Landside Emission Input Output methods port pollution pollution saving energy waste (tools) Number of berths, the length of the Cargo throughput, NO emissions, Port performance Secondary [32] ● ● ● ● terminal, the number of staf, and the PR SO emissions, and solid waste 18 Chinese ports evaluation data sources total fxed assets containers Determinant factors of the survey are lean management, green operational Highly correlated input variables PCA practices, green behavior (green participation and green compliance), Port performance and green climate Kaohsiung container [27] ● ● ● ● Survey evaluation Determinant factors of the survey are port (Taiwan) lean management, green operational Green performance (fnancial and practices, green behavior (green PR nonfnancial) participation, green compliance), and green climate Noise of moving Draught, speed, and Primary data Industrial port of [45] ● ● PR Sound emitted ships in port areas ship-to-microphone distance sources Livorno (Italy) Container features are cycle, type, weight, special (e.g., hazard shipping), agreement (between stakeholders), Secondary Port of Altamira Hierarchical [51] ● Truck scheduling ● vessel departure time, distance (of two Container groups data sources, (Mexico) and Port of clustering containers in the yard), customs survey Genoa (Italy) clearance, dwell time, and fnal destination Total gross weight of goods, air Port performance Energy consumption and number of Hierarchical Secondary 24 European [30] ● ● ● pollutant emissions, and the rank of evaluation employees clustering, PR data sources container ports ports in terms of eco-efciency Indoor air quality A liner between CO concentration and load (number of Te reference fow rate of the Secondary [36] ● prediction ● ● NN (BP) Egypt and Saudi cars) ventilation system data sources (RORO) Arabia ports ETA features (date, time, and Secondary [38] ● Berthing ● weekday) and ship features (ship type SVM Arrival time of vessels — data sources and length) Air quality Fine particulate mass and fne Primary data [33] ● ● PR Air quality Long Beach (US) prediction particulate composition sources Scheduled arrival, departure, and load/ unload start time, planned berthing Primary data Hamburg container [12] ● AGV ● ● place, planned position of front and NN (BP) Availability of AGVs sources terminal (Germany) rear of the ship, and number of containers to load and unload Principal components (container Container truck Highly correlated data of trafc and Secondary Waigaoqiao port [52] ● ● PCA truck volume, other vehicles volume, emissions particle number concentrations (PNC) data sources (China) and PNC data) Ports of Ambarlı, Air quality Type of pollutant, the operating mode, Emissions (SO , NO , CO , VOC, Secondary 2 x 2 [34] ● ● PR Izmir, Mersin, and prediction and gross tonnage of ships PM, and CO) data sources Kocaeli (Turkey) Energy consumption of hoist, gantry, Secondary Casablanca port [47] ● ● RTG crane ● ● PR General energy consumption of RTG and trolley data sources (Morocco) Air quality Meteorological data, air quality data, Emissions (PM , PM , SO , O , Secondary 2.5 10 2 3 [35] ● ● RNN and LSTM Busan port (Korea) prediction and shipping activity data NO , CO) data sources Hourly data of energy (electricity) LSTM, NN (BP), Secondary A navigation route in [43] ● Berthing ● ● Day-ahead prices of energy prices and load demands Elman, RBF data sources Australia Air quality, rate of treatment of wastewater, standard-reaching rate of nearshore water, green coverage Highly correlated input variables PCA rate in developed areas, and expenditure on energy-saving Secondary Port performance [26] ● ● ● ● investments per capita data sources, 15 Chinese seaports evaluation Air quality, rate of treatment of survey wastewater, standard-reaching rate of Hierarchical Te rank of ports based on nearshore water, green coverage rate in clustering environmental sustainability features developed areas, and expenditure on energy-saving investments per capita Journal of Advanced Transportation 9 Figure 7. As can be observed in the fgure, supervised Furthermore, considering truck emissions in the port of Hamburg, the authors of reference [50] developed a multi- learning with 70% is the most used type of technique while unsupervised learning is the second option with 30% of objective model for interterminal truck routing problems and utilized a machine learning technique as part of the collected articles. Particularly, polynomial regression (PR) decision support system. Moreover, using two real container with 30% and neural networks (NN) with 27.5% are the most terminals, i.e., the port of Altamira (Mexico) and the port of used tools. Terefore, according to the main categorization Genoa (Italy) as case studies, the authors of reference [51] of ML techniques discussed in Section 3.3, there is no article proposed a methodological framework to reduce empty using classifcation nor reinforcement learning among the truck trips to minimize the deviation from their preferred collected articles. time slots and turnaround times in container terminals and reduce emissions. Te authors of reference [52] studied the 3.3.1. Supervised Learning. Supervised learning, commonly relationship between trafc volume and the particle number called predictive learning, is used for labelled datasets in concentrations (PNC) caused by emissions of container which the response of a scenario or example is known [53]. It trucks in the port of Waigaoqiao (China). For this, they enables several regression tools (e.g., polynomial regression, combined a machine learning technique with statistical neural networks, and k-nearest neighbour) for predicting methods to characterize the variation of particles in the the behaviour of a dataset. Moreover, some classifcation port area. tools (e.g., Bayesian network, logistic regression, and de- cision tree) are other applications of supervised learning 3.3. ML Techniques to Promote Green Port Operations (RQ3). when the output is categorical [54]. As seen in Table 1, only Researchers or practitioners who seek to apply ML in regression-related algorithms have been developed in the scope of this review. maritime port operations should possess the fundamental competency of selecting an algorithm that is appropriate for Regression is a supervised learning technique that aims to identify the correlation between variables and predict the a given task or problem. However, conceptualizing a way toward using ML to improve the performances of port continuous output variables based on one or more predictor variables. In this regard, to evaluate port efciency in terms operations is challenging in the absence of expertise or prior research of a similar type, especially when taking into ac- of environmental problems, [24] used polynomial regression count the numerous algorithms that have been ofered in the for predicting the amount of NOx emission based on port technical literature. assets, berth quantity, and geographical location of 17 port ML is mainly classifed into three diferent types, i.e., enterprises in China. Te authors showed a beneft of the supervised learning, unsupervised learning, and re- polynomial regression model for ports performance eval- inforcement learning [53]. Given that division, a systematic uation and found that the medium-sized and large-scale ports should focus on emissions reduction compared to literature review by [54] illustrated the machine learning techniques used in industrial applications so far which are small-sized ports that should focus on improving the service level and full resource utilization. Similarly, [25] considered organized in Table 2. Te diferent tools related to (i) su- pervised learning with classifcation and regression algo- the number of berths, the length of the terminal, the number rithms, (ii) unsupervised learning with clustering and of staf, and the total fxed assets of 18 ports in China as the dimensionality reduction algorithms, and (iii) re- input variables of their regression model to predict NOx and inforcement learning, are presented in the table. In order to SOx emissions as well as solid waste and energy con- highlight current and emerging trends and, more impor- sumption in the selected ports. Based on the results of tantly, to guide researchers or practitioners in the selection a regression model, the authors found that economic de- velopment positively impacts green efciency. Reference of ML techniques, the table demonstrating the sub- classifcation of ML algorithms is used in this review to map [31] used berth length, the number of cranes, terminal area, and amount of cargo handled as the independent variables of techniques when analyzing the collected articles. For further information on the tools, see the reference of the table. their model for predicting emissions as well as port per- formance evaluation. Te authors found, by applying a re- To provide a better vision of the problems (RQ2) and the techniques (RQ3) discussed in this SLR, Table 1 is presented. gression model, that although ECA regulation reduces Table 1 summarizes the main characteristics of the reviewed emissions, it signifcantly harms port productivity due to articles based on the following categories: the application area losing cargoes. within the port (i.e., seaside, yard, landside, and overall port), For evaluating the port of Kaohsiung in Taiwan in terms the research scope, environmental problems (i.e., emissions, of environmental sustainability, [27] used several input water pollution, noise pollution, energy saving, renewable variables including lean management, green operational practices, green behaviour (green participation and green energy, and solid waste), the machine learning technique, involved factors (input and output), the data collection compliance), and green climate to predict green perfor- mance (fnancial and nonfnancial). Using the results of method, and the case study. Te information provided in this table is discussed in Sections 3.1, 3.2, and 3.3. a regression model, the authors concluded that lean man- agement positively impacted green operations and green Considering the ML techniques used in the collected articles, a hierarchical categorization including the ML behaviour. Green operational practices had a positive in- techniques, the subdomains, and the tools is shown in fuence on both green behaviour and green performance. 10 Journal of Advanced Transportation Table 2: Categorization and used machine learning techniques in industrial applications [54]. ML domain ML subdomains Algorithms Tools (i) K-means (i) Spatial cluster (SC) (i) Local outlier factor (LOF) Clustering (ii) K-median (ii) Gaussian mixture model (ii) Neighbour-based clustering (NBC) (iii) Hierarchical clustering (HC) (GMM) (iii) Parzen windows (PW) Unsupervised (i) t-distributed stochastic neighbour (i) Principal component analysis (i) Kernel principal component learning embedding (t-SNE) Dimensionality (PCA) analysis (K-PCA) (ii) Uniform manifold approx. and projection reduction (ii) Linear discriminant analysis (LDA) (ii) Singular value decomposition (UMAP) (iii) Kernel density estimator (KDE) (SVD) (iii) Self-organizing maps (SOM) (i) Neural networks (NN) (i) NN, multilayer perception (ii) NN, back propagation (BP) (MLP) (iii) NN, convolutional neural network (ii) NN, radial basis function (RBF) (i) Locally weighted regression (LWR) (CNN) (iii) NN, recurrent neural network (ii) Support vector machine (SVM)- regressor (iv) NN, extreme learning machine (RNN) (iii) Gradient boosting (GBoost) Regression (ELM) (iv) Linear regression (LR) (iv) Random forest (RF)- regressor (v) NN, long-short term memory (v) Polynomial regression (PR) (v) K-nearest neighbor (KNN)-regressor Machine (LSTM) (vi) Fuzzy regression (FR) (vi) Gaussian process regression (GPR) learning (vi) NN, deep learning (DL) (vii) Bayesian regression (BR) Supervised learning (vii) NN, adaptive neuro-fuzzy (viii) Lasso regression (LASSO) inference system (ANFIS) (i) Adaptive support vector (i) Decision tree (DT) machine (ASVM) (i) K-nearest neighbor (KNN) (ii) Gradient boosting (GBoost) (ii) Learning vector quantization (ii) Quadratic discriminant analysis (QDA) (iii) Naive bayes (NB) (LVQ) (iii) Random forest (RF) Classifcation (iv) Bayesian network (BN) (iii) Linear discriminant analysis (iv) Logistic regression (LogR) (v) Kernel method (KM) (LDA) (v) Pattern recognition (PattR) (vi) Multi-layer perception (MLP) (vi) Support vector machine (SVM) (iv) Stochastic gradient descent (SGD) (i) Approximate dynamic (i) Adaptive heuristic critic (AHC) programming (ADP) (i) State-action-reward-state-action (SARSA) Reinforcement (ii) Deep deterministic policy gradient (ii) Proximal policy optimization (ii) Temporal diference learning (TD) learning (DDPG) (PPO) (iii) Trust region policy optimization (TRPO) (iii) Q-learning (QL) (iii) Deep Q- learning (DQL) Journal of Advanced Transportation 11 2 2 2 1111 111 1 11 K- PR NN SVM GBoost RF KNN HC means SC PCA KDE SOM (30%) (27.5%) (5%) (2.5%) (2.5%) (2.5%) (7.5%) (5%) (2.5%) (10%) (2.5%) (2.5%) Regression Classifcation Clustering Dimensionality (70%) (0%) (15%) Reduction (15%) Supervised Unsupervised Reinforcement (70%) (30%) (0%) Machine Learning Techniques Figure 7: Categorization of ML techniques (subdomains, algorithms, and tools) for promoting environmental sustainability in port logistics. Reference [30] used the total energy consumption of ports [49] developed a forecasting engine for truck arrivals to and the number of employees as the input variables of their logistics nodes, i.e., empty container depots and packing model to predict the total gross weight of goods, air pollutant facilities or terminals that mitigate greenhouse gas emissions emissions, and the eco-efciency rank of ports. With a re- from truck congestions in the landside. In doing so, they proposed a neural network model by taking historical data of gression model, they revealed that the energy consumption variable had a signifcant diverse correlation with the eco- truck arrival time, administrative wait time, intermediate wait time, node-specifc forecasting parameters (e.g., dis- efciency of ports. Moreover, [33] used the fne particulate mass and the fne particulate composition as the input patching modes and storage policies), and external fore- variables of their model to predict air quality. Te authors casting parameters (e.g., weather information and trafc concluded that polynomial regression models provided information) as the inputs. Te authors showed the beneft useful analysis for air quality management. of neural networks in the smoothed peak workloads at the Te authors of reference [34] considered the type of nodes due to adaptive truck routing and reduced pollutant, the operating mode, and the gross tonnage of waiting times. ships to predict the amount of emission. Based on the re- Intending to manage the energy consumption of RTG gression analysis, they found that innovative methods cranes, the authors of reference [46] utilized neural networks and a support vector machine and considered the average of proposed by the International Maritime Organization (IMO) such as carbon capture and storage systems, in- the previous day load, the average of the previous week load, creasing energy efciency, and emissions converting tech- the same hour load for the previous day, and the previous nologies had a signifcant impact on emissions reduction. hour load as the input variables of their model to predict Te authors of reference [35], by using long short-term RTG crane demand of one hour. Tey revealed that the memory (LSTM) and the recurrent neural network (RNN), efectiveness of the neural networks model was signifcantly used meteorological data, air quality data, and shipping high when the estimation of the number of crane moves and activity data as the input variables to predict emissions in container gross weight was accurate. Furthermore, to predict ports. Te authors indicated that besides meteorological data the general energy consumption of RTGs, the authors of reference [47] proposed a regression model based on the and air quality data, ship activities, as one of the main sources of emissions in port areas, should be considered in energy consumption of hoist RTG, gantry RTG, and trolley RTG. Te authors showed huge air pollution decrease and the prediction model to enhance the performance. Using neural networks, the authors of reference [36] developed cost-saving on energy by the forecasting model. In other a predictive model for controlling the CO concentration in work, through an adaptive neuro-fuzzy inference system RORO ships indoors. Tey considered CO concentration, (ANFIS), the authors of reference [48] developed a model to load (number of cars), and the reference fow rate of the minimize swings of RTG during loading/unloading of ventilation system as the input variables of their model. Te containers and cargo in the yard and seaside area which leads authors concluded that neural networks models combined to prevent emission of hazardous materials into the air and with other methodologies such as fuzzy controlling and water. Tey considered trolley position, trolley speed, particle swamp optimization signifcantly guarantee the loading angle, angular velocity, and the driving force of the trolley as the input variables of their neural network model. robustness of the indoors CO concentration reduction in RORO ships to an allowable limit. Te authors of reference Te authors showed that the ANFIS control method was BP (12.5%) LSTM (5%) Elman (2.5%) RBF (2.5%) RNN (2.5%) ANFIS (2.5%) 12 Journal of Advanced Transportation sources of noise from ships [55]. For this sake, the authors of robust and quick-response under diferent rope lengths and working conditions, but not reliable enough when the noise reference [45] proposed a regression model based on draught, distance from a recording microphone, and speed to evaluate was strong. In the same area, to predict the demand for battery-electric AGVs, the authors of reference [12] took the the correlations among the variables and predict sound scheduled arrival, departure and work started, planned emitted from moving ships in port areas. Te authors in- berthing place, the planned position of front and rear of the dicated that ship draught was not an infuencing parameter for ship, and the number of containers to load and unload as the noise emissions. Also, the authors concluded that for the noise input variables of their neural networks model. Te authors assessment in port areas, the right placement of the noise reported a beneft of the neural networks model for checking source that provides precise input data plays an essential role in improving the output of an acoustic model. the availability of AGVs in the horizontal transportation area of ports. In the seaside area, the authors of reference [40] pre- dicted the engine exhaust emissions using the polynomial 3.3.2. Unsupervised Learning. Unsupervised learning or descriptive learning is used for unlabeled datasets that the regression based on the maximum power output and shaft speed of ships during berthing operations. Tey proposed response to a scenario or example is unknown [53]. It en- a forecasting model that was signifcantly accurate for dif- ables several clustering tools (e.g., hierarchical clustering, k- ferent engine types at berth, manoeuvring, and sea. More- means, and fuzzy c-means) for recognizing the behaviour of over, to develop the berth allocation planning to manage a dataset that there is no historical output. Moreover, some real-time data and air emissions reduction, [38] used ETA dimensionality reduction tools (e.g., principal component features (date, time, and weekday) and ship features (ship analysis and self-organizing maps) are known as other applications of unsupervised learning that minimize the type and length) as the inputs of their support vector ma- chine model to build a predicting system for vessel arrivals. volume of datasets to an efcient computation process. Te authors concluded that the use of additional features (e.g., weekday) and discarding irrelevant inputs (e.g., the (1) Clustering. Clustering is an unsupervised learning technique that considers a set of selected features to group shipping line) have a positive infuence on the performance of the SVM model. With the same purpose, using a re- objects with similar attributes. Te purpose of the clustering gression model, [41] utilized ship identifcation, position, technique is to construct clusters where data objects within speed, course, timestamp, heading, and navigational status the same cluster are similar to one anther but diferent from of ships. Tey reported the beneft of regression analysis to the objects in other clusters. In this regard, for bench- model the spatial extent (the active area) of the emissions at marking ports in terms of environmental sustainability, diferent temporal resolutions (hourly and daily). In a sim- hierarchical clustering is used to create dendrograms or cluster trees. For instance, the authors of references [26, 30] ilar work, using the net tonnage, deadweight tonnage, actual handling volume, time of ships arrival, and efciency of evaluated and benchmarked several ports in Europe and China considering energy consumption, rate of wastewater facilities as the input variables, [42] used fve regression tools of the machine learning technique including gradient boost treatment, standard-reaching rate of nearshore water, the (i.e., Gboost), backpropagation neural network, linear re- green coverage rate in developed areas, and expenditure on gression, k-nearest neighbour, and random forest to predict energy-saving investments per capita. Based on the revealed energy consumption and emissions from ships during clusters, they both concluded that the ports in the same berthing operation. Te authors found that the time of ships cluster with the best performance in terms of technical ef- arrival without infuencing the performance of the model fciency showed a better eco-efciency performance than could be eliminated to reduce the difculty of data collec- other clusters. tion. Tey also concluded that when the efciency of fa- Spatial clustering by splitting spatial data into a series of meaningful subclasses aims to consider the selected features cilities was doubled, the energy consumption of ships was reduced by 34.17% at berth and 8.41% in overall port areas. to group spatial objects in the same cluster that are similar to each other and dissimilar to those in diferent clusters. In this To manage the energy consumption of an all-electric ship (AES), the authors of reference [43] proposed several regard, to investigate the spatial characteristics of emissions neural network-based models including the Elman, back- from the port container distribution (PCD), the authors of propagation (BP), the radial basis function (RBF), and long reference [32] used this type of clustering and considered short-term memory (LSTM). Tey considered hourly data of several parameters such as CO emission driver factors of the electricity prices and load demands as the input variables. city where the port is located, gross domestic product, total Te authors revealed that the LSTM method could predict resident population, the number of port berths, total im- the hourly price of electricity onshore accurately. As a result, ports, total exports, the frst industrial value, the secondary the method combined with an optimization model resulted industrial value, the primary industrial value, gross in- in the minimum cost and emission of the AES. dustrial production, fxed assets investment in the tertiary Ship sources, one of the main sources of noise emissions in industry, per capita income, railway freight volume, highway the port area (i.e., roads, railways, ships, port activities, and freight volume, and waterway freight volume. With the industrial plants), are from all the activities related to the spatial clustering technique, they reported that ports with movement and stationing of ships. Engines, funnels, and similar geographical locations showed a similar pattern of ventilation, as well as transit in port regimes, are the main PCD carbon emissions. Journal of Advanced Transportation 13 In the yard area, to manage trucks operations in con- climate to evaluate ports in terms of green performance tainer terminals and reduce empty truck trips, it is important (fnancial and nonfnancial). Tey showed the beneft of PCA as a data preprocessing tool for fnding the relationship to identify container features. In doing so, the authors of reference [51] performed a dendrogram and clustered between principal components of an equation in polynomial containers based on several input variables including the regression models. Moreover, for ranking ports based on cycle of import/export, the ISO type of the container, the several environmental sustainability factors, [26] utilized weight of the container, special (e.g., hazard shipping), the PCA to identify the independent variables. Tey proposed agreement between stakeholders, vessel departure time, the the rate of treatment of wastewater, the standard-reaching distance of two containers in the yard, customs clearance for rate of nearshore water, the green coverage rate in developed import/export, the dwell time of the container, and the fnal areas, and expenditure on energy-saving investments per destination of container in the hinterland. Te authors capita as the input variables. Te work concluded that PCA showed the beneft of container clustering in terms of re- helped in reducing the dimension of indicators when ducing the number of trucks for moving the same number of combined with a hierarchical clustering model. Te authors of reference [44] developed a computer vision-based model containers. In the same area, the authors of reference [37] developed a k-means model based on the number of RTGs, to detect ships entering into the imaging area at the seaside blocks, handling container specifcations, stevedoring full or and help them with automatic berthing. Tey considered the empty category, handling volume for a task, and the number observed video as the input variable of the PCA and sep- of clusters to manage the relationship between RTG crane arated the foreground object (ship) from the background teams and the given block sets. Te authors concluded that scene of each video frame as the outputs. Te authors re- the k-means model was an efcient tool for clustering block ported the beneft of PCA for reducing the dimensions of communities and dispatching RTG cranes in the yard area. image features. Moreover, to identify the relationship be- Furthermore, the authors of reference [50] utilized a k- tween the trafc and particle number concentrations (PNC) means as part of a decision support system to provide data from container truck emissions in the yard, the authors representative solutions for a multiobjective interterminal of reference [52] applied the PCA and proposed container truck volume, other vehicles’ volume, and PNC data as truck routing problem. Tey used the number of clusters and a solution archive as the inputs and the cluster centroids as uncorrelated variables for characterizing the variation of the output of the model. Te authors showed that using k- particles. Tey found that the method had a high perfor- means inside their multiobjective algorithm was a suitable mance in dimensionality reduction when combined with clustering approach for reducing the set of solutions and, a Pearson correlation analysis. Tey also concluded that thus, making the decision process more manageable. dimensionality reduction signifcantly reduced the com- putation cost and data collection difculties. (2) Dimensionality Reduction. Dimensionality reduction in machine learning is a data preprocessing technique that 3.4. Advantages and Disadvantages. Regarding Table 1 that refers to reducing the number of input variables in a dataset summarized the main characteristics of the reviewed articles to minimize computational costs and increase speed [53]. It based on the following categories: the application area enables several tools (e.g., principal component analysis within the port (i.e., seaside, yard, landside, and overall (PCA), kernel density estimator (KDE), and self-organizing port), the research scope, environmental problems (i.e., map (SOM)) for dataset volume reduction [54]. emissions, water pollution, noise pollution, energy saving, Regarding benchmarking and evaluating ports, [29] used renewable energy, and solid waste), the machine learning SOM combined with a data envelopment analysis (DEA) to technique, involved factors (input and output), the data propose similar groups of ports in terms of environmental collection method, and the case study, Table 3 reports the efciency. In doing so, they took the number of quay cranes, advantages and disadvantages of the used ML techniques in acres, berth, depth, the number of calls, throughput, the collected articles. deadweight tonnage, and CO emissions as well as the inputs of the SOM. Te authors concluded that the SOM was a suitable tool for reducing the dimensions of the features to 3.5. Open Issues and Future Works. Tis subsection discusses some important future research directions and open view- a simple visualized map. For the same purpose, the authors of reference [28] implemented the KDE tool to measure points toward the application of ML techniques in the environmental sustainability of maritime ports. Future re- pollution incidents intensity among the ports. Te KDE selected a common distribution and estimated the param- search may concentrate on analyzing the efectiveness of eters for the density (intensity) function from the data various machine learning (ML) algorithms for reducing the sample. Te authors indicated that KDE was an accurate tool impact of environmental concerns related to solid wastes or for generating a smooth surface variation and mapping the noise pollution near port cities, as there are numerous spatial distribution of ports pollution probability density of pollutants but few solutions for those issues as of yet. Te use a geographic area. Reference [27] applied PCA for reducing of ML approaches to help renewable energy technology is the dimension of highly correlated input variables of another important study area that can be tackled by future a survey to a few independent factors including lean academics. Te use of optimization techniques like meta- heuristics, mathematical programming, and heuristic ap- management, green operational practices, green behaviour (green participation and green compliance), and green proaches to aid decision making for planning and carrying 14 Journal of Advanced Transportation Table 3: Advantages and disadvantages of the used ML techniques for promoting environmental sustainability in maritime port logistics. ML techniques Advantages Disadvantages Related articles (i) Works on any size of the dataset (i) Te correct polynomial degree should be chosen for Polynomial regression (PR) (ii) Gives information about the relevance of [24, 25, 27, 30, 31, 33, 34, 40–42, 45, 47] a good bias features (i) Hardware dependent (i) Efciency (ii) Complex algorithms (ii) Continuous learning Neural networks (NN) (iii) Black-box nature [12, 35, 36, 42, 43, 46, 48, 49] (iii) Data retrieval (iv) Approximate results (iv) Multitasking (v) Data-dependency (i) Works very well on nonlinear problems Support vector machine (ii) Easily adaptable (i) Requires feature scaling [38, 46] (SVM) (iii) Ignores outliers (i) Performs very well on medium and small datasets (ii) Easy to interpret (i) Sensitive to outliers Gradient boost (GBoost) (iii) Prevents overftting (ii) Hardly scalable [42] (iv) A great approach for enhancing classifcation (iii) Poor results on unstructured data and regression solutions (i) Accurate (i) Te number of trees should be chosen Random forest (RF) (ii) Powerful [42] (ii) No interpretability (iii) Works very well on linear/nonlinear problems (i) Fast (i) Need feature scaling K-nearest neighbor (KNN) [42] (ii) Easy to implement (ii) Data should be cleaned (i) Easy to implement Hierarchical clustering (HC) (i) Long runtime [26, 30, 51] (ii) Does not require the number of clusters (i) Requires the number of clusters (i) Easy to implement (ii) Requires initial values (ii) Large datasets K-means (iii) Requires dimensionality reduction tools if the [37, 50] (iii) Convergence number of dimensions is high (iv) Diferent shapes and sizes with generalization (iv) Does not ignore outliers (i) Does not require the number of clusters (i) Poor results for datasets with various densities Spatial clustering (SC) (ii) Diferent shaped clusters (ii) Poor results for unstructured data [32] (iii) Ignores outliers (iii) Not deterministic (i) Eliminates correlated features (i) Less interpretable Principal component (ii) Enhances algorithm performance (ii) Data has to be uniformed [26, 27, 44, 52] analysis (PCA) (iii) Reduces overftting (iii) Loss of information (iv) Enhances visualization Kernel density estimator (i) Smooth visualization (i) Biased at the boundaries [28] (KDE) (ii) Works with various shapes and sizes (ii) Information loss by oversmoothing (i) Interpretable Self-organizing maps (SOM) (i) Requires initial weights [29] (ii) Applicable for large datasets Journal of Advanced Transportation 15 implementation of ML techniques to promote environ- out port operations that incorporate environmental aspects is also something we propose as an extension given that the mentally sustainable practices at ports, we propose as an extension the consideration of optimization techniques such focus of this SLR was on the implementation of ML tech- niques to promote environmentally sustainable practices as metaheuristics, mathematical programming, and heu- in ports. ristic approaches to aid decision making for planning and executing port operations incorporating environmental aspects. Finally, a complementary SLR on green maritime 4. Conclusions shipping can be a relevant research direction to provide Tis article presented the results of a systematic literature insights and analyse the current contributions of ML in that review of the recent literature on machine learning for application domain. promoting environmental sustainability in ports. Te review explored contributions of fve years (2017–2021) with the Data Availability aim of capturing the most recent approaches to foster en- No data were used to support this study. vironmental sustainable port operations. It categorized the identifed articles based on their application area within the port as well as the application of the ML approach. Using the Conflicts of Interest PRISMA protocol and bibliometric tools, the research Te authors declare that they have no conficts of interest. framework was constituted on the major considerations of impacts, techniques, and problems. In general, the chal- lenges and barriers of machine learning to aid decision Acknowledgments making and promote more sustainable green practices at Te third author of the manuscript acknowledges the ports were discussed. By analyzing the 29 identifed articles support of the National Agency of Research and Develop- with their methodological approaches, this review sum- ment (ANID), Chile, through the grant no. FONDECYT marized the academic contributions considering the three N.1210735, the program STIC-AMSUD-22-STIC-09, and main dimensions including machine learning, port logistics, the grant no. FOVI220133 “Digital Ports of Next and environmental sustainability. Te articles that used Generation”. regression models were dominant in the literature, while LSTM and RNN were the most recent approaches. Also, in terms of environmental indicators, investigations on References emissions and energy consumption were predominant [1] E. Lalla-Ruiz, L. Heilig, and S. Voß, “Environmental sus- among collected articles. tainability in ports,” Sustainable Transportation and Smart Concerning the type of approaches, supervised learning Logistics: Decision-Making Models and Solutions, Vol. 65–89, was the most used type of technique while unsupervised Elsevier, , Amsterdam, Netherlands, 2018. learning was the second option with about 30% of collected [2] A. Vega-Muñoz, G. Salazar-Sepulveda, J. F. Espinosa-Cristia, articles. Dimensionality reduction tools as data pre- and J. Sanhueza-Vergara, “How to measure environmental processing techniques reduced computation costs as they performance in ports,” Sustainability, vol. 13, no. 7, p. 4035, minimized the volume of data in databases. Clustering techniques were benefcial to constructing clusters where [3] H. Davarzani, B. Fahimnia, M. Bell, and J. Sarkis, “Greening ports and maritime logistics: a review,” Transportation Re- data objects within the same cluster were similar to one search Part D: Transport and Environment, vol. 48, pp. 473– anther but diferent from the objects in other clusters. 487, 2016. Moreover, the information on ML-based prediction algo- [4] M. Ghobaei-Arani, M. Shamsi, and A. A. Rahmanian, “An rithms such as polynomial regression and neural networks efcient approach for improving virtual machine placement showed a high contribution to promoting environmental in cloud computing environment,” Journal of Experimental indicators including greenhouse gas emissions, energy and Teoretical Artifcial Intelligence, vol. 29, no. 6, consumption, and water pollution. pp. 1149–1171, 2017. Te outcomes of this SLR can be used as a starting point [5] Y. Zheng, J. Zhao, and G. Shao, “Port city sustainability: for investigations regarding the application of machine a review of its research trends,” Sustainability, vol. 12, no. 20, learning techniques for enhancing environmental sustain- p. 8355, 2020. ability in ports and provide a guideline for researchers to [6] H. Geerlings and T. Vellinga, “Sustainability,” in Ports and Networks, pp. 296–314, Routledge, Oxfordshire, UK, 1 edi- fnd research gaps in the literature. In the light of this review, tion, 2018. future studies may focus on examining the performance of [7] F. Schulte, E. Lalla-Ruiz, R. G. Gonzalez-Ram ´ ´ırez, and S. Voß, diferent ML algorithms for mitigating the impact of envi- “Reducing port-related empty truck emissions: a mathemati- ronmental concerns related to solid wastes or noise pollution cal approach for truck appointments with collaboration,” near port cities, as there are several pollutants, but there are Transportation Research Part E: Logistics and Transportation only a few contributions for solving those problems yet. Review, vol. 105, pp. 195–212, 2017. Another relevant research direction is the one related to [8] HLEG, “Independent High-Level Expert Group on Artifcial supporting renewable energy technologies based on ML Intelligence Set up by the European Commission,” 2019, techniques that can be addressed by future researchers. https://www.aepd.es/sites/default/fles/2019-12/ai-defnition. Moreover, given that the scope of this SLR was on the pdf. 16 Journal of Advanced Transportation [9] A. Shakarami, A. Shahidinejad, and M. Ghobaei-Arani, “An [24] J. Sun, Y. Yuan, R. Yang, X. Ji, and J. Wu, “Performance autonomous computation ofoading strategy in Mobile Edge evaluation of Chinese port enterprises under signifcant en- vironmental concerns: an extended DEA-based analysis,” Computing: a deep learning-based hybrid approach,” Journal Transport Policy, vol. 60, pp. 75–86, 2017. of Network and Computer Applications, vol. 178, Article ID [25] L. Wang, C. Peng, W. Shi, and M. Zhu, “Carbon dioxide 102974, 2021. emissions from port container distribution: spatial charac- [10] F. Jazayeri, A. Shahidinejad, and M. Ghobaei-Arani, “Au- teristics and driving factors,” Transportation Research Part D: tonomous computation ofoading and auto-scaling the in the Transport and Environment, vol. 82, Article ID 102318, 2020. mobile fog computing: a deep reinforcement learning-based [26] C. Wang, H. Haralambides, and L. Zhang, “Sustainable port approach,” Journal of Ambient Intelligence and Humanized development: the role of Chinese seaports in the 21st century Computing, vol. 12, no. 8, pp. 8265–8284, 2021. Maritime Silk Road,” International Journal of Shipping and [11] S. Samoili, M. Lopez ´ Cobo, E. Gomez, ´ G. De Prato, Transport Logistics, vol. 13, no. 1/2, p. 205, 2021. F. Mart´ınez-Plumed, and B. Delipetrev, “AI Watch - defning [27] S. Y. Kuo and P. C. Lin, “Determinants of green performance Artifcial Intelligence. Towards an operational defnition and in container terminal operations: a lean management,” taxonomy of artifcial intelligence,” in Joint Research Cen- Journal of Cleaner Production, vol. 275, Article ID 123105, treEuropean Commission, Brussels, Belgium, 2020. [12] S. Holly, A. Nieße, M. Troschel ¨ et al., “Flexibility management [28] S. Cheon, A. Maltz, and K. Dooley, “Te link between eco- and provision of balancing services with battery-electric au- nomic and environmental performance of the top 10 U.S. tomated guided vehicles in the Hamburg container terminal ports,” Maritime Policy and Management, vol. 44, no. 2, Altenwerder,” Energy Inform, vol. 3, no. 1, p. 26, 2020. pp. 227–247, 2017. [13] N. H. Prevljak, “HHLA container terminals use machine [29] Y. S. Park, N. M. A. Mohamed Abdul Ghani, F. Gebremikael, learning technology to boost productivity,” Ofshore Energy, and G. Egilmez, “Benchmarking environmental efciency of Broussard, Louisiana, 2020. ports using data mining and RDEA: the case of a U.S. con- [14] I. de la Peña Zarzuelo, M. J. Freire Soeane, and B. Lopez ´ tainer ports,” International Journal of Logistics Research and Bermudez, ´ “Industry 4.0 in the port and maritime industry: Applications, vol. 22, no. 2, pp. 172–187, 2019. a literature review,” Journal of Industrial Information In- [30] C. Quintano, P. Mazzocchi, and A. Rocca, “Examining eco- tegration, vol. 20, Article ID 100173, 2020. efciency in the port sector via non-radial data envelopment [15] G. Yeo, S. H. Lim, L. Wynter, and H. Hassan, “MPA-IBM analysis and the response based procedure for detecting unit project SAFER: sense-making analytics for maritime event segments,” Journal of Cleaner Production, vol. 259, Article ID recognition,” INFORMS Journal on Applied Analytics, vol. 49, 120979, 2020. no. 4, pp. 269–280, 2019. [31] Y. T. Chang, H. K. Park, S. Lee, and E. Kim, “Have emission [16] E. F. France, N. Ring, J. Noyes et al., “Protocol- control areas (ECAs) harmed port efciency in Europe?” developingmeta-ethnography reporting guidelines Transportation Research Part D: Transport and Environment, (eMERGe),” BMC Medical Research Methodology, vol. 15, vol. 58, pp. 39–53, 2018. no. 1, pp. 103–114, 2015. [32] L. Wang, Z. Zhou, Y. Yang, and J. Wu, “Green efciency [17] A. Tong, K. Flemming, E. McInnes, S. Oliver, and J. Craig, evaluation and improvement of Chinese ports: a cross- “Enhancing transparency in reporting the synthesis of efciency model,” Transportation Research Part D: Trans- qualitative research: entreq,” BMC Medical Research Meth- port and Environment, vol. 88, Article ID 102590, 2020. odology, vol. 12, no. 1, pp. 181–188, 2012. [33] D. J. Eatough, P. Cropper, W. Keeton et al., “Apportionment [18] B. C. O’Brien, I. B. Harris, T. J. Beckman, D. A. Reed, and of PM2.5 adjacent to the I-710 harbor freeway in long Beach, D. A. Cook, “Standards for reporting qualitative research: CA,” Journal of the Air and Waste Management Association, a synthesis of recommendations,” Academic Medicine, vol. 89, vol. 70, no. 3, pp. 260–282, 2020. no. 9, pp. 1245–1251, 2014. [34] A. Ekmekçioglu, ˘ K. Unlugenço ¨ glu, ˘ and U. B. Çelebi, “Con- [19] K. Flemming, A. Booth, K. Hannes, M. Cargo, and J. Noyes, tainer ship emission estimation model for the concept of “Cochrane Qualitative and Implementation Methods Group green port in Turkey,” Journal of Engineering for the Maritime guidance series—article 6: reporting guidelines for qualitative, Environment, vol. 236, no. 2, pp. 504–518, 2021. implementation, and process evaluation evidence syntheses,” [35] H. Hong, H. Jeon, C. Youn, and H. Kim, “Incorporation of Journal of Clinical Epidemiology, vol. 97, pp. 79–85, 2018. shipping activity data in recurrent neural networks and long [20] A. Liberati, D. G. Altman, J. Tetzlaf et al., “Te PRISMA short-term memory models to improve air quality predictions statement for reporting systematic reviews and meta-analyses around busan port,” Atmosphere, vol. 12, no. 9, p. 1172, 2021. of studies that evaluate healthcare interventions: explanation [36] H. Agamy, M. Abdelgeliel, M. Mosleh, K. Elserafy, and and elaboration,” BMJ, vol. 339, 2009. N. A. N. Mohamed, “Neural fuzzy control of the indoor air [21] D. Moher, A. Liberati, J. Tetzlaf et al., “Preferred reporting quality onboard a RO–RO ship garage,” International Journal items for systematic reviews and meta-analyses: the PRISMA of Fuzzy Systems, vol. 22, no. 3, pp. 1020–1035, 2020. statement,” PLoS Medicine, vol. 6, no. 7, Article ID e1000097, [37] G. Song, B. Li, and Y. He, “Container terminal logistics generalized computing architecture and green initiative [22] V. Garousi, K. Petersen, and B. Ozkan, “Challenges and best computational pattern performance evaluation,” Information, practices in industry-academia collaborations in software vol. 10, no. 12, pp. 383–424, 2019. engineering: a systematic literature review,” Information and [38] P. Cammin, M. Sarhani, L. Heilig, and S. Voß, “Applications Software Technology, vol. 79, pp. 106–127, 2016. of real-time data to reduce air emissions in maritime ports,” in [23] S. Pahlevan-Sharif, P. Mura, and S. N. R. Wijesinghe, “A Design, User Experience, and Usability. Case Studies in Public systematic review of systematic reviews in tourism,” Journal of and Personal Interactive Systems. HCII 2020, A. Marcus and Hospitality and Tourism Management, vol. 39, pp. 158–165, E. Rosenzweig, Eds., Vol. 12202, Springer Cham, New York, 2019. NY, USA, 2020. Journal of Advanced Transportation 17 [39] J. Yu, G. Tang, X. Song et al., “Ship arrival prediction and its [55] L. Fredianelli, M. Bolognese, F. Fidecaro, and G. Licitra, value on daily container terminal operation,” Ocean Engi- “Classifcation of noise sources for port area noise mapping,” Environments, vol. 8, no. 2, p. 12, 2021. neering, vol. 157, pp. 73–86, 2018. [40] S. Jahangiri, N. Nikolova, and K. Tenekedjiev, “An improved [56] H. Du and Y. Jiang, “Backup or reliability improvement strategy for a manufacturer facing heterogeneous consumers emission inventory method for estimating engine exhaust in a dynamic supply chain,” IEEE Access, vol. 7, Article ID emissions from ships,” Sustainable Environment Research, 50419, 2019. vol. 28, no. 6, pp. 374–381, 2018. [57] A. Kara and I. Dogan, “Reinforcement learning approaches [41] B. Goldsworthy, H. Enshaei, and S. Jayasinghe, “Comparison for specifying ordering policies of perishable inventory sys- of large-scale ship exhaust emissions across multiple reso- tems,” Expert Systems with Applications, vol. 91, pp. 150–158, lutions: from annual to hourly data,” Atmospheric Environ- ment, vol. 214, Article ID 116829, 2019. [58] A. Kuhnle, J. Jakubik, and G. Lanza, “Reinforcement learning [42] Y. Peng, H. Liu, X. Li, J. Huang, and W. Wang, “Machine for opportunistic maintenance optimization,” Production learning method for energy consumption prediction of ships Engineering, vol. 13, no. 1, pp. 33–41, 2018. in port considering green ports,” Journal of Cleaner Pro- [59] M. F. Monaco, L. Moccia, and M. Sammarra, “Operations duction, vol. 264, Article ID 121564, 2020. Research for the management of a transhipment container [43] S. Wen, T. Zhao, Y. Tang et al., “Coordinated optimal energy terminal: the Gioia Tauro case,” Maritime Economics and management and voyage scheduling for all-electric ships Logistics, vol. 11, no. 1, pp. 7–35, 2009. based on predicted shore-side electricity price,” IEEE [60] Q. Tan, Y. Tong, S. Wu, and D. Li, “Modeling, planning, and Transactions on Industry Applications, vol. 57, no. 1, scheduling of shop-foor assembly process with dynamic pp. 139–148, 2021. cyber-physical interactions: a case study for CPS-based smart [44] Y. Zhan and T. Liu, “Weighted RPCA based background industrial robot production,” International Journal of Ad- subtraction for automatic berthing,” in Proceedings of the vanced Manufacturing Technology, vol. 105, no. 9, pp. 3979– Chinese Control Conference (CCC), pp. 3519–3524, Guangz- 3989, 2019. hou, China, July 2019. [61] G. Wong, T. Greenhalgh, G. Westhorp, J. Buckingham, and [45] M. Nastasi, L. Fredianelli, M. Bernardini, L. Teti, F. Fidecaro, R. Pawson, “RAMESES publication standards: meta-narrative and G. Licitra, “Parameters afecting noise emitted by ships reviews,” Journal of Advanced Nursing, vol. 69, no. 5, moving in port areas,” Sustainability, vol. 12, no. 20, p. 8742, pp. 987–1004, 2013. [62] Z. Zhu, N. Anwer, Q. Huang, and L. Mathieu, “Machine [46] F. Alasali, S. Haben, V. Becerra, and W. Holderbaum, “Day- learning in tolerancing for additive manufacturing,” CIRP ahead industrial load forecasting for electric RTG cranes,” Annals, vol. 67, no. 1, pp. 157–160, 2018. Journal of Modern Power Systems and Clean Energy, vol. 6, no. 2, pp. 223–234, 2018. [47] S. Fahdi, M. Elkhechaf, and H. Hachimi, “Machine learning for cleaner production in port of Casablanca,” Journal of Cleaner Production, vol. 294, Article ID 126269, 2021. [48] Z. Wang, Y. Shi, and S. Li, “Research on anti-swaying of crane based on t-s type adaptive neural fuzzy control,” in Pro- ceedings of the Chinese Control and Decision Conference (CCDC), pp. 5503–5508, Shenyang, China, June 2018. [49] A. Hill and J. W. Bose, ¨ “A decision support system for im- proved resource planning and truck routing at logistic nodes,” Information Technology and Management, vol. 18, no. 3, pp. 241–251, 2017. [50] L. Heilig, E. Lalla-Ruiz, and S. Voß, “Multi-objectiveinter- terminal truck routing,” Transportation Research Part E: Logistics and Transportation Review, vol. 106, pp. 178–202, [51] C. Caballini, M. D. Gracia, J. Mar-Ortiz, and S. Sacone, “A combined data mining – optimization approach to manage trucks operations in container terminals with the use of a TAS: application to an Italian and a Mexican port,” Transportation Research Part E: Logistics and Transportation Review, vol. 142, Article ID 102054, 2020. [52] H. Zhao, H. He, W. Lu, and Y. Hao, “Characterizing the variation of particles in varied sizes from a container truck in a port area,” Environmental Monitoring and Assessment, vol. 192, no. 12, p. 787, 2020. [53] K. Murphy, Machine Learning: A Probabilistic PerspectiveTe MIT Press, Cambridge, MA, USA, 2012. [54] M. Bertolini, D. Mezzogori, M. Neroni, and F. Zammori, “Machine Learning for industrial applications: a compre- hensive literature review,” Expert Systems with Applications, vol. 175, Article ID 114820, 2021.

Journal

Journal of Advanced Transportation – Hindawi Publishing Corporation

Published: Mar 3, 2023

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Machine Learning for Promoting Environmental Sustainability in Ports

Machine Learning for Promoting Environmental Sustainability in Ports

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Machine Learning for Promoting Environmental Sustainability in Ports

Machine Learning for Promoting Environmental Sustainability in Ports

References (53)

Abstract

Journal

Recommended Articles

There are no references for this article.

Our policy towards the use of cookies