Access the full text.
Sign up today, get DeepDyve free for 14 days.
Loading next page...
/lp/association-for-computing-machinery/artificial-intelligence-in-underwater-digital-twins-sensor-networks-HlBxrmU1l3
References
References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.
- Publisher
- Association for Computing Machinery
- Copyright
- Copyright © 2022 Copyright held by the owner/author(s). Publication rights licensed to ACM.
- ISSN
- 1550-4859
- eISSN
- 1550-4867
- DOI
- 10.1145/3519301
- Publisher site
- See Article on Publisher Site
Abstract
Artificial Intelligence in Underwater Digital Twins Sensor Networks ZHIHAN LV, Department of Game Design, Faculty of Arts, Uppsala University, Sweden DONGLIANG CHEN, College of Computer Science and Technology, Qingdao University, China HAILIN FENG, School of Information Engineering, Zhejiang A & F University, China WEI WEI, School of Computer Science and Engineering, Xi’an University of Technology, China HAIBIN LV, North China Sea Offshore Engineering Survey Institute, Ministry of Natural Resources North Sea Bureau, China The particularity of the marine underwater environment has brought many challenges to the development of underwater sensor networks (UWSNs). This research realized the effective monitoring of targets by UWSNs and achieved higher quality of service in various applications such as communication, monitoring, and data transmission in the marine environment. After analysis of the architecture, the marine integrated communication network system (MICN system) is constructed based on the maritime wireless Mesh network (MWMN) by combining with the UWSNs. A distributed hybrid fish swarm optimization al- gorithm (FSOA) based on mobility of underwater environment and artificial fish swarm (AFS) theory is proposed in response to the actual needs of UWSNs. The proposed FSOA algorithm makes full use of the perceptual communication of sensor nodes and lets the sensor nodes share the information covered by each other as much as possible, enhancing the global search ability. In addition, a reliable transmission protocol NC-HARQ is put forward based on the combination of network coding (NC) and hybrid automatic repeat request (HARQ). In this work, three sets of experiments are performed in an area of 200 × 200 × 200 m. The simulation results show that the FSOA algorithm can fully cover the events, effectively avoid the blind movement of nodes, and ensure consistent distribution density of nodes and events. The NC-HARQ proto- col proposed uses relay nodes for retransmission, and the probability of successful retransmission is much higher than that of the source node. At a distance of more than 2,000 m, the successful delivery rate of data packets is as high as 99.6%. Based on the MICN system, the intelligent ship constructed with the digital twins framework can provide effective ship operating state prediction information. In summary, this study is of great value for improving the overall performance of UWSNs and advancing the monitoring of marine data information. CCS Concepts: • Computer systems organization → Embedded systems; Additional Key Words and Phrases: Marine monitoring, underwater sensor networks, digital twins, artificial intelligence This work was supported by National Natural Science Foundation of China (No. 61902203). Authors’ addresses: D. Chen, College of Computer Science and Technology, Qingdao University, Qingdao 266071, China; email: cdlord@qq.com; W. Wei, School of Computer Science and Engineering, Xian University of Technology, Xian 710048, China; email: weiwei@xaut.edu.cn; H. Lv, North China Sea Offshore Engineering Survey Institute, Ministry of Natural Resources North Sea Bureau, Qingdao, China; email: lvhaibinsoa@gmail.com. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than the author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from permissions@acm.org. © 2022 Copyright held by the owner/author(s). Publication rights licensed to ACM. 1550-4859/2022/04-ART39 $15.00 https://doi.org/10.1145/3519301 ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. 39:2 Z. Lv et al. ACM Reference format: Zhihan Lv, Dongliang Chen, Hailin Feng, Wei Wei, and Haibin Lv. 2022. Artificial Intelligence in Underwater Digital Twins Sensor Networks. ACM Trans. Sen. Netw. 18, 3, Article 39 (April 2022), 27 pages. https://doi.org/10.1145/3519301 1 INTRODUCTION As scholars from all over the world pay more attention to marine technology to an unprecedented strategic height, marine resource survey, marine environmental monitoring, and sea area secu- rity can be realized through satellites, ships, and submarines [1]. But these methods are unable to complete these tasks over a long time, at close range, seamless, and in real-time. As an underwater monitoring network system composed of acoustic communication sensor nodes, UWSN has now become an important part of ocean informatization [2–4]. The use of sensor networks can obtain marine data for a long time, monitor the marine environment, and cooperate with traditional ship platforms to jointly complete more complex tasks [5]. In recent years, the research of UWSNs mainly involves underwater communication technology, layout and networking, routing proto- cols, and location tracking [6]. Communication and positioning are the two core key technologies that support UWSNs. Underwater acoustic communication provides a link channel for informa- tion exchange among different nodes. On the one hand, underwater acoustic positioning provides position calibration for the nodes; on the other hand, as the dynamic node of the underwater sen- sor network, the underwater unmanned submarine can carry a variety of equipment to perform underwater tasks, and its maneuverability greatly expands the application field of the UWSNs [ 7]. The digital twins is considered to be an effective means to realize the interactive integration of the manufacturing information world and the physical world. It makes full use of the physical models, sensor updates, and operating history to complete the mapping of physical entities in the virtual space, thereby reflecting the life cycle process of physical entities [ 8–10]. As thecoretech- nology for creating and running digital twins, simulation technology is the basis for digital twins to realize data interaction and data fusion. On this basis, the digital twins has to rely on and integrate other new technologies to be online with the sensor, so as to ensure its fidelity, instantaneity, and closed-loop characteristics. The Internet of Things (IoT) allow to access the information about the physical world, and digital twins can use, analyze, optimize, and test this information [11]. Data mining and IoT technology are the main driving factors of digital twins. With the gradual maturity of technology, digital twins has begun to be applied in fields other than manufacturing [ 12]. By continuously integrating the latest information, digital twins technology can bring the greatest value, and combine the real data with virtual models for analysis, thereby reducing production interruptions and costs. For example, engineers can access real-time data, simulation results, and solutions using the digital twins, efficiently performing hundreds of operational tasks from a long distance [13]. The sensor industry is showing explosive growth in the world, but it is also facing more new challenges with the vigorous development of intelligent manufacturing. For the underwater envi- ronment with high complexity, the underwater acoustic channel will have a serious impact on the communication among the sensor nodes, and the existing underwater routing strategy is greatly affected by the underwater environment [ 14]. Therefore, more in-depth research is needed in the field of reliable transmission and routing technology for underwater acoustic sensor networks. In addition, when the multiple access mechanism of UWSNs is explored, it is not only necessary to consider the competing access requirements of nodes within the network, but also to avoid harm to other network users, achieving parallel transmission of multiple communications and maximize network capacity and increasing the spatial reuse rate of spectrum. After the sensor data of the ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. Artificial Intelligence in Underwater Digital Twins Sensor Networks 39:3 underwater site collected by the current sensor are obtained, the sensor data is processed by the digital twins inference model. In this way, the status data of the underwater station is obtained, thereby improving the accuracy and efficiency of obtaining the operating status of the station and reducing labor costs. Moreover, after the status data of the underwater station is obtained, the dig- ital twins display model further visualizes the operating status of the station, which is convenient for users to intuitively obtain the operating status of the station and is beneficial to improve the user experience. At present, various countries in the world have been making breakthroughs in the field of UWSNs research. The more representative projects include Seaweb, Ocean-TUNE of the US Navy, and SUNRISE and CommsNet 12 of Europe. They have accumulated a lot of experience in the ex- periment, and have tried to add mobile nodes such as AUV to the network, achieving good results. But there is no research on fully self-organized underwater mobile cluster networking. The key to realizing unmanned mobile cluster networking is whether the nodes are sufficiently intelligent, which requires a large number of intelligent algorithm support. The research of underwater mo- bile clusters has to consider not only the intelligence of clusters, but also a series of complex issues that affect network performance. The marine monitoring network is explored, and the informa- tion transmission mechanism and network coverage of UWSNs are discussed in this study. Taking sensor nodes as agents, a multiple access mechanism based on the propagation characteristics of underwater acoustic signals is put forward with the guidance of the power allocation strategy of network node game. Furthermore, a reliable underwater data transmission protocol is proposed to enhance the reliability and robustness of underwater data transmission, and its performance is verified by simulation experiments. Finally, the digital twins of the ship are generated through data-driven and physical model-driven methods to realize the prediction of ship navigation. The relatively systematic research in this study compensates the lack of information transmission ef- ficiency in the field of underwater sensors, thereby promoting the construction and improvement of MICN system. 2 EXPERIMENTAL AND COMPUTATIONAL DETAILS 2.1 Design of MICN System 2.1.1 Architecture Design of the MICN System. The collection, analysis, processing, and effec- tive transmission of marine information is the core of marine technology. The marine network sys- tem is mainly divided into the marine network and the underwater network. The marine network assists marine users to expand more marine services by obtaining monitoring data, and the underwater network is mainly to monitor and transmit various marine data [15, 16]. For the transmission of marine data information, it is necessary to cross the water-air interface, and the efficient integration of the underwater acoustic network and the radio network is required in the data information transmission network architecture. The current single underwater or mar- itime network system mainly relies on the Internet and satellite communication technology to transmit the collected marine data information to the land. The marine users only can obtain the required information from transmission by a land data center. The MICN system leads to very high transmission costs, high energy consumption of the network, and obvious delay [17]. There- fore, a new marine network architecture has to be designed to collect the underwater monitoring data and the efficient acquisition of UWSNs data effectively by sea users. Based on MWMN, an MICN system integrating the offshore and underwater networks is proposed by combining with the UWSNs. The overall architecture is shown in Figure 1. The MICN system consists of three parts: the MWMN and intelligent data gateway (IDG) and the underwater wireless sensor network (UWWSN). ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. 39:4 Z. Lv et al. Fig. 1. Architecture of MICN System. In a traditional wireless local area network (WLAN), each client accesses the network through a wireless link connected to an access point (AP). Users have to first visit a fixed AP to realize communication with each other. Such network structure is called the single-hop network [18–20]. In an MWMN, any wireless device node can act as an AP and a router synchronously, each node has the ability of sending and receiving signals and directly communicates with one node or more peer nodes. The robustness of the Mesh network is more in contrast to the single-hop network, because it does not depend on the performance of a single node. Failure of any node in a single-hop network can cause the breakdown of the entire network. There is one or several data transmission paths for each node in the Mesh network [21–23]. Mesh adopts the standard 802.11 standard protocol, which can be widely compatible with wireless client terminals. The maritime Mesh network is mainly composed of communication nodes. The data collected by the sensor net- work can be transmitted and shared among different nodes through voice over Internet protocol (VoIP). UWSN is mainly a self-organizing network constructed by underwater sensor nodes with com- munication and computing capabilities. Using the underwater acoustic communication method based on high-speed multi-carrier transmission technology, the node realizes the collection of un- derwater resources and environmental information by carrying different sensors. Furthermore, the collected data is transmitted to the underwater sink node, and then uploaded to the marine smart data gateway [24]. The node concept of UWWSN has been expanded from traditional sensors to new concept node forms including autonomous underwater robots, ships, submarines, torpedoes, mines, and surface buoys. The sensor network system usually is composed of sensor nodes, sink nodes (receivers), and management nodes (servers). The data under the monitor of sensor nodes are transmitted in multiple hops along the sensor backbone nodes (cluster heads). The monitored data may be processed by multiple nodes, then routed to the sink node after multiple hops, and fi- nally reach the management node through the Internet or satellite [25–27]. Thesensorcan be ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. Artificial Intelligence in Underwater Digital Twins Sensor Networks 39:5 Fig. 2. The structure of UWSNs and the internal structure of nodes. configured and managed through the management node, the monitoring tasks can be issued, and the monitoring data can be collected. Underwater sensor nodes are mainly composed of a main controller or central processing unit (CPU), and the controller is connected to the sensor through an interface circuit. The controller receives the sensor data and stores it in the memory, and then sends the data to other network nodes through the underwater acoustic modem after processing. The structure of UWSNs and the internal structure of nodes are shown in Figure 2. Since UWSNs and the maritime Mesh network use different communication signals, the net- working and access methods of the network are different, and the network protocols are not compatible. The corresponding gateway is designed for different networking to realize the in- tercommunication between the underwater network and the maritime network. The main part of the gateway is composed of three parts: underwater acoustic modem, very high frequency (VHF) module, and Beidou+GPS positioning module. The underwater acoustic modem adopts or- thogonal frequency division multiplexing (OFDM) modem, which realizes communication, networking, and positioning synchronously. The VHF module adopts a wireless data transmission station, and the Beidou+GPS positioning module adopts a dual-mode positioning module. The gateway has to make a data upload decision before the sensor nodes upload the collected data. Sensor nodes realize the energy-saving and efficient transmission of collected data by executing the upload decision of the gateway. 2.1.2 Network Functional Architecture. Corresponding to the MICN system, the network system functional architecture also includes three modules. The marine Mesh network is responsible for sea surface information service and underwater data transmission; the intelligent gateway is to collect and upload the underwater data, completing protocol conversion and data transfer; and UWSNs is to collect the underwater data information [28]. The marine underwater environment (Figure 3) includes various types of artificial and natural underwater acoustic systems, which simultaneously share only a few hundred kilohertz of under- water acoustic channel bandwidth. The sound waves emitted and received by marine organisms will affect the acoustic sensor system. Therefore, the sensor network in the MICN system should ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. 39:6 Z. Lv et al. Fig. 3. The marine underwater environment. have the ability to perceive the spectrum of other underwater acoustic systems, and can realize multi-user spectrum sharing and multi-channel spectrum access and management. Spectrum sens- ing is to detect the underwater spectrum occupancy. In the UWSNs where underwater mammals are the main users, if the distribution density of sensing users is too large, it may not be possible to obtain sufficient and accurate sensing data. The positioning algorithm is a sub-link in the spectrum sensing module, which can position the underwater mammals found in the spectrum detection and pass the position information to the power control module [29–31]. The operation result of spec- trum sensing is undertaken as the basis, and the channel quality measurement index is finally set by describing the different characteristics of the idle frequency band. In UWSNs, interference among sensor nodes can be avoided by the power control only, which reduces the interference and impacts of underwater mammals on the main user and improving the network throughput. In sensor nodes, the sensor is responsible for data collection; the processor stores and process the data, realizing control of the entire node; and the wireless communication module is for transmitting data and control information. The battery is generally deemed as an energy supply module and is mainly to provide energy for the first three modules. Since the data collected by UWSNs usually does not change for a long time, it is necessary to implement a data compression strategy to reduce the energy consumption during transmission of a large amount of redundant data. When the underwater node uploads the collected data to the gateway, the gateway makes the decision with the best energy saving effect based on the data upload decision mechanism at this time. In the MICN designed in this study, the gateway will send the data to the MWMN for transmission after the data collected underwater is received. There will be a big difference for marine WMN due to the wireless channel characteristics of the UWWSN, and the communication protocol different from VHF-WMN is required. In general, there are differences between marine WMNs and underwater sensors in terms of transmission media and communication protocols. Therefore, the MICN system should be equipped with a protocol conversion intelligent gateway to realize the monitor of data information and management of UWSNs by sea users. 2.2 Intelligent Coverage and Transmission Optimization of 3D UWSNs 2.2.1 Topological Structure Model of UWSNs. At present, the more common UWSNs include two structures: a two-dimensional (2D) network for static monitoring and a three-dimensional ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. Artificial Intelligence in Underwater Digital Twins Sensor Networks 39:7 Fig. 4. UWSNs under dynamic monitoring. (3D) network for dynamic monitoring [32, 33]. The 2D network structure is formed based on the fuzzy perception model and is generally composed of nodes deployed on the sea level or anchored on the seabed, as shown in Figure 4. In the 2D network structure, there are lots of sensor nodes to detect the surrounding area of interest on the seafloor plane. Each sensor node is connected to the joint point of the underwater network through the acoustic link according to the correspond- ing clustering algorithm, and the gateway node further transmits the collected data to the data processing base station on the surface. The underwater gateway node can directly communicate with the surface base station in the shallower ocean area; while the underwater gateway node has to communicate with the surface base station through a transducer that can support long- distance communication in the deep ocean area. When the amount of underwater data is large and the coverage area is wide, the data between the gateways is fused based on the fuzzy data fusion mechanism, and then the fused data is transmitted to the main gateway node through relay multi-hop transmission [34]. Compared with the 2D network, the 3D UWSNs are more powerful with more complex network structure. The network architecture is shown in Figure 5. In the 3D network structure, sensor nodes are distributed in sea areas of different depths. They are fixed sensor nodes on the anchor on the seabed or on the platform on the water surface through cables, and the latter method is easier. It is precisely because sensor nodes are deployed at different sea depths that the 3D network monitors the underwater area more extensively and comprehensively. In terms of data transmission, routing algorithms are generally adopted to determine a reliable path from sensor nodes to surface base stations [35–37]. If the distance between the node and the surface base station is relatively long, data can be transmitted by using a multi-hop relay. 2.2.2 Network Smart Coverage based on Swarm Intelligence Algorithm. The coverage control of UWSNs can be described as by determining the coverage deployment strategy or topological structure of sensor nodes, so that the sensor nodes can cooperate with each other to realize the comprehensive management of the monitoring area. The influence of underwater environment characteristics, channel bandwidth, and node energy supply will hinder the effective coverage of the network. In addition, special underwater environmental factors make it difficult to substitute ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. 39:8 Z. Lv et al. Fig. 5. UWSNs under 3D monitoring. the battery of sensor nodes, and the network coverage loopholes caused by energy consumption problems will continue to increase. Reducing the energy consumption of sensor nodes can prolong the life cycle of the network. It can avoid the premature end of the life cycle of the network by keeping the balance of energy consumption of the nodes in the network due to the premature disappearance of the energy consumption of some nodes. According to the mobility of underwater sensor nodes, network coverage deployment algo- rithms can currently be classified into three categories: static deployment, self-adjusting deploy- ment, and mobile-assisted deployment. The static deployment assumes that all sensor nodes are static after the initial deployment and fixed on the sea surface buoy or anchored on the seabed [ 38]. The strategy of static deployment is more energy-efficient and easy to operate, but such manual deployment of sensor nodes is too difficult to operate underwater and is not suitable for scenarios where the network scale is expanded or the topology changes. If the deployment of sensor nodes is achieved by adjusting the depth of the automatic inflatable buoy, the network coverage can adopt the self-adjusting deployment to achieve better coverage of the monitoring area by the network. 3D random deployment does not require the coordination of ground base stations. The sensor nodes are randomly deployed at the bottom of the target area, and then the depth is randomly selected. The difference between the bottom random deployment and the 3D random deployment is that the location of each sensor node will be informed by the ground base station [39]. Sen- sor nodes can adjust the depth by applying repulsive force until the sensor monitoring overlap area among nodes disappears. If there are other portable underwater mobile sensor nodes in the monitoring area, the nodes in UWSNs can communicate with the mobile nodes and cooperate to complete underwater monitoring tasks. Compared with ordinary sensor nodes, underwater sen- sor nodes are more flexible, and they can move autonomously and collect data as needed, but its energy consumption is the most obvious. It can effectively avoid the energy loss caused by blind movement of sensor nodes while the range of sensing and monitoring is expanded to enable connected nodes to share information. The deployment of sensor nodes in the underwater monitoring area is described, and a distributed ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. Artificial Intelligence in Underwater Digital Twins Sensor Networks 39:9 hybrid FSOA that allows sensor nodes to move autonomously toward the monitored object is proposed in this study [40]. If there are n sensor nodes in the underwater monitoring area A, s is the i-th sensor node in the network. If the event e is described as the dynamic point of the monitoring object, the set of events in A can be expressed as E = {e |e ∈ A, i = 1, 2...m}. Any sensor node is able to sense, move, and i i s c s c communicate, which can be described as B = (r , r ,l , P ),ofwhich r and r represent the sens- j j j j j j ing radius and communication radius of sensor node s , respectively, l represents the maximum j j moving step length, and P refers to the current position of sensor node. Sensor nodes can sense surrounding events and obtain the number of covered events of surrounding nodes by communi- cating with neighboring nodes. It is assumed that the coverage model of each sensor node in A can be regarded as a sphere, the node coordinates are at the center of the sphere, the sensing radius s c is expressed as r , and the communication radius is represented by r . The communication radius j j is set to two times or more of the sensing radius to ensure network connectivity. The Euclidean distance d (e , s ) between the event e and sensor node s can be expressed as Equation (1)below: i j i j 2 2 2 d (e , s ) = (x − x ) + (y − y ) + (z − z ) (1) i j j j j i i i In the equation above, (x ,y , z ) are the coordinates of sensor node s and (x ,y , z ) are the j j j j i i i coordinates of the event e . If the probability of e being covered by s is represented by p (e , s ),it i i j i j can be expressed as a binary function using the Boolean sensor coverage model, as follows: 1 d (e , s ) ≤ r i j p (e , s ) = (2) i j 0 other Similar to the calculation of the coverage area of a 2D sensor network, the probability P (e , S ) of e being covered by s in a 3D network is expressed as Equation (3): i j P (e , S ) = p (e , s ) ∨ p (e , s ) ∨...∨ p (e , s ) = 1− (1− P (e , s )) (3) i i 1 i 2 i N i i i=1 For an event e , its relative effectiveness is set as the coverage degree and can be described as follows: p (e , s ) i j D (e ) = (4) A i 1+ I (d (e , s )) ≤ r e ∈E i j s ∈S In the equation above, I () represents the indicator function, and I (d (e , s )) ≤ r is the i j e ∈E number of adjacent events e . The effective coverage D (e ) and network coverage η (E ) of the event set can be calculated to assess the performance of the node deployment algorithm proposed in this study. D (e ) A i D (e ) = (5) D (e ) A j e ∈E H (E ) n η (E ) = α + β (6) logm n In the above Equations (4) and (5), H (E ) is the coverage entropy of the event set, and refers to the number of events covered by sensor nodes. H (E ) = D (e ) log (7) A i D (e ) e ∈E ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. 39:10 Z. Lv et al. The objective of sensor node is to place the sensor nodes, reaching the maximal value η (E ). When H (E ) reaches the maximal value logm, η (E ) can be 1. AFSA adopts a bottom-up optimization idea to design the perception and behavior mechanism of a single individual, and then place one entity or a group of entities in the environment, allow- ing them to solve problems in the interaction of the environment [41]. Artificial fish have four behaviors: foraging, clustering, rear-end collision, and random. Each fish will try before moving to decide which action can be taken. The steps of the AFS algorithm are summarized as follows. The settings are initialized, including the size of the population, the initial position of the artificial fish, the field of view (FOV) of the artificial fish, the step length, and the crowding factor. The fitness value of each individual in the initial fish school is calculated, and the optimal artificial fish state and its assignment are selected to the bulletin board. Each individual is evaluated to select the behavior to be performed, including foraging, clustering, rear-end collision, and random behavior. The behavior of artificial fish is implemented and a new fish school is generated. All individuals are evaluated to check if any individual is better than the bulletin board, which should be replaced with the individual if yes. When the optimal solution on the bulletin board reaches within the sat- isfactory error range or when the upper limit of the number of iterations is reached, the operations of the algorithm is completed, or otherwise step 3 can be repeated. For the deployment of network area coverage, sensor nodes can be deemed as the artificial fish in AFSA, and events can be considered as food. According to the principle of fish school system operation mode, a distributed hybrid FSOA is proposed for UWSNs. In the monitoring area A, the allowable congestion σ (s ) at sensor node s is defined as follows: i j σ (s ) = ψ × N (s ) (8) i e i In the Equation (8)above, ψ refers to the expected coverage of a single event, and N (s ) refers e i to the number of events covered, which can be expressed as follows: N (s ) = p (e , s ) (9) e i i j e ∈E The number of nodes within the communication range and sensing range of sensor node s are s s N (s ) and N (s ), respectively, which can be written as follows: i i ne co N (s ) = card (λ(s )) (10) i j ne N (s ) = card (γ (s )) (11) i j co In the above equations, λ(s ) and γ (s ) are the sets of nodes s within the communication radius j j j and sensing range, respectively. If the number of partners in the visible domain of s is set to N (s ) > 0, the optimal node s j i opt ne in the information pool C can be determined: sum s = arg max N (s ) (12) opt k ne s ∈C k sum If there are more events covered at s anditisnot toocrowded,thenitcan move to the opt direction of partner s : opt X − X opt i X = X + rand (l ) × (13) next i X − X opt i In the Equation (13)above, l refers to the value of the moving step, and X and X are the i opt position vectors of s and s , respectively. i opt ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. Artificial Intelligence in Underwater Digital Twins Sensor Networks 39:11 Table 1. Overall Description of FSOA Algorithm FSOA s c Input: B = (r , r ,l , P ), IterNum j j j j i+1 Output: P S = {s , s ... s }← the random deployment nodes in UWSNs 1 2 n for k = 1, 2...T do max N (s ) ← event covered by s e i i N (s ) ← number of nodes within the sensing range of s e i i if N (s ) > 0 then ne i then, the set C can be obtained; sum The nodes in the set C are sorted according to the number of events covered to find the collection ψ ; sum s = arg max{N (s )} opt k ne s ∈C k sum The rear-end collision is performed to move closer to s ; opt else for N = 1, 2...N do prey max The foraging is performed to move randomly; if N (s ) > N (s ) then break; e e i end for end if If the end condition is met, the iteration is stopped. end for Foraging refers to the search activity for sensor nodes within the communication radius, which is to ensure that the node status is continuously optimized. In the initial stage, a larger step size is used to improve the convergence speed of the algorithm, and then the step size is reduced in a gradual way to transform the global search into a local search. Finally, a fine search is performed near the optimal position to improve the optimization accuracy of the algorithm. In this study, the dynamic adjustment idea of l is expressed as follows: l = l × a + l (14) Iter Iter−1 min Iter a = exp −k × (15) IterNum In the two equations above, Iter represents the current iteration number, IterNum refers to the maximum iteration number, and l is the maximum movement step. The movement step size de- pends on the value a. The FSOA algorithm proposed keeps the maximum step size at the initial stage of operation, gradually changes from large to small, and then maintains the minimum. The description of the AFS-based FSOA is given in Table 1. 2.2.3 Multiple Access Mechanism for Sensor Networks. Since the communication system of UWSNs may cause harm to marine mammals, the environmental perception and dynamic power control are integrated into UWSNs [42–44]. Among them, the environmental perception module is to detect and locate underwater mammals, using the passive positioning algorithm (PLA);and the dynamic power control module is to schedule the data transmission, using the power control algorithm (PoCA). Next, the above two algorithms proposed will be explained in detail. In the environment perception module, the passive positioning algorithm on the basis of dis- tance measurement adopts the distance between the target node and anchor node to solve the ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. 39:12 Z. Lv et al. Fig. 6. Key elements of PLA algorithm. location information of the target. The distance measurement of the existing positioning algo- rithm depends on the signal received by the anchor node, and it has to feed back a signal with a time stamp when the target receives the signal. As the target node, mammals can’t provide the above parameters through interaction [45]. Therefore, a target search algorithm based on the re- ceived signal strength is proposed in this study to passively locate mammals. The algorithm can search for the possible sound power of animals, and screen the sound power, and finally obtain the most likely sound power in a limited space. In the signal detection stage, it is assumed that the anchor node has detected the signal and completed the classification of mammals, and PLA will enter the information interaction stage. Each anchor node sends an information packet containing its received signal information and its own position information. In the final target positioning stage, each anchor node independently solves the position information of mammals through the PLA algorithm. In the localization algo- rithm, it has to consider the sound power range of the target first, and then search for the sound source sound power according to a certain step. For each sound source under test, the calculation equation for the distance between the anchor node and the positioning target is given as follows: 10 log = A (d, f ) (16) A (d, f ) = 10k · logd+ · log α (f ) (17) In the above equations, the distance between the anchor node and the positioning target is denoted as d; f refers to the acoustic signal frequency; A refers to the channel gain; k refers to the propagation coefficient, which is set to 1.5; α is the absorption coefficient; and P and P are the s r sound power and the received power of the measured sound source, respectively. The key elements and main processes of the PLA algorithm are shown in Figures 6 and 7, respec- tively. After the distance between the target and the anchor node is obtained, any two distances should satisfy d + d > d according to the relationship theorem of the three sides of the trian- 1 2 AB gle. For this algorithm, it has to calculate the difference of the distance between the sound source target and different anchor nodes, and each difference can establish a hyperboloid equation. Any three hyperboloid equations are selected, and the intersection point is the location of the position- ing target. Finally, it has to verify whether the intersection point corresponds to the sound power, so as to obtain the accurate sound source position information. ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. Artificial Intelligence in Underwater Digital Twins Sensor Networks 39:13 Fig. 7. Main process of PLA algorithm. For the location P (x ,y , z ) of the target sound source, if there is only one P corresponding to s s s s it, it is taken as the target sound source location. If there are more than one P , the distance error between the anchor node and the target is calculated with the equations below: 2 2 2 (x − x ) + (y − y ) + (z − z ) − d = ε (18) s a s a s a 1 a 2 2 2 (x − x ) + (y − y ) + (z − z ) − d = ε (19) s s s 2 b b b b 2 2 2 (x − x ) + (y − y ) + (z − z ) − d = ε (20) s c s c s c 3 c 2 2 2 ( ) ( ) ( ) x − x + y − y + z − z − d = ε (21) s d s d s d 4 d The PLA algorithm adopts the coordinate of the minimal ε as the position coordinate of the sound source target, which means that the position coordinate has the smallest positioning error. Sensor nodes can obtain the reasonable data packets using the PoCA algorithm proposed based on the gain information of the channel and the position information of underwater mammals. ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. 39:14 Z. Lv et al. Animals do not always appear in the coverage of UWSNs, so the power control mechanism can work in two different modes: shared mode and exclusive mode. In the sharing mode, each node hopes to get the fastest response to its own request in order to maximize utility. Therefore, the transmission power allocation can be regarded as a competitive optimization. The utility function is defined as Equation ( 22), so as to achieve the goal of PoCA. The function includes two aspects: maximizing throughput and avoiding excessive energy consumption. If each node shares the same bandwidth (B), it can be ignored when the utility function is constructed. h · p i i u (p , P ) = log 1+ − α (22) i i −i i N−1 h · p + σ i i j=1 In the Equation (22)above, h and p are the channel gain and transmission power on link I, i i respectively; P refers to the transmission power of other links except link i; σ represents the −i noise power; and α refers to the cost required for transmission power. The power allocation can be expressed as follows: h · p i i maxu (p , P ) = log 1+ − α · p (23) i i −i i i N−1 h · p + σ i i j=1 s.t. p ∈ [0, P ] (1 ≤ i ≤ N ) i max h · p i i ≥ SINR (1 ≤ i ≤ N ) (24) th N−1 h · p + σ i i j=1 h · p ≤ T mi i i=1 In the above equation, P refers to the maximum transmit power; SINR is the acoustic max th modem decoding threshold; T is the mammalian behavior influence threshold; and SINR is the th channel gain between the sending node and the animal. If the game theory is adopted to understand the transmission power allocation, the optimal transmission power of each sending node can be taken as a response function of the players in the game, which can be expressed as follows: ∂u h i i = − − α = 0 (25) N−1 ∂p h · p + h · p + σ i i i i j=1 N−1 2 h · p + σ i i j=1 p = − (26) α h i i α = 1/P is assumed to ensure p ≤ P . All sending nodes can obtain their own optimal trans- i max i max mission powers through the PCA algorithm. After p is obtained, the following matrix equation can be obtained: ∗ −1 P = H · G (27) In the above equation, H refers to a square matrix of N × N; P refers to the solution of Nash equilibrium, and G is a N × 1 vector. The node can transmit data packets in exclusive mode when no animals are detected within the UWSNs. At this time, the main task of the PoCA algorithm is to lower the energy consumption ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. Artificial Intelligence in Underwater Digital Twins Sensor Networks 39:15 while maximizing the throughput of the network. In the exclusive mode, the power allocation op- timization can be expressed as Equation (28), and the two constraints can be expressed as Equation (29): h · p i i maxu (p , P ) = log 1+ − α · p (28) i i −i i i N−1 h · p + σ i i j=1 s.t. p ∈ [0, P ] (1 ≤ i ≤ N ) i max (29) h · p i i ≥ SINR (1 ≤ i ≤ N ) th N−1 h · p + σ i i j=1 Compared with the shared mode, the exclusive mode removes the third constraint used to limit the impacts on animals, and the node can send data packets at the maximum power. The Bellhop model is adopted to describe the underwater acoustic propagation characteristics, so as to verify the performance of the PLA and PoCA algorithms proposed in the underwater environment. The experimental data come from the Ocean-TUNE underwater network test bed (located in the Long Island Strait) deployed in a real environment. The reflection coefficients of the bottom and water surface are set to 1 and−1, respectively, and four anchor nodes are deployed at a water depth of 50 m. 2.2.4 Reliable Transmission Protocol for Underwater Acoustic Sensors Based on Coding. Under- water information transmission will cause high bit error rate due to special environmental factors, so it is necessary to design a reliable transmission protocol based on error control strategies. The traditional error control strategies are mainly classified into four types: error detection (ED), au- tomatic retransmission request (ARQ), forward error correction coding (FEC),and HARQ [46–48]. In HARQ, the number of retransmission and the bit error rate can be lowered using the FEC, and the packet data transmission is guaranteed by the ARQ retransmission. Therefore, the HARQ mechanism can be determined as a compromise solution, which can correct the errors au- tomatically due to the capability of error correction. The sender is required to retransmit the data under the condition of out of the range of the error correction. It not only improves the reliability of the system, but also elevates the transmission efficiency of the system. A reliable transmission protocol NC-HARQ is proposed based on the combination of NC and HARQ, which makes full use of the cooperative nodes and the hierarchical gains of multi-node transmission in the network, and finally obtains the optimal transmission system. The data send- ing process and receiving process of NC-HARQ protocol are shown in Figure 8. During the retrans- mission, if the sending node receives the successfully decoded information, but the relay node does not receive the next hop node confirmation, the sending node will terminate the retransmission, and the relay node with better channel conditions will perform the data retransmission. Based on the automatic feedback of ARQ, the NC-HARQ protocol can reduce transmission failures caused by random burst packet loss, thereby improving the reliability of the system. In this study, the routing protocols are divided into energy-balanced and non-energy-balanced routing protocols. According to the different selection methods of the route candidate node set, the non-energy-balanced routing is divided into sending-based and receiving-based hybrid rout- ing, and energy-balanced routing is divided into variable power and fixed power routing accord- ing to whether the node has variable transmission power. For underwater routing based on the receiving end, the selection of the next hop sending node set is determined by the receiving node that receives the data packet. Under this system, the sender no longer stores the information of neighbor nodes. At the beginning of data transmission, the sending node packs part of the control ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. 39:16 Z. Lv et al. Fig. 8. The data sending process and receiving process of NC-HARQ protocol. information into the packet header. This control information is generally the status information of the current sending node. In a simple three-node scenario, the node R is located between the node S and the node D, and the node D can directly send data packets to the node D; p , p ,and p are the packet loss SD SR RD rates of the channels S → D, S → R,and R → D, respectively. For ARQ, it has to wait for feedback confirmation after a data packet is sent, and it will be resent without feedback confirmation. In this study, the maximum number of retransmissions is set to 1. The redundancy of ARQ can be expressed as Equation (30)below: η = (1− p ) + 2p = 1+ p (30) ARQ SD SD SD The successful delivery rate of ARQ can be calculated with the equation below: PDR = (1− p ) + p (1− p ) = 1− p (31) ARQ SD SD SD SD The equation below can be obtained after the energy consumption is normalized: ARQ 1 EN = = (32) ARQ PDR 1− p ARQ SD ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. Artificial Intelligence in Underwater Digital Twins Sensor Networks 39:17 The average end-to-end delay can be written as Equation (33)below: L 2d SD DE = η × + /PDR (33) ARQ ARQ ARQ R ν In NC-HARQ, the node S sends data packets to the node D and the relay node R synchronously, and the data retransmission is completed by the node R, so it only needs to ensure that the node S correctly receives the data of the node R. The redundancy of NC-HARQ can be expressed as follows: 6 K 8 K K η = N p + N (i ) p (i ) + (p (i, j)) (34) NC−HARQ 1 1 m m 7 m=2 i=1 a=7 c=1 b=1 The success rate of NC-HARQ could be written as follows: K K PDR = p + p (i ) + p (i ) + p (i ) + (p (i, j)) (35) NC−HARQ 1 3 4 6 7 i=1 c=1 The normalized energy consumption of NC-HARQ could be calculated with Equation (36)below. NC−HARQ EN = (36) NC−HARQ PDR NC−HARQ The average end-to-end delay can be defined as the equation below. 6 K 8 K ( ) ( ) ( ) ( ) T p + T i p i + T b p b 1 1 m m a a m=2 i=1 a=7 b=1 DE = (37) NC−HARQ K · PDR NC−HARQ The classic Throp model is undertaken as the underwater acoustic wave absorption model, and the noise spectrum measured by Wenz is taken as the noise model to prove the effectiveness of the NC-HARQ protocol designed in this study. The acoustic velocity (v) = 1500 m/s, the length of data packet (L)= 225 Bytes, and the channel bandwidth (B)= 15 kHz. For the transmission system with deletion codes, two cases (K = 4 and K = 8) are simulated to compare the performances of the three protocols, which are FEC, HARQ, and NC-HARQ. 2.3 Marine Monitoring Based on Digital Twins Ship-Based Sensors 2.3.1 Framework of Digital Twins Network. The goal is to build a marine information service system from the perspective of multidimensional data visualization by taking the marine space resources, marine environmental data resources, and marine integrated business management data resources as the core under the support of digital twins technology. The marine business results data are presented in the form of “a picture” to promote the centralization of marine business management and intelligent application services, so that the marine business management and natural resource auxiliary management can be supported better [49, 50]. The marine environment is complex and changeable, so the monitoring and prediction of the ship’s condition requires multi- disciplinary knowledge integration. The physical model of the ship in the digital twin system can generate the state prediction information. The goal of safe navigation can be achieved by providing early warning of potential dangerous situations of ships. Based on virtual reality (VR) and computer graphics, a virtual marine environment is constructed and the ship operation process is simulated in real time. As a dynamic virtual representation of a real ship, digital twins uses comprehensive knowledge from sensor data and physical models to simulate the ship’s operating status. A ship digital twins ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. 39:18 Z. Lv et al. framework based on a virtual reality system is proposed. The real sea area database and hydrody- namic database are first developed to ensure the consistency of the marine environment and ship motion characteristics with physical entities. Then, a virtual ship based on the sensor data and physical model is built, and the real sea environment and the virtual ship are finally rendered in the simulator. A framework of digital twins based on ship simulator is proposed in this study. First, a real sea area database and a hydrodynamic database are established for the reproduction of the real ocean environment and the prediction of ship motion. The sensor data contains the real-time motion state and control equipment information of the real ship, which are directly transmitted to the navigation virtual reality system for driving the virtual ship. The fluid data information obtained from the sensing equipment is introduced into the physical model for state prediction, and the data collected by the actual ship can support the comparison and verification. The real sea area database is to render the marine environment information to realize the digital twins of ships. 2.3.2 Ship-Based Sensors and Virtual Ship. The ship simulator realized by VR technology can simulate the marine environment and the operation process of the ship. The modeling of ships and the marine environment includes not only 3D digital models such as port buildings, various types of ships, and navigation aids, but also various physical models such as ship motion, weather conditions, wind, and waves. Based on the six-degree-of-freedom (6-DOF) motion equation of a rigid body, the dynamic model of the ship can be established as the equations below: η = [x,y, z, φ, θ, ϕ] (38) v = [u,v,w,p,q, r] (39) η refers to the 3D position and three-axis heading angle in the geographic coordinate system; and v refers to the three-axis speed and the three-axis angular velocity in the ship coordinate system. Based on the theory of time domain unification and modularization, the ship kinematics model constructed is described as follows: M = f + f + f + f + f + f + f v H P R wind current exc drif t (40) f = −A (∞) v − B (∞) v − K (t − τ ) [v (τ ) − Ue ]dτ H 1 ⎪ 0 − Cη + f NL−Hull (v ) In the Equation (40)above, wind, current, exc,and drift refer to the environmental interference force caused by wind, ocean current, first-order waves, and second-order waves, respectively; f , f ,and f refer to the forces generated by hull, propeller, and rudder, respectively; A, B,and K are P R the additional mass matrix, damping coefficient matrix, and impulse response function of the hull, respectively; C is the hull radiation force matrix; and f represents the nonlinear viscous NL−Hull (v ) force of the hull. The marine environment is numerically solved based on the WAVEWATCH-III model to ensure that the real marine environment is consistent with the marine environment in the digital twins. The environment vector field is given as follows: ∧ ∧ ∂N ∂ ∂ S +∇ (c + U )N + N + N = (41) k θ x д ∂t ∂k ∂θ σ ∂σ ∂d ∂U k = − − k (42) ∂d ∂s ∂s 1 ∂σ ∂d ∂U = − − k (43) k ∂d ∂m ∂m ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. Artificial Intelligence in Underwater Digital Twins Sensor Networks 39:19 Fig. 9. Node deployment based on the FSOA. In the equation, ∇ is the Hamilton operator, N refers to the wave action density spectrum, c represents the wave group velocity; U refers to the flow velocity; k and θ are the number and direction of the wave, respectively; s and m are the coordinates of θ in two directions perpendicular to each other. 3 RESULTS AND DISCUSSION 3.1 Performance of UWSNs Node Deployment Algorithm In this study, multiple sets of Monte Carlo simulation experiments are carried out on the deploy- ment of marine (3D) nodes on the Matlab platform. The experimental processing platform is In- tel(R) Core(TM) i7-4785T CPU 2.6GHz. Three sets of experiments are performed in an area of 200× 200× 200 m to check the effectiveness of the FSOA algorithm proposed in a static environ- ment. The settings of the three groups of experiments are as follows: 1 5nodes and35eventsare unevenly T-type distributed; 2 5 nodes and 35 events are randomly distributed; and 3 5nodes and 35 events are linearly unevenly distributed. The running results of the FSOA algorithm in the three sets of experiments are shown in Figures 9(a) ∼ 9(c), respectively. In the Figure 9, the blue sphere represents the 3D coverage of sensor nodes, and the distributed stars represent events. It can be seen that the FSOA algorithm can comprehensively cover the events, while ensuring that the distribution density of nodes and events is basically the consistent. The traditional particle swarm optimization (PSO) - based underwater sensor self- deployment algorithm and the FSOA algorithm proposed are adopted for the deployment of sensor ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. 39:20 Z. Lv et al. Fig. 10. The comparison of the coverage of node events and the total moving distance of the two algorithms. nodes. In the three sets of experiments, the comparison of the coverage of node events and the to- tal moving distance of the two algorithms in the running process is shown in Figure 10(a) and Figure 10(b), respectively. In the first set of experiments, the coverage of the FSOA algorithm can reach 0.78 after five iterations when 35 events are distributed in a T-shape. The dynamic adjust- ment capability of the PSO algorithm is relatively good, but the FSOA algorithm realizes more stability throughout the deployment. In the second and third sets of experiments, the FSOA algo- rithm is better than the PSO algorithm in terms of event coverage and total movement distance. Therefore, the FSOA algorithm not only exhibits a wider coverage capability, but also overcomes the blindness of the traditional algorithm for random search, with fast convergence speed and better stability. In summary, both algorithms require to move significantly in the initial deployment phase, so the initial total node movement distance is relatively large. As the deployment further deepens, the range of movement gradually decreases. The FSOA algorithm realizes information sharing among different nodes on the basis of the information pool, so that the blind movement of nodes can be avoided effectively. Compared with the PSO algorithm, the FSOA can obviously reduce the total movement distance of nodes during the deployment. 3.2 Evaluation on Multiple Access Performance of UWSNs Two anchor node deployment scenarios are set, and 1,000 points in the monitoring range are randomly selected as the possible locations for animals to comprehensively assess the performance of the PLA positioning algorithm. A UWSN simulation software Aqua-Sim based on NS2 is adopted to complete the simulation, and some of the physical layer modules are modified to meet the needs of transmission power adjustment. The coordinates of the four anchor nodes in scene one are: A (0,0,0), B (2000,0,0), C (0,2000,0), and D (2000,0,0), respectively. Standard triangular pyramid structure is set in the scene two. With reference to the speed range of mammals, four unequal speeds are set as v1 = 5m/s,v2 = 10 m/s, v3 = 15 m/s, and v4 = 20 m/s, respectively. The actual position of the animal is compared with the position result obtained by the positioning algorithm to assess the performance of the PLA algorithm. The magnitude and direction of the movement speed of underwater mammals will affect the positioning success rate, but the speed does not have a functional relationship. In the PLA determination algorithm proposed, the absolute frequency deviation of the received signal increase to varying degrees as the movement speed of mammals ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. Artificial Intelligence in Underwater Digital Twins Sensor Networks 39:21 increases. The calculated average positioning success rate is 89.4% in scenario one and 87.3% in scenario two. Therefore, the deployment of anchor nodes will have an impact on the positioning success rate. The average positioning error of mammals at different speeds in the two scenes is illustrated in Figure 11 below. Furthermore, the performance of the PoCA algorithm is evaluated through simulation experi- ments, including the throughput of the network under different v values and the influence of the detected mammalian motion on the network performance. The curves of the effective throughput of the network at different v values in the two scenarios are shown in Figure 12. When there are several mammals, the sensor node will switch to shared mode; and in the exclusive mode where no mammals are detected, node requests that allow parallel transmission will be postponed. In the case where mammals are most severely affected (v = v1), the effective throughput of the network can still reach 85% of the throughput without mammals. At different speeds (v values), the effec- tive throughput of network of the PoCA algorithm can be increased by 30%∼ 50% compared with the classic underwater medium access control (MAC) protocol and slotted floor acquisition multiple access (SFAMA) protocol. Therefore, the PoCA algorithm can effectively improve the capacity of multiplexing space, allowing more parallel transmission. End-to-end delay is an important indicator for evaluating the network performance. It is eval- uated in different speed scenarios through simulation experiments in this study, and the results are shown in Figure 13. It can be found that the presence of underwater mammals enhances the end-to-end time delay. When the moving direction is the same, the moving speed of the mammal is lower, which causes a longer time delay. The reason is that the slow movement speed of the mam- mal means that the impact on the network lasts for a long time, during which the transmission of many data packets is delayed, resulting in an increase in end-to-end delay. 3.3 Performance Analysis of NC-HARQ Reliable Transmission Protocol The comparison results of the transmission redundancy of the three transmission protocols at different distances are shown in Figure 14(a). Due to the impacts from integer quantization, the redundancy is higher when the distance between nodes is shorter and K = 4; the change of the value of K has less influence on the redundancy when the distance between nodes increases. As the distance increases, the redundancy of the FEC protocol increases more. Figure 14(b) illustrates the comparison result of the success rate of the data packet transmission of the three protocols at different distances. The FEC protocol encodes K original data packets and sends them, and all K data packets fail to be sent when the number of encodings can’t resist the high packet loss rate in the channel, so the overall transmission success rate is low. HARQ adds a feedback retransmission link, which improves the transmission success rate in contrast to the FEC. But only one data packet is retransmitted, so all errors can’t be corrected. The NC-HARQ protocol proposed applies a relay node for retransmission, and the probability of successful retransmission is obviously higher in contrast to that of the source node. When the distance is larger than 2,000 m, the success rate of data packets is as high as 99.6%. The relationship between the normalized energy consumption of data packet transmission and the communication distance is analyzed under the three protocols, as shown in Figure 15(a). HARQ improves the successful delivery rate of the system compared to FEC, the normalized energy con- sumption is slightly reduced. In contrast to the other two protocols, the NC-HARQ protocol pro- posed increases the redundancy, but improves the successful delivery rate of the system greatly, so the energy consumption required for the successful delivery of a data packet is relatively lower. Therefore, the efficiency of the system has been effectively improved. The end-to-end time delay under the three protocols is shown in Figure 15(b). It can be found that the FEC protocol can re- duce the delay in the feedback propagation process by increasing the transmission redundancy. ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. 39:22 Z. Lv et al. Fig. 11. The average positioning error of mammals at different speeds in the two scenes. Fig. 12. Comparison of effective throughput of the network at different v values. The average end-to-end delay under the NC-HARQ protocol is the lowest among the three proto- cols, which may be because that the system shows a higher transmission success rate and a shorter data packet transmission path, which improves the overall delay of the system. Performance of the data transmission system is tested in the NS-3 underwater simulation sys- tem to further analyze the reliability of the NC-HARQ protocol. The transmission success rate of data packet and end-to-end delay under the three protocols are shown in Figures 16(a) and 16(b), respectively. In the underwater simulation environment, the NC-HARQ protocol can adjust the redundancy based on the current state of the channel according to the feedback, and can realize the data retransmission using the relay node. It effectively guarantees the reliability of the system in contrast to the other two protocols. 3.4 Ship Navigation Analysis based on Digital Twins The real purpose of the sea area database is to realize the reproduction of the marine environment in the sea area and time period, and the actual measurement data of the sensor is taken as ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. Artificial Intelligence in Underwater Digital Twins Sensor Networks 39:23 Fig. 13. Comparison of end-to-end delay at different v values. Fig. 14. The transmission redundancy and success rate of the three transmission protocols at different distances. input to keep the control signals of the virtual ship and the real ship consistent. Different marine environments are set in this study to predict, so as to verify the feasibility of applying digital twins technology in real seas. The comparison result between the real sensor data and the predicted nav- igation trajectory is shown in Figure 17. The control input of all data is kept consistent, the line T in the figure refers to the real trajectory data recorded by global position system (GPS); and the line I is the trajectory prediction without considering the environmental factors. Lines II∼ IV are the trajectory prediction results when the wind, ocean current, and waves are considered only; and line V is the trajectory prediction result when the three environmental factors are comprehensively considered. The Figure 17 indicates that the trajectory V considers all environmental impacts, so it can simulate an environment closer to the real marine environment, thereby more accurately predicting the trajectory of the ship. From the above information, it can be concluded that under the influence of a single factor, the error of the predicted trajectory quickly accumulates due to the inaccurate description of the environment, resulting in a complete deviation from the true value. In the case of considering all environmental factors, the accuracy of predicting the trajectory is the ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. 39:24 Z. Lv et al. Fig. 15. Normalized energy consumption and end-to-end delay of three transmission protocols at different distances. Fig. 16. Comparison of performances of three transmission protocols under NS-3 underwater simulation system. highest. This result illustrates the importance of the marine environmental field for accurate prediction. Since the prediction accuracy is the highest in the fully coupled environment, the prediction of the heading, speed, and sway is based on the prediction value in the fully coupled environment. 4 CONCLUSIONS In UWSNs, coverage preservation technology only exists in the static deployment framework. The full coverage of target events is realized by sensor nodes based on the theoretical basis of AFS in consideration of the mobility of the underwater environment. Artificial FSOA not only shows a wider coverage capability, but also overcomes the blindness of traditional random search algorithms, with fast convergence speed and better stability. The environment perception and dynamic power control are integrated into UWSNs, the underwater mammals are detected and located using the PLA algorithm, and the data transmission is scheduled with the PoCA ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. Artificial Intelligence in Underwater Digital Twins Sensor Networks 39:25 Fig. 17. The comparison result between the real sensor data and the predicted navigation trajectory. algorithm. The PoCA algorithm can effectively improve the capacity of multiplexing space, al- lowing more parallel transmission. A decision-making mechanism for underwater network data upload that optimizes energy consumption is proposed, aiming at the energy limitation of UWSNs. The NC-HARQ protocol uses the advantages of ARQ automatic feedback to reduce transmission failures caused by random burst packet loss, thereby improving reliability of the system. The av- erage end-to-end delay under the NC-HARQ protocol is generally low, which may be because that the system shows a higher transmission success rate and a shorter data packet transmission path, which improve the overall delay of the system greatly. The marine communication network is analyzed from the overall level in this study, and it is found that optimizing the network coverage and data transmission of data sensors will help improve the efficiency of the network. However, the technology discussed in this study has not been applied in a large area, and the defense module of intruder detection has not been integrated into UWSNs, which will be discussed in detail in the following research. REFERENCES [1] T. Wilberforce, Z. El Hassan, A. Durrant, et al. 2019. Overview of ocean power technology. Energy 175 (2019), 165–181. [2] G. Han, X. Long, C. Zhu, et al. 2019. A high-availability data collection scheme based on multi-AUVs for underwater sensor networks. IEEE Transactions on Mobile Computing 19, 5 (2019), 1010–1022. [3] S. Li, W. Qu, C. Liu, et al. 2019. Survey on high reliability wireless communication for underwater sensor networks. Journal of Network and Computer Applications 148 (2019), 102446. [4] S. Karim, F. K. Shaikh, B. S. Chowdhry, et al. 2021. GCORP: Geographic and cooperative opportunistic routing protocol for underwater sensor networks. IEEE Access 9 (2021), 27650–27667. [5] B. C. Renner, J. Heitmann, and F. Steinmetz. 2020. ahoi: Inexpensive, low-power communication and localization for underwater sensor networks and μAUVs. ACM Transactions on Sensor Networks (TOSN) 16, 2 (2020), 1–46. [6] H. Huang and Y. R. Zheng. 2018. Node localization with AoA assistance in multi-hop underwater sensor networks. Ad Hoc Networks 78, (2018), 32–41. [7] J. Yan, Z. Xu, X. Luo, et al. 2018. Feedback-based target localization in underwater sensor networks: A multisensor fusion approach. IEEE Transactions on Signal and Information Processing over Networks 5, 1 (2018), 168–180. [8] F. Tao, M. Zhang, Y. Liu, et al. 2018. Digital twin driven prognostics and health management for complex equipment. CIRP Annals 67, 1 (2018), 169–172. [9] R. B. Roy, D. Mishra, S. K. Pal, et al. 2020. Digital twin: Current scenario and a case study on a manufacturing process. The International Journal of Advanced Manufacturing Technology 107, 9 (2020), 3691–3714. [10] K. Ding, F. T. S. Chan, X. Zhang, et al. 2019. Defining a digital twin-based cyber-physical production system for autonomous manufacturing in smart shop o fl ors. International Journal of Production Research 57, 20 (2019), 6315–6334. ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. 39:26 Z. Lv et al. [11] Y. He, J. Guo, and X. Zheng. 2018. From surveillance to digital twin: Challenges and recent advances of signal pro- cessing for Industrial Internet of Things. IEEE Signal Processing Magazine 35, 5 (2018), 120–129. [12] M. Jacoby and T. Usländer. 2020. Digital twin and Internet of Things—current standards landscape. Applied Sciences 10, 18 (2020), 6519. [13] C. Koulamas and A. Kalogeras. 2018. Cyber-physical systems and digital twins in the Industrial Internet of Things [cyber-physical systems]. Computer 51, 11 (2018), 95–98. [14] N. Goyal, M. Dave, and A. K. Verma. 2019. Protocol stack of underwater wireless sensor network: Classical approaches and new trends. Wireless Personal Communications 104, 3 (2019), 995–1022. [15] A. Zolich, D. Palma, K. Kansanen, et al. 2019. Survey on communication and networks for autonomous marine sys- tems. Journal of Intelligent & Robotic Systems 95, 3 (2019), 789–813. [16] T. C. Smythe and J. McCann. 2018. Lessons learned in marine governance: Case studies of marine spatial planning practice in the US. Marine Policy 94, (2018) 227–237. [17] C. A. Trasviña-Moreno, R. Blasco, Á. Marco, et al. 2017. Unmanned aerial vehicle based wireless sensor network for marine-coastal environment monitoring. Sensors 17, 3 (2017), 460. [18] R. K. Garg, M. V. D. Nair, S. Singhal, et al. 2021. Compact CPW-fed asymmetric UWB antenna with sufficient WLAN- band rejection. Progress In Electromagnetics Research C 108, (2021), 201–210. [19] R. M. Gomathi and J. M. L. Manickam. 2018. Energy-efficient shortest path routing protocol for underwater acoustic wireless sensor network. Wireless Personal Communications 98, 1 (2018), 843–856. [20] S. Karim, F. K. Shaikh, B. S. Chowdhry, et al. 2021. GCORP: Geographic and cooperative opportunistic routing protocol for underwater sensor networks. IEEE Access 9, (2021), 27650–27667. [21] A. Aydoğdu, T. J. Hoar, T. Vukicevic, et al. 2018. OSSE for a sustainable marine observing network in the Sea of Marmara. Nonlinear Processes in Geophysics 25, 3 (2018), 537–551. [22] C. Günther, A. Temming, J. Santos, et al. 2021. Small steps high leaps: Bio-economical effects of changing cod end mesh size in the North Sea brown shrimp fishery. Fisheries Research 234 (2021), 105797. [23] J. Berli, M. Bunel, and C. Ducruet. 2018. Sea-land interdependence in the global maritime network: The case of Aus- tralian port cities. Networks and Spatial Economics 18, 3 (2018), 447–471. [24] Y. Jo and S. Lim. 2019. Design and implementation of heterogeneous surface gateway for underwater acoustic sensor network. International Journal of Electrical and Computer Engineering 9, 2 (2019), 1226. [25] J. Liu, W. Guan, G. Han, et al. 2019. A dynamic surface gateway placement scheme for mobile underwater networks. Sensors 19, 9 (2019), 1993. [26] N. Goyal, M. Dave, and A. K. Verma. 2019. Protocol stack of underwater wireless sensor network: Classical approaches and new trends. Wireless Personal Communications 104, 3 (2019), 995–1022. [27] N. Goyal, J. K. Sandhu, and L. Verma. 2019. Machine learning based data agglomeration in underwater wireless sensor networks. International Journal of Management, Technology and Engineering 9, 6 (2019), 240–245. [28] J. E. Smith and N. Pinter-Wollman. 2021. Observing the unwatchable: Integrating automated sensing, naturalistic observations and animal social network analysis in the age of big data. Journal of Animal Ecology 90, 1 (2021), 62–75. [29] E. Yalcin, O. Arslan, and M. Aymelek. 2020. Developing a policy management algorithm for ship provision: A Delphi technique integrated with hesitant fuzzy set (DTIHFS) approach. Maritime Policy & Management 47, 8 (2020), 1097– [30] D. Nikolic, N. Stojkovic, and N. Lekic. 2018. Maritime over the horizon sensor integration: High frequency surface- wave-radar and automatic identification system data integration algorithm. Sensors 18, 4 (2018), 1147. [31] T. Yang, J. Chen, and N. Zhang. 2020. AI-empowered maritime Internet of Things: A parallel-network-driven ap- proach. IEEE Network 34, 5 (2020), 54–59. [32] G. Santana Sosa, J. Santana Abril, J. Sosa, et al. 2020. Design of a practical underwater sensor network for offshore fish farm cages. Sensors 20, 16 (2020), 4459. [33] Y. Zhang, M. Wang, J. Liang, et al. 2017. Coverage enhancing of 3D underwater sensor networks based on improved fruit y fl optimization algorithm. Soft Computing 21, 20 (2017), 6019–6029. [34] Z. Wan, S. Liu, W. Ni, et al. 2019. An energy-efficient multi-level adaptive clustering routing algorithm for underwater wireless sensor networks. Cluster Computing 22, 6 (2019), 14651–14660. [35] Y. Guo, Q. Han, and X. Kang. 2020. Underwater sensor networks localization based on mobility-constrained beacon. Wireless Networks 26, 4 (2020), 2585–2594. [36] Y. Zhuang, C. Wu, H. Wu, et al. 2020. Event coverage hole repair algorithm based on multi-AUVs in multi-constrained three-dimensional underwater wireless sensor networks. Symmetry 12, 11 (2020), 1884. [37] J. Zhang, M. Cai, G. Han, et al. 2020. Cellular clustering-based interference-aware data transmission protocol for underwater acoustic sensor networks. IEEE Transactions on Vehicular Technology 69, 3 (2020), 3217–3230. [38] S. Mehta and A. Malik. 2020. A swarm intelligence based coverage hole healing approach for wireless sensor networks. EAI Endorsed Transactions on Scalable Information Systems 7 (2020), 26. ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022. Artificial Intelligence in Underwater Digital Twins Sensor Networks 39:27 [39] J. P. Devarajan and T. P. Robert. 2020. Swarm intelligent data aggregation in wireless sensor network. International Journal of Swarm Intelligence Research (IJSIR) 11, 2 (2020), 1–18. [40] Y. Feng, S. Zhao, and H. Liu. 2020. Analysis of network coverage optimization based on feedback k-means clustering and artificial fish swarm algorithm. IEEE Access 8, (2020) 42864–42876. [41] M. Neshat, G. Sepidnam, M. Sargolzaei, et al. 2014. Artificial fish swarm algorithm: A survey of the state-of-the-art, hybridization, combinatorial and indicative applications. Artificial Intelligence Review 42, 4 (2014), 965–997. [42] Z. Jin, J. Wang, and Y. Su. 2017. A routing protocol of underwater cognitive acoustic networks for marine mammals. Journal of Xi‘an Jiaotong University 51, 2 (2017), 33–39. [43] N. Saeed, A. Celik, T. Y. Al-Naffouri, et al. 2018. Energy harvesting hybrid acoustic-optical underwater wireless sensor networks localization. Sensors 18, 1 (2018), 51. [44] Z. Jin, Y. Ma, Y. Su, et al. 2017. A Q-learning-based delay-aware routing algorithm to extend the lifetime of underwater sensor networks. Sensors 17, 7 (2017), 1660. [45] B. Mishachandar and S. Vairamuthu. 2021. An underwater cognitive acoustic network strategy for efficient spectrum utilization. Applied Acoustics 175, (2021), 107861. [46] R. M. Gomathi and J. M. L. Manickam. 2018. Energy efficient shortest path routing protocol for underwater acoustic wireless sensor network. Wireless Personal Communications 98, 1 (2018), 843–856. [47] J. Zhang, M. Cai, G. Han, et al. 2020. Cellular clustering-based interference-aware data transmission protocol for underwater acoustic sensor networks. IEEE Transactions on Vehicular Technology 69, 3 (2020), 3217–3230. [48] A. Khasawneh, M. S. B. Abd Latiff, O. Kaiwartya, et al. 2018. A reliable energy-efficient pressure-based routing pro- tocol for underwater wireless sensor network. Wireless Networks 24, 6 (2018), 2061–2075. [49] V. Arrichiello and P. Gualeni. 2020. Systems engineering and digital twin: A vision for the future of cruise ships design, production and operations. International Journal on Interactive Design and Manufacturing (IJIDeM) 14, 1 (2020), 115– [50] T. Y. Pang, J. D. Pelaez Restrepo, C. T. Cheng, et al. 2021. Developing a digital twin and digital thread framework for an ‘Industry 4.0’Shipyard’. Applied Sciences 11, 3 (2021), 1097. Received April 2021; revised October 2021; accepted December 2021 ACM Transactions on Sensor Networks, Vol. 18, No. 3, Article 39. Publication date: April 2022.
Journal
ACM Transactions on Sensor Networks (TOSN) – Association for Computing Machinery
Published: Apr 18, 2022
Keywords: Marine monitoring