A Dynamic Hierarchical Routing Method for Underwater Acoustic Sensor Networks Based on BP Neural Network

Abstract

A dynamic hierarchical routing method for underwater acoustic sensor networks based on BP neural networks, which is related to underwater communication. The underwater acoustic sensor network is divided into a top layer, a middle layer, and a bottom layer. BP neural networks are used to extract node and environmental feature information, and the network is hierarchically trained to obtain a dynamic hierarchical routing model based on BP neural networks. Nodes in different routing layers design different weight calculation methods according to the importance of information, the level to which they belong, and the depth information of the nodes to form a node forwarding function F F . Nodes are activated by underwater sound sources and transmit information to the surface sink nodes. Radio communication is carried out between sink nodes to transmit data to the onshore control center. The sink nodes process the underwater information sensed by the nodes within their corresponding water areas, and the BP neural network trains the dynamic hierarchical model for hierarchical prediction. The prediction results are broadcast by the sink nodes to each node, and the nodes dynamically adjust their own working modes to improve the quality of underwater communication, reduce energy holes, and extend the network life

Description

Technical Field

[0001] The present invention relates to BP neural networks and underwater communication, and in particular to a dynamic hierarchical routing method for an underwater acoustic sensor network based on a BP neural network. Background Art

[0002] As a main means for humans to study underwater environments such as the ocean, underwater acoustic sensor networks have been widely applied in many fields such as underwater exploration, military surveillance, data collection, and underwater disaster prevention. In the face of the characteristics of underwater acoustic sensor networks such as low bandwidth, long delay, high energy consumption, and limited energy, designing a more efficient and reliable underwater routing protocol to improve network throughput, reduce end-to-end transmission delay, and other issues has become one of the main directions of current development.

[0003] With the continuous development and update of artificial intelligence technology, various model algorithms have been widely applied in fields such as computer vision, natural language processing, and medical technology. Among them, the backpropagation (BP) neural network, as a classic and mature artificial neural network, has a relatively simple principle and consumes less computing resources. Its network structure includes three layers: an input layer, a hidden layer, and an output layer. By continuously correcting the weights of each layer through the forward propagation of signals and the backpropagation of errors until the error is reduced to a threshold. The BP neural network can be used to implement functions such as inference, prediction, and classification (Zhang Xudong. Research on the implementation method of backpropagation neural network [D]. Tianjin University, 1995.). Due to the changing underwater environment, underwater acoustic nodes are easily affected by water body movement, biological activities, and other emergencies, resulting in their inability to work properly or temporarily withdraw from the network, thereby affecting the efficiency and success rate of underwater data transmission. Using a BP neural network to predict the hierarchical situation and lifespan of network nodes can better improve the ability to cope with the instability of underwater acoustic nodes and enhance the overall robustness and risk resistance of the underwater communication network.

[0004] Given that the positioning technology of underwater acoustic nodes is relatively difficult, Hai Yan et al. (Hai Y, et al. DBR: Depth-Based Routing for Underwater Sensor Networks[C], 2008.) proposed a strategy for routing selection using the depth information of nodes, obtaining the node depth value through the pressure sensors carried by the nodes; during data forwarding, only target nodes with a depth less than the current node are selected for forwarding, which can effectively avoid the need to obtain global location information. Due to the special network environment of underwater acoustic sensor networks, the load levels of underwater sensor nodes vary, resulting in some nodes dying prematurely and forming network holes, causing intermittent interruptions and additional delays in information transmission, and further leading to the accelerated death of the surrounding nodes, seriously affecting the overall lifespan of the network. To provide a good solution and make the overall energy load of the network more balanced, research has been conducted on hierarchical routing algorithms for dense underwater acoustic sensor networks.

[0005] Among various hierarchical routing algorithms for underwater acoustic sensor networks, Jafri M R et al. (Jafri M R, et al. iAMCTD: Improved Adaptive Mobility of Courier Nodes in Threshold-Optimized DBR Protocol for Underwater Wireless Sensor Networks[J]. International Journal of Distributed Sensor Networks, 2014.) designed different-level data forwarding functions in combination with the on-demand principle after simple hierarchical division in the vertical direction. This method takes into account the overall load balance of the network, reduces network energy consumption and propagation loss while taking into account network throughput. Liu Yang et al. (Liu Yang, et al. Routing Algorithm for River Underwater Sensor Networks Based on Hierarchy[J]. Computer Applications, 2016) combined the river environment, conducted a mechanical analysis of the tethered nodes, and proposed a hierarchical routing strategy that is superior to the depth-based routing algorithm in terms of network redundancy and packet loss rate, and the network survival period is also significantly improved.

[0006] To sum up, current research on the hierarchical problem of underwater acoustic sensor network routing mostly divides by propagation distance centered on nodes or makes simple fixed hierarchical division in terms of depth, and no relevant research has been conducted by combining the BP neural network and the hierarchical routing technology of underwater acoustic sensor networks from the perspective of the global deployment of the overall network. Summary of the Invention

[0007] The object of the present invention is to provide a dynamic hierarchical routing method for underwater acoustic sensor networks based on BP neural networks. Combining underwater environmental factors, a node data forwarding function is designed to select the best receiving node, improve the dynamic hierarchical routing mechanism, and apply it to solve the problems of energy holes and "communication dead zones" easily generated in dense underwater acoustic sensor networks, so as to obtain a dynamic hierarchical routing forwarding method that takes into account network load balancing and can improve the overall network survival period.

[0008] The present invention includes the following steps:

[0009] 1) Assume that the sensor nodes of the underwater acoustic sensor network are randomly distributed, denoted as node 1, node 2, node 3,.... The network is divided into three-layer structures: the top layer (L1), the middle layer (L2), and the bottom layer (L3) in sequence from the water surface to the bottom and from the lower level to the higher level; multiple sink nodes are distributed on the water surface, denoted as sink1, sink2, sink3,.... The sink nodes can transmit information to the shore-based control center through radio communication; each sink node maintains an information table of the network nodes in its own water area. Considering the node drift, the number of nodes in the sub-network within its corresponding water area is not fixed;

[0010] Among them, the sink nodes have strong computing capabilities, are responsible for processing the information forwarded by all nodes within their corresponding water areas, and are also responsible for calculating the hierarchical situation of all nodes. They are powered by solar energy and do not consider energy consumption issues;

[0011] 2) The hierarchical data packet contains the hierarchical results predicted by the dynamic hierarchical routing model based on the BP neural network. The sink nodes perform dynamic hierarchical division of the network by periodically broadcasting the hierarchical data packets downward, and the broadcast range can reach the lower boundary of the entire network, that is, each node in the network can receive the updated data of the hierarchical results;

[0012] The prediction steps of the above-mentioned dynamic hierarchical routing model based on the BP neural network are specifically as follows:

[0013] 2.1) Match the characteristic information such as the initial hierarchical situation, node depth, and remaining energy of all nodes in the overall network with the simulated hierarchical results after different rounds of transmission as model samples to establish a sample library;

[0014] 2.2) Randomly divide the sample library into a training set and a test set for training;

[0015] 2.3) Select the node depth, remaining energy, and initial level as characteristic information as the input, and the level to which the node belongs after R rounds of transmission as the output to establish a dynamic hierarchical routing model based on the BP neural network;

[0016] 2.4) Train the dynamic hierarchical routing model based on the BP neural network according to the training set;

[0017] 2.5) Test the trained dynamic routing hierarchical model based on the BP neural network according to the test set, calculate the error Loss between the predicted value and the simulated value, backpropagate the error Loss, and iteratively adjust the model parameters until a numerically small and stable error Loss is obtained to get the model parameters;

[0018] 2.6) Input the node feature information to be predicted into the dynamic hierarchical routing model based on the BP neural network, and the output result is the predicted level of the node after R rounds of transmission;

[0019] 3) After the underwater acoustic node receives the hierarchical data packet, extract the hierarchical information corresponding to its own node ID information in the hierarchical data packet, record its own level, and update and maintain the information table of its neighboring nodes according to the latest hierarchical result;

[0020] 4) All underwater acoustic nodes can be triggered by the sound source and become the source node to originate the data packet after being triggered; the non-Sink node containing the data packet in the routing is the sending node; set the emergency degree threshold of the data, the hard threshold is represented by V th and the soft threshold is represented by V ts . Then, the routing request process of the sending node when sending data is divided into three cases in combination with the data emergency degree: if the data emergency degree is greater than the hard threshold V th , immediately perform data transmission; if the data emergency degree is less than the soft threshold V ts , discard the data and do not transmit; if the data emergency degree is between V th and the soft threshold V ts , further consider the remaining energy of the sending node; if the remaining energy of the sending node is greater than the remaining energy threshold R t , perform transmission; if it is less than R t , wait to consider transmission again in the next round of forwarding, the emergency degree of this data packet decreases accordingly, and repeat the current step 4);

[0021] 5) If the node in step 4) confirms the need to forward the data packet, to reduce the same-layer forwarding of the data packet, the sending node will select the best next-hop forwarding node from the neighbor node information table as the receiving node for data forwarding;

[0022] 6) The receiving node is updated to the new sending node;

[0023] 7) The sending node judges whether the next hop can reach the Sink node according to whether the communication range contains the Sink node. If it has not reached, repeat steps 5) and 6) until it is successfully transmitted to the Sink node; otherwise, it is regarded as a communication interruption;

[0024] 8) Considering that in different routing hierarchical structures, the environmental conditions, energy consumption, belonging levels, and optimization objectives of nodes are different, the signal-to-noise ratio, remaining energy, and depth parameters for calculating the weights of neighbor nodes will change. A forwarding function is designed to select the best receiving node. The forwarding weights of all neighbor nodes within the transmission range of a node are represented as W1, W2, W3, … respectively. Then, the node to be forwarded will select the neighbor node with the largest weight value as the receiving node for forwarding. The forwarding function is expressed as: F F = W max ; If the Sink node is included within the transmission range, then F F = W sink ;

[0025] In step 8), according to the different characteristics of the environmental conditions of the water layers where the nodes are located, their own energy consumption, and forwarding optimization objectives, the signal-to-noise ratio, remaining energy, and depth parameters for calculating the weights change. Then, the expression of the forwarding function F F is different in different layers of the three-layer routing structure. The specific expression formula is:

[0026] 8.1) Layer L1: The nodes within the layer have a shallow depth and are close to the water surface. They are significantly affected by interface interference, ship activities, and noise. To ensure signal quality, the forwarding function should pursue the minimum propagation loss to achieve the best signal-to-noise ratio. The nodes within the layer are close to the Sink node, and energy holes are likely to occur. Considering the remaining energy information of the nodes, the formation of energy holes is reduced, and the network lifetime is extended. Then, the forwarding function F F is expressed as:

[0027]

[0028]

[0029] In the formula, w1 is a weight constant, which can be adjusted according to the specific scenario; E res is the remaining energy of the node; dr i is the Euclidean distance between the transceiver nodes; L is the level to which the node belongs, and L = 1 in L1; LSNR is the non-localization signal-to-noise ratio; P t is the constant transmission power; dl is the depth difference between the transceiver nodes; f is the signal frequency (kHz); A is the path attenuation loss, and the product of it and the environmental noise N constitutes the environmental attenuation coefficient;

[0030] 8.2) Layer L2: The data forwarding of the nodes within the layer is the most frequent, and the number of nodes is large. If an energy hole appears, it will cause a large-scale and rapid death of the surrounding nodes. Therefore, it is necessary to consider the influence of both transmission energy consumption and transmission efficiency to reduce the influence of energy holes and maximize the realization of network energy balance so as to effectively improve the network lifetime. Then, the forwarding function FF It is expressed as:

[0031]

[0032] In the formula, w2 is a weight constant, which can be adjusted according to the specific scenario; E res is the remaining energy of the node; dl is the depth difference between the sending and receiving nodes; L is the layer to which the node belongs, and in L2, L = 2;

[0033] 8.3) Layer L3: The in-layer channel loss is not large, the node forwarding load is large, there are many marine organisms, the underwater interference has a great impact, and it is also affected by the mineral distribution. It pursues long-distance (mainly in the vertical direction) transmission to reduce the impact of flooding. Then the forwarding function F F It is expressed as:

[0034]

[0035] In the formula, w3 is a weight constant, which can be adjusted according to the specific scenario; E res is the remaining energy of the node; LSNR is the non-positioning signal-to-noise ratio; dl is the depth difference between the sending and receiving nodes; l is the node depth; L is the layer to which the node belongs, and in L3, L = 3;

[0036] 9) If the current sending node has no receiving node for transmission and cannot directly transmit to the Sink node, the data packet will be returned to the previous hop node. After deleting the current node from the information table of the neighbors of the previous hop node, then repeat steps 5), 6), and 7).

[0037] The present invention divides the underwater acoustic sensor network into a top layer, a middle layer, and a bottom layer, uses a BP neural network to extract node and environmental feature information, trains the network layer by layer, and obtains a dynamic hierarchical routing model based on the BP neural network. The model is used for hierarchical prediction of routing. The underwater acoustic sensor nodes in different routing layers design different weight calculation methods according to the information importance level, the layer to which they belong, and the node depth information, to form the node forwarding function F F . After being activated by an underwater sound source, the node transmits information to the surface convergence (Sink) node. The Sink nodes communicate via radio, and finally transmit the data to the onshore control center. Each Sink node is responsible for processing the underwater information sensed by the nodes within its corresponding water area range, and uses the dynamic hierarchical model trained by the BP neural network according to the remaining energy, depth information, and neighbor topology structure of the nodes for hierarchical prediction. The prediction results are broadcast by the Sink node to each node, and the node dynamically adjusts its own working mode to achieve the effects of improving underwater communication quality, reducing energy holes, and extending the network life.

[0038] The present invention can effectively balance the load of network nodes in underwater acoustic data transmission, thereby reducing the occurrence of network energy holes and extending the overall network life and survival period. Considering the problems of unbalanced node load and possible "communication dead zones" in the data transmission process in dense underwater acoustic sensor networks, the present invention intends to combine the on-demand principle and a hierarchical data forwarding mode, and perform dynamic hierarchical routing based on a BP neural network, aiming to obtain an underwater acoustic network routing selection method that can balance the network load energy and extend the overall network life.

[0039] The present invention has the following outstanding advantages:

[0040] 1) In underwater acoustic communication, the physical layer loss and the total energy consumption of the entire network are related to the distance, and the distance and energy are in an exponential relationship. Short-distance transmission can effectively save energy, but using only the distance as the standard for selecting transmission may result in "communication dead zones" problems. By adopting a hierarchical routing strategy, data is transmitted layer by layer and hop by hop, avoiding repeated transmission among nodes in the same layer, effectively solving the "communication dead zones" problem, and improving network connectivity, reliability, and fault tolerance;

[0041] 2) Design a dynamic hierarchical mechanism, considering the situations of node death and node movement due to water flow, etc., which causes deviation from the original layer. Dynamically update the modulation routing layer from the perspective of the overall network global deployment, improve the flexibility of the mechanism, ensure the network performance when the network density decreases, and at the same time extend the network survival period;

[0042] 3) Considering the working characteristics of nodes at different layers, design forwarding functions for nodes at different layers, optimize the routing scheme based only on depth, and can reduce the "communication dead zones" of the network, balance the network load, and improve the network life.

[0043] 4) Use a BP neural network to predict the hierarchical situation and life of network nodes. While avoiding collecting information of all nodes, it can effectively improve the overall robustness and risk resistance ability of the underwater communication network. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a data transmission scenario diagram under dynamic routing layer of an underwater acoustic sensor network

[0045] Figure 2 It is a flowchart of the dynamic hierarchical routing of an underwater acoustic sensor network based on a BP neural network of the present invention.

[0046] Figure 3 It is a comparison diagram of the hierarchical results of the dynamic hierarchical routing method of an underwater acoustic sensor network based on a BP neural network of the present invention at different network life stages.

[0047] Figure 4This is a comparison graph of the simulation hierarchical results and the BP neural network prediction hierarchical results of the underwater acoustic sensor network routing dynamic hierarchical method based on the BP neural network of the present invention. Detailed implementation manners

[0048] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0049] As Figure 1 and Figure 2 shown, the embodiments of the present invention include the following steps:

[0050] 1) Assume that the underwater acoustic sensor nodes in the underwater acoustic sensor network are randomly distributed, denoted as node 1, node 2, node 3, …, node i, …, and the network routing is divided into three-layer structures of the top layer (L1), the middle layer (L2) and the bottom layer (L3) from the water surface to the bottom and from the lower level to the higher level. There are n Sink nodes distributed on the water surface, denoted as sink1, sink2, sink3, …, sinkn; the Sink nodes can transmit information to the shore-based control center through radio communication; each Sink node maintains an information table of the network nodes in its own water area. Considering the node drift, the number of sub-network nodes in its corresponding water area range is not fixed; as Figure 1 shown, there are 11 randomly distributed nodes underwater, denoted as node 1, node 2, node 3, …, node 11, L1 has 3 nodes, L2 has 6 nodes, and L3 has 2 nodes; the underwater data collected by the nodes will all be forwarded to the water surface aggregation node, as shown in the figure as sink1;

[0051] Among them, sink1 has strong computing power and is responsible for processing the information forwarded by node 1, node 2, node 3, …, node 11. It is powered by solar energy and the energy consumption problem is not considered;

[0052] 2) The hierarchical data packet contains the hierarchical information of each node. Sink1 periodically broadcasts the hierarchical data packet downward, and the ID of the node corresponds one-to-one with the belonging layer; the broadcast range of sink1 can reach the lower boundary of the entire network, that is, node 1, node 2, node 3, …, node 11 can all receive the update information of their own layers;

[0053] 3) Before the data transmission starts, a prediction hierarchical model is pre-trained on the shore, as Figure 2 shown:

[0054] 3.1) First, establish a model sample library according to the specific water area of the actual scene;

[0055] 3.2) Randomly divide the sample library into a training set and a test set, and train the BP neural network model;

[0056] 3.3) Take the depth, remaining energy, and level information of the node as input quantities, and take the level it belongs to after R rounds of transmission as the output;

[0057] 3.4) Define the loss function Loss as the difference between the predicted value and the actual simulation value, and backpropagate the difference to continuously adjust the parameters of the BO neural network model;

[0058] 3.5) Determine whether the loss function Loss has converged. If it has not converged, repeat steps 3.2), 3.3), 3.4), and 3.5); if it has converged, the prediction hierarchical model training is completed and can be used in the next step;

[0059] 4) After nodes 1, 2, 3, …, 11 receive the hierarchical data packets, extract the hierarchical information corresponding to their own IDs in the packets, record their own levels, and update and maintain the information tables of their own neighbor nodes according to the latest hierarchical results;

[0060] 5) Each node may be triggered by a sound source. As Figure 1 shown, after node 11 is triggered, it becomes the source node. Its routing request process is as Figure 2 shown. Set the emergency degree threshold of the required data. The hard threshold is represented by V th and the soft threshold is represented by V ts . Combine the data emergency degree into three cases: If the data emergency degree is greater than the hard threshold V th , immediately perform data transmission; if the data emergency degree is less than the soft threshold V ts , discard the data and do not transmit; if the data emergency degree is between V th and the soft threshold V ts , further consider the remaining energy of node 11. If the remaining energy of node 11 is greater than the remaining energy threshold R t , perform transmission; if it is less than R t , wait to consider transmission in the next round of forwarding. At the same time, the emergency degree of the current data packet decreases accordingly, and repeat the current step until the data packet is discarded or confirmed for transmission;

[0061] 5) If node 11 confirms the need to forward data packets in step 4), node 11 enters the data forwarding stage. As Figure 2 shown, node 11 calculates the weights of neighbor nodes according to the forwarding function, and selects the best next-hop forwarding node from the information tables of neighbor nodes for data forwarding; as Figure 1 shown, node 11 belongs to L3, and there is a node in L2 within its transmission range, then it will be transmitted to node 8 in L2;

[0062] 6) When node 8 forwards data, although both node 5 and node 6 are its neighbor nodes, according to the forwarding function, it is calculated that node 6 is the best neighbor node, and data packets are delivered to it; repeat steps 4) and 5) until the data packets are successfully transmitted to sink1;

[0063] Considering that in different routing hierarchical structures, the environmental conditions, energy consumption, levels, and optimization objectives of nodes are different, and the SNR, remaining energy, and depth parameters for calculating the weights of neighbor nodes change, the forwarding functions are also different; the forwarding weights of neighbor nodes within the transmission range of a node are respectively expressed as W1, W2, W3,..., W k , …, then the node to be forwarded will select the node with the largest weight value among the neighbor nodes for forwarding, that is, the forwarding function F F = W max ; if the Sink node is included within the transmission range, then F F = W sink ; F F The specific expression formula is:

[0064] 6.1) Layer L1: The nodes within the layer have a shallow depth and are close to the water surface, and are significantly affected by interface interference, ship activities, and noise. To ensure signal quality, the forwarding function should pursue the minimum propagation loss to achieve the best SNR. The nodes within the layer are close to the Sink node, and energy holes are likely to occur. Considering the remaining energy information of the nodes, reducing the formation of energy holes and extending the network lifetime, the forwarding function F F is expressed as:

[0065]

[0066]

[0067] In the formula, w1 is a weight constant, which can be adjusted according to the specific scenario; E res is the remaining energy of the node; dr i is the Euclidean distance between the sending and receiving nodes; L is the level to which the node belongs, and L = 1 in L1; LSNR is the SNR without positioning; P t is the constant transmission power; dl is the depth difference between the sending and receiving nodes; f is the signal frequency (kHz); A is the path attenuation loss, and the product of it and the environmental noise N constitutes the environmental attenuation coefficient.

[0068] 6.2) Layer L2: The data forwarding of the nodes within the layer is the most frequent, and the number of nodes is large. If an energy hole appears, it will cause a large - scale and rapid death of the surrounding nodes. Therefore, it is necessary to consider the influence of both transmission energy consumption and transmission efficiency to reduce the influence of energy holes and maximize the realization of network energy balance to effectively improve the network lifetime. Then the forwarding function F F is expressed as:

[0069]

[0070] Wherein, w2 is a weight constant and can be adjusted according to specific scenarios; E res is the remaining energy of the node; dl is the depth difference between the sending and receiving nodes; L is the layer to which the node belongs, and in L2, L = 2.

[0071] 6.3) Layer L3: The in-layer channel loss is not large, the node forwarding load is large, there are many marine organisms, the underwater interference has a great impact and is affected by the mineral distribution. It pursues long-distance (mainly vertical direction) transmission to reduce the impact of flooding. Then the forwarding function F F is expressed as:

[0072]

[0073] Wherein, w3 is a weight constant and can be adjusted according to specific scenarios; E res is the remaining energy of the node; LSNR is the signal-to-noise ratio without positioning; dl is the depth difference between the sending and receiving nodes; l is the node depth; L is the layer to which the node belongs, and in L3, L = 3.

[0074] 7) After every 20 rounds of transmission, sink1 outputs the updated node hierarchical information according to the BP neural network prediction model, that is, the underwater routing hierarchy changes dynamically, and is updated by broadcasting downward through hierarchical packets;

[0075] 8) The training steps of the dynamic hierarchical routing model based on the BP neural network are as follows:

[0076] 8.1) For all nodes in the overall network, such as Figure 1 shown, that is, the initial layer L, node depth l, and remaining energy E res and other characteristic information of nodes 1, node 2, node 3,..., node 11 are matched with the simulated hierarchical results after R rounds of transmission (R = 20, 40, 60, 80...), and used as the model sample S to establish a sample library;

[0077] 8.2) The sample library is randomly divided into a training set S train and a test set S test for training;

[0078] 8.3) Select the node depth l, remaining energy E res and layer L as characteristic information and use them as the input X. Take the layer L R to which the node belongs after R rounds of transmission as the output Y, and establish a dynamic hierarchical routing model based on the BP neural network;

[0079] 8.4) Train the dynamic hierarchical routing model according to the training set S train ;

[0080] 8.5) According to the test set S test Test the dynamically routed hierarchical model after training, calculate the error e between the predicted value of L and the simulated value of L, backpropagate the error, and continuously iterate to adjust the model parameters until a numerically small and stable error is obtained to get the model parameters;

[0081] 8.6) Input the node feature information l, E res and L of the node to be predicted into the dynamic hierarchical routing model, and the output result is the predicted level L of the node after R rounds of transmission R .

[0082] Next, the feasibility of the method of the present invention is verified by computer simulation.

[0083] To simulate the underwater acoustic sensor network, as shown in Figure 3 (a), randomly arrange the underwater sensor node network topology model. There are a total of 64 nodes, denoted as node 1, node 2, node 3,..., node i,..., node 64; 4 Sink nodes, denoted as sink1, sink2, sink3, sink4; initially set that there are 16 L1 nodes, 24 L2 nodes, and 24 L3 nodes.

[0084] Assume that the maximum depth of the underwater acoustic sensor network is 1 km, the maximum horizontal range is 1 km, the maximum communication distance between two nodes is 200 m, and it is determined that the data packet is successfully transmitted to the Sink node, otherwise it is judged that the transmission fails. The power consumption of the node in the sending, receiving, and idle modes is 2 W, 0.1 W, and 10 mW respectively, and the node is initially equipped with 70 J.

[0085] The algorithm steps are as follows:

[0086] (1) Randomly select node i as the initial data sending node.

[0087] (2) Judge whether the next hop can reach the Sink node according to the depth information of the sending node. If it can, directly forward the data packet to the nearest Sink node to successfully complete the data transmission; if the next hop cannot directly reach the Sink node, then classify all the normal working live nodes within its communication range as the neighbor nodes of the sending node, calculate the neighbor node weights according to the forwarding function, and select the node j with the largest weight as the receiving node for data forwarding.

[0088] In step (2), the specific formula of the forwarding function is as follows:

[0089]

[0090] (3) Update the receiving node as the new sending node and repeat step (2).

[0091] (4) The successful or failed transmission of data packets to the Sink node is regarded as the end of a transmission round. After every 20 rounds of transmission, the level of the node is updated according to the level result predicted by the BP neural network dynamic hierarchical model.

[0092] (5) When the node acts as a sending node, the transmission power is 2W; when the node acts as a receiving node, the receiving power is 0.1W; when idle, the working power of the node is 10mW. If the remaining energy of the node is less than the energy threshold, the node is determined to be in a dead state, unable to communicate, and exits the network.

[0093] (6) Update the number of remaining live nodes in the network, the network density, and the average remaining energy of the network. The specific calculation formulas are as follows:

[0094]

[0095]

[0096] Where N alive represents the number of live nodes in the network, and N dead represents the number of dead nodes in the network, and e n represents the remaining energy of node n .

[0097] (7) When the network density is less than or equal to 20%, it is determined that the network is dead, that is, the network survival period is defined as the time when the network density drops from 100% to 20%.

[0098] To better display the effect of the dynamic hierarchical forwarding method, the method described in the invention is verified by computer simulation as follows:

[0099] As Figure 3 (b) shows, after 300 rounds of transmission, that is, after 15 updates of the hierarchical boundary, the network hierarchical result is significantly different from the initial network hierarchy shown in Figure 3 (a): the number of nodes in L1 is reduced to 8, the number of nodes in L2 is increased to 35, and the number of nodes in L3 is reduced to 21; as Figure 3 (c) shows, after 2000 rounds of transmission, the network has 62 L2 nodes and 2 L3 nodes, and there are no L1 nodes.

[0100] To reduce the computing pressure and load of the Sink node and verify the machine learning feasibility of the method described in the invention, the effect of dynamic hierarchical prediction using the BP neural network is simulated by computer as follows:

[0101] As Figure 4 shows, using the BP neural network to extract the feature information of the nodes and perform model training, the predicted hierarchical result obtained is asFigure 3 As shown in (d), compared with the actual simulation layering results, only the layer of 1 node was wrongly predicted among 64 nodes, and the accuracy rate could reach 98.4375%.

[0102] Based on the BP neural network, the present invention combines the on-demand principle and the hierarchical routing algorithm to give a dynamic hierarchical routing and forwarding scheme in an underwater acoustic sensor network based on the BP neural network; adopting the hierarchical idea, data is transmitted hop by hop layer by layer according to the layer, solving the "communication dead zone" problem and improving the network connectivity, reliability and fault tolerance; in underwater acoustic communication, the physical layer loss and the total energy consumption of the whole network are related to the distance, and the distance and the energy are in an exponential relationship. Considering the distance between nodes, short-distance transmission can effectively save energy; taking the remaining energy of the node as a measure index for electing candidate nodes, balancing the network energy consumption and improving the network lifetime; avoiding the difficulty of obtaining global node information, using the prediction and classification functions of the BP neural network to effectively simulate and predict the dynamic layering results of the underwater route, reducing the computational complexity while improving the network stability. The present invention uses the dynamic hierarchical routing algorithm to solve the problems of unbalanced network energy load and "communication dead zone", reducing the energy consumption of the underwater acoustic cooperative communication network, improving the communication efficiency and extending the network survival period.

Claims

1. A dynamic hierarchical routing method for an underwater acoustic sensor network based on a BP neural network, characterized in that: The sensor nodes of the underwater acoustic sensor network are randomly distributed, represented as node 1, node 2, node 3, …, and the network is sequentially divided into three-layer structures: the top layer L1, the middle layer L2, and the bottom layer L3 from the water surface to the bottom and from the lower level to the higher level; multiple sink nodes are distributed on the water surface, represented as sink1, sink2, sink3, …; The sink nodes transmit information to the shore-based control center through radio communication; each sink node maintains an information table of the network nodes in its own water area. Considering the node drift, the number of nodes in the sub-network within its corresponding water area is not fixed; Select the node depth, remaining energy, and initial level as feature information as the input, and use the level of the node after R rounds of transmission as the output to establish a dynamic hierarchical routing model based on a BP neural network; According to the different characteristics of the water layer environment conditions where the node is located, its own energy consumption situation, and different forwarding optimization objectives, the signal-to-noise ratio, remaining energy, and depth parameters for weight calculation change, and then the forwarding function F F is expressed differently in different layers of the three-layer routing structure. The specific expression formula is: (1) Layer L1: The depth of the nodes within the layer is shallow and they are close to the water surface. They are significantly affected by interface interference, ship activities, and noise. To ensure signal quality, the forwarding function should pursue the minimum propagation loss to achieve the best signal-to-noise ratio; the nodes within the layer are close to the Sink node, and energy holes are likely to occur. Considering the remaining energy information of the nodes, reducing the formation of energy holes, and extending the network lifetime, the forwarding function F F is expressed as: Where, w1 is a weight constant, which is adjusted according to specific scenarios; E res is the remaining energy of the node; dr i is the Euclidean distance between the transceiver nodes; L is the level to which the node belongs, and L = 1 in L1; LSNR is the positioning-free signal-to-noise ratio; P t is the constant transmission power; dl is the depth difference between the transceiver nodes; f is the signal frequency; A is the path attenuation loss, and the product of it and the environmental noise N constitutes the environmental attenuation coefficient; (2) Layer L2: The data forwarding among nodes within the layer is the most frequent, and the number of nodes is large. If an energy hole appears, it will cause a large-scale and rapid death of surrounding nodes. Therefore, it is necessary to consider the influence of both transmission energy consumption and transmission efficiency at the same time to reduce the influence of energy holes, achieve network energy balance, and thus improve the network lifetime. Then, the forwarding function F F is expressed as: where w2 is a weight constant, adjusted according to specific scenarios; E res is the remaining energy of the node; dl is the depth difference between the sending and receiving nodes; L is the layer to which the node belongs, and in L2, L = 2; (3) Layer L3: The in-layer channel loss is not significant, the node forwarding load is high, there are many marine organisms, the underwater interference has a great impact, and it is also affected by the mineral distribution. To pursue long-distance transmission and reduce the impact of flooding, the forwarding function F F is expressed as: where w3 is a weight constant, which is adjusted according to the specific scenario; E res is the remaining energy of the node; LSNR is the positioning-free signal-to-noise ratio; dl is the depth difference between the transceiver nodes; l is the depth of the node; L is the layer to which the node belongs, and in L3, L = 3.

2. The dynamic hierarchical routing method for an underwater acoustic sensor network based on a BP neural network according to claim 1, the method further includes: The hierarchical data packet contains the hierarchical result predicted by the dynamic hierarchical routing model based on the BP neural network. The sink node dynamically hierarchizes the network by periodically broadcasting the hierarchical data packet downward, and the broadcast range reaches the lower boundary of the entire network, that is, each node in the network can receive the updated data of the hierarchical result.

3. The dynamic hierarchical routing method for an underwater acoustic sensor network based on a BP neural network according to claim 1, the method further includes: All underwater acoustic nodes are triggered by the sound source and become source nodes to originate data packets after being triggered; non-Sink nodes that contain data packets in the route are sending nodes; set the emergency level threshold of the data, the hard threshold is represented by V th and the soft threshold is represented by V ts . Then, the route request process of the sending node when sending data is divided into three cases according to the data emergency level: if the data emergency level is greater than the hard threshold V th , then data transmission; if the data emergency level is less than the soft threshold V ts , then discard the data and do not transmit; If the data urgency level is at V th and the soft threshold V ts then consider the remaining energy of the sending node; if the remaining energy of the sending node is greater than the remaining energy threshold R t , then perform transmission; if it is less than R t , then wait to consider transmission again during the next round of forwarding, and the urgency level of this data packet will be reduced accordingly, and repeat the above method.

Images