Multi-cluster-head optimal multi-hop routing path scheme acquisition method and related equipment

Abstract

The invention discloses a multi-cluster-head optimal multi-hop routing path scheme acquisition method and related equipment, and the method comprises the steps: achieving the reasonable switching of a working state and a dormant state of a node through the quantitative analysis of the coverage redundancy of a three-dimensional space of a sensor node; carrying out clustering analysis on network nodes by using an adaptive density peak value clustering algorithm to realize adaptive clustering of WSNs, adaptively adjusting truncation distance parameters by using the adaptive density peak value clustering algorithm according to residual energy of the network nodes, and replacing traditional distance measurement by using a covariance distance; using an improved zebra optimization algorithm to optimally configure clustering data fusion model parameters, and performing intra-cluster node data fusion so as to reduce the total network data communication amount; and establishing a multi-cluster-head routing communication model, and designing a discrete zebra optimization algorithm to solve the model so as to obtain a multi-cluster-head optimal multi-hop routing path scheme. Experimental results show that the total network communication amount and the total energy consumption of the algorithm are obviously reduced, and the network lifetime is greatly prolonged.

Description

Technical Field

[0001] The present invention belongs to the technical field of wireless sensor network optimization, and particularly relates to a method for obtaining an optimal multi-hop routing path scheme for multiple cluster heads and related devices. Background Art

[0002] As a network system formed by a large number of micro wireless communication sensor nodes self-organizing, wireless sensor networks (WSNs) can collect and process various information of monitored objects in real time, and are increasingly becoming one of the most important and crucial links in earthquake disaster monitoring and early warning systems. However, limited by the limited energy and data processing capabilities of sensor nodes, how to more effectively reduce the total network communication volume, network energy consumption, and extend the network lifetime has become a bottleneck restricting the development of WSNs.

[0003] Research shows that by using the storage and computing capabilities of some nodes to fuse the data collected by other nearby nodes, the data collection efficiency can be improved and network energy can be saved. Optimizing the clustering topology structure and formulating a reasonable data transmission mechanism are the current research hotspots in data fusion of WSNs. In terms of the clustering topology structure, LEACH (low energy adaptive clustering hierarchy), as a classic routing protocol, first proposed the concept of clustering and solved the problem of network load balancing well. However, since this protocol randomly selects cluster heads, problems such as "energy holes" are likely to occur. Subsequently, a large number of routing protocols based on the clustering idea have emerged one after another, such as the EAUCA protocol, the CDEIR protocol, etc. Most of these protocols use a fixed clustering method, which is easy to implement but may lead to problems such as uneven clustering structure in the later stage. Zhang Zhaohui et al. proposed a semi-fixed clustering routing protocol, which tries to keep the number of nodes in each cluster the same by combining the network energy consumption situation, ensuring the uniformity of the clustering structure. However, this protocol artificially sets the threshold for selecting cluster heads, increasing the implementation difficulty. Yang Jing et al. proposed a clustering routing algorithm based on IHBA and fuzzy C-means for a more complex three-dimensional application scenario. This algorithm uses the fuzzy C-means clustering algorithm to divide the three-dimensional network topology structure, with good scalability. However, this routing protocol cannot effectively avoid the appearance of extremely small clusters and extremely large clusters. In terms of the data transmission mechanism, transforming the multi-hop data communication problem between cluster heads into an objective optimization problem and using intelligent optimization algorithms to solve it is the current research hotspot. Sun Aijing et al. established a multi-hop routing model based on indicators such as node energy and position distance, and used the particle swarm optimization algorithm (PSO) to optimize the model; there is also a routing clustering algorithm based on the ant colony algorithm, which uses the ant colony algorithm to find the optimal multi-hop path for data transmission; Zhao Xiaoqiang et al. proposed a routing protocol based on simulated annealing, which uses the simulated annealing algorithm to find the optimal set of cluster heads and realizes the data transmission path planning. The above research combines the advantages of the strong global optimization ability of intelligent optimization algorithms and obtains relatively reasonable network data transmission paths. However, most of these routing protocols do not fully consider the defect that intelligent optimization algorithms are prone to fall into, and the obtained data transmission paths may not be optimal.

[0004] How to better utilize information such as the location and energy of network nodes to achieve network adaptive clustering, reduce the total network communication volume by deep learning nodes to collect data, and find the best data communication path is worthy of further in-depth research. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for obtaining a multi-cluster-head optimal multi-hop routing path scheme and related devices to solve the technical problem of the inaccuracy of the existing method for obtaining the optimal network data transmission path based on wireless sensor networks.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A method for obtaining a multi-cluster head optimal multi-hop routing path scheme, comprising the following steps: Analyze the three-dimensional space coverage redundancy of nodes in a wireless sensor network by using a node working state control algorithm to determine the working nodes in the wireless sensor network; Perform spatial clustering analysis on the working nodes in the wireless sensor network by using an adaptive density peak clustering algorithm to achieve adaptive dynamic clustering of the wireless sensor network; For the clustered wireless sensor network, establish a clustered data fusion model based on deep learning, and use an improved zebra optimization algorithm to optimize the configuration of the parameters of the clustered data fusion model to fuse the data of the working nodes within the cluster; Based on the fused data of the working nodes within the cluster, establish a multi-cluster head routing communication model; Solve the multi-cluster head routing communication model to obtain the multi-cluster head optimal multi-hop routing path.

[0007] Further, the determination steps of the working nodes in the wireless sensor network are as follows: Uniformly divide the three-dimensional space network of the wireless sensor to determine the working node candidate set and the network edge node set; First, determine the working state of the nodes in the working node candidate set, and then determine the working state of the nodes in the network edge node set; The steps for determining the working state of the nodes in the working node candidate set are as follows: Compare the absolute node coverage probability of the node with the largest node density in each grid in the working node candidate set with the lowest coverage rate of the given monitoring area; If the absolute node coverage probability of the node is less than the lowest coverage rate of the given monitoring area, set the state of the node to "working", and if the absolute node coverage probability of the node is greater than or equal to the lowest coverage rate of the given monitoring area, set the state of the node to "sleep"; After the determination is completed, update the node state in the grid, and then determine the node with the largest density in the new grid; The determination method of the working state of the nodes in the network edge node set is the same as that of the nodes in the working node candidate set.

[0008] Further, the determination condition for switching the determination of the working state of the nodes in the working node candidate set and the network edge node set is: The grid coverage rate of the grid where the new node with the largest density is located after the update is less than the lowest coverage rate of the given monitoring area.

[0009] Further, the adaptive density peak clustering algorithm is an improvement based on the density peak clustering algorithm, including adaptively adjusting the truncation distance parameter according to the remaining energy of network nodes and using covariance distance to replace the traditional distance metric. The expression of the truncation distance is:

[0010] In the formula, is the cut-off distance, λ is the proportionality coefficient, is representing the remaining energy of the node.

[0011] Further, the clustered data fusion model is constructed based on the BP neural network. After the BP neural network model is constructed, the Sink node trains the model parameters and feeds back the training results to the cluster head. The cluster head extracts the data features of the nodes in the cluster using the trained BP neural network, and transmits the fused data to the Sink node through the multi-hop routing path. The cluster head is the center of each cluster after adaptive dynamic clustering by the adaptive density peak clustering algorithm.

[0012] Further, the improved zebra optimization algorithm adds a small-range depth search strategy, an extreme reverse defense evolution strategy for peripheral isolated individuals, and a perturbation evolution method for pioneer zebras on the basis of the zebra optimization algorithm. The parameters of the clustered data fusion model are assigned to the zebra individuals of the improved zebra optimization algorithm and the objective function is defined to complete the parameter optimization.

[0013] Further, the multi-cluster-head routing communication model is:

[0014] In the formula, represents the distance between and represents the distance between and represents the distance between and where c in is the sequence label of among all cluster heads, and k is the sequence number of among all multi-hop cluster heads;

[0015] In the second aspect, a system for obtaining a multi-cluster-head optimal multi-hop routing path scheme is provided, including a node determination module, a clustering module, a fusion module, a model construction module, and a solution module, where: Node determination module: used to analyze the three-dimensional spatial coverage redundancy of nodes in a wireless sensor network by using a node working state control algorithm, and determine the working nodes in the wireless sensor network; Clustering module: used to perform spatial clustering analysis on the working nodes in the wireless sensor network through an adaptive density peak clustering algorithm, and realize the adaptive dynamic clustering of the wireless sensor network; Fusion module: for the clustered wireless sensor network, establish a clustered data fusion model based on deep learning, and use an improved zebra optimization algorithm to optimize the configuration of the parameters of the clustered data fusion model, and fuse the data of the working nodes within the cluster; Model construction module: based on the data of the working nodes within the fused cluster, establish a multi-cluster head routing communication model; Solution module: solve the multi-cluster head routing communication model to obtain the best multi-hop routing path of the multi-cluster head..

[0016] A terminal device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above method are implemented.

[0017] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0018] Compared with the prior art, the present invention has the following beneficial technical effects: The present invention provides a method for obtaining a multi-cluster head best multi-hop routing path scheme. By quantitatively analyzing the three-dimensional spatial coverage redundancy of sensor nodes, it realizes the reasonable switching of the working and sleeping states of nodes; designs an adaptive density peak clustering algorithm, which adaptively adjusts the cut-off distance parameter according to the remaining energy of network nodes, and uses the covariance distance to replace the traditional distance metric; uses the adaptive density peak clustering algorithm to perform clustering analysis on network nodes to realize the adaptive clustering of WSNs. In the data preprocessing stage, an improved zebra optimization algorithm is used to optimize the configuration of the parameters of the clustered data fusion model and perform data fusion of the nodes within the cluster to reduce the total network data communication volume. In the data routing and transmission stage, a multi-cluster head routing communication model is established, and a discrete zebra optimization algorithm is designed to solve the model to obtain a multi-cluster head best multi-hop routing path scheme. Experimental results show that compared with other data collection algorithms, the total network communication volume and total energy consumption of this algorithm are significantly reduced, and the network lifetime is greatly improved.

[0019] Preferably, a method for controlling the working state of wireless sensor network nodes is proposed, and a working mechanism for node working and sleeping states is designed, that is, while meeting the requirements of network monitoring coverage rate, as few nodes as possible are in the working state, which can balance the energy of network nodes and make the working nodes more evenly deployed in the monitoring area.

[0020] Preferably, during the node state determination process, the state of internal nodes in the grid is determined first, and then the state of edge nodes is determined, which can accelerate the node working state determination speed on the premise of meeting the minimum monitoring coverage rate.

[0021] Preferably, the improved zebra optimization algorithm adds a small-range depth search strategy, enabling the population to explore deeper areas; the outlying isolated individuals adopt the extreme reverse defense evolution strategy, expanding the sample search space; the pioneer zebras adopt the perturbation evolution method, which is more conducive to the population jumping out of local extreme points, so that IZOA still has a large value in the later stage of the algorithm, thus making the convergence accuracy of IZOA higher. Description of the Drawings

[0022] Figure 1 It is a flowchart of a method for obtaining the best multi-hop routing path scheme with multiple cluster heads of the present invention; Figure 2 It is a schematic diagram of the coverage of the node sensing area; Figure 3 It is a flowchart of the implementation of the node working state control algorithm; Figure 4 It is a schematic diagram of the individual coding of DZOA; Figure 5 a is a schematic diagram of the clustering result of the IZOA-DL network in a certain round; Figure 5 b is a schematic diagram of the data transmission path planning result between cluster heads; Figure 6 It is a curve graph of the change of the number of network nodes and the network survival period; Figure 7 It is a curve graph of the number of surviving nodes changing with the number of working rounds; Figure 8 It is a curve graph of the change of the remaining energy of the network and the number of working rounds; Figure 9 It is a curve graph of the change of the data volume received by the sink node and the number of working rounds. Detailed Embodiments

[0023] To enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0024] It should be noted that similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0025] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0026] The present invention will be further described in detail below with reference to the accompanying drawings: As Figure 1 shown, a method for obtaining a multi-cluster head optimal multi-hop routing path scheme includes the following steps: Step 1, analyze the three-dimensional space coverage redundancy of nodes in the wireless sensor network by using a node working state control algorithm to determine the working nodes in the wireless sensor network; Specifically, the steps for determining the working nodes in the wireless sensor network are as follows: Uniformly divide the three-dimensional space network of the wireless sensor to determine the candidate set of working nodes and the set of network edge nodes; First, determine the working state of the nodes in the candidate set of working nodes, and then determine the working state of the nodes in the set of network edge nodes; The steps for determining the working state of the nodes in the candidate set of working nodes are as follows: Compare the absolute node coverage probability of the node with the largest node density in each grid in the candidate set of working nodes with the lowest coverage rate of the given monitoring area; If the absolute node coverage probability of the node is less than the minimum coverage rate of the given monitoring area, set the node status to "working"; if the absolute node coverage probability of the node is greater than or equal to the minimum coverage rate of the given monitoring area, set the node status to "sleeping". After the determination is completed, update the node status in the grid, and then determine the node with the largest density in the new grid. The determination method of the working status of the nodes in the network edge node set is the same as that of the nodes in the working node candidate set. The specific steps are as follows: Calculate the absolute coverage rate of the edge node with the largest node density and compare it with the minimum coverage rate of the given monitoring area. If the absolute node coverage probability of the node is less than the minimum coverage rate of the given monitoring area, set the node status to "working"; if the absolute node coverage probability of the node is greater than or equal to the minimum coverage rate of the given monitoring area, set the node status to "sleeping". After the determination is completed, update the node status, and then determine the new node with the largest density until the probability formed by the current network edge point set is less than the minimum coverage rate of the given monitoring area, and end the entire determination process.

[0027] The determination condition for the switching of the working status of the nodes in the working node candidate set and the network edge node set is: The grid coverage rate of the grid where the new node with the largest density is located after the update is less than the minimum coverage rate of the given monitoring area.

[0028] Step 2: Perform spatial clustering analysis on the working nodes in the wireless sensor network through the adaptive density peak clustering algorithm to achieve the adaptive dynamic clustering of the wireless sensor network. The adaptive density peak clustering algorithm is an improvement based on the density peak clustering algorithm, including adaptively adjusting the truncation distance parameter according to the remaining energy of the network nodes and using the covariance distance to replace the traditional distance metric. Step 3: For the clustered wireless sensor network, establish a clustered data fusion model based on deep learning, and use the improved zebra optimization algorithm to optimize the configuration of the parameters of the clustered data fusion model to fuse the data of the working nodes in the cluster. Specifically, the clustered data fusion model is constructed based on the BP neural network. After the construction of the BP neural network model, the Sink node trains the model parameters and feeds back the training results to the cluster head. The cluster head uses the trained BP neural network to extract the data features of the nodes in the cluster and transmits the fused data to the Sink node through the multi-hop routing path. The cluster head is the center of each cluster after adaptive dynamic clustering through the adaptive density peak clustering algorithm.

[0029] The improved zebra optimization algorithm adds a small - range depth search strategy, an extreme reverse defense evolution strategy for peripheral isolated individuals, and a perturbation evolution method for pioneer zebras based on the zebra optimization algorithm; Assign the parameters of the clustered data fusion model to the zebra individuals of the improved zebra optimization algorithm and define the objective function to complete the parameter optimization.

[0030] Step four, based on the working - node data within the fused clusters, establish a multi - cluster - head routing communication model; The multi - cluster - head routing communication model is:

[0031] In the formula, represents the distance between and represents the distance between and represents the distance between and In it, c is the sequence label of among all cluster - heads, and k is

[0032] the sequence number of

[0033] in all multi - hop cluster - heads.

[0034] In an embodiment of the present invention, a method for obtaining a multi - cluster - head optimal multi - hop routing path scheme is provided. Taking earthquake disaster monitoring and early warning as an application scenario, in the monitored area, sensor nodes capable of sensing low - frequency vibrations are densely deployed. Drawing on the idea of the "round" in the LEACH protocol, the "round" of the data dynamic fusion algorithm proposed in this paper mainly includes three stages: data acquisition stage, data pre - processing, and data routing and transmission. For the convenience of problem description, it is assumed that N sensor nodes are randomly and densely deployed in a three - dimensional cube monitoring area with side length M. Each sensor node is fixed in position, the position information is known, and it has the same communication and storage and computing capabilities. Outside the monitoring area, several Sink nodes (converging nodes) are deployed. The Sink nodes have sufficient energy and can achieve interactive communication with the remote earthquake disaster monitoring system through relay devices such as unmanned aerial vehicles. The steps are as follows: 1.1 Data acquisition stage 1.1.1 Node working - state control Randomly and densely deploying a large number of sensor nodes within the monitoring area can effectively extend the network's survival period. However, if all nodes are in the working state, it will seriously waste node resources. Therefore, a working and sleeping state mechanism for nodes is designed, that is, while meeting the requirements of network monitoring coverage rate, as few nodes as possible are in the working state. For sensor node (1 ≤ i ≤ N), all nodes within its communication radius r form its sensing area .

[0035] (1) In the formula, is the distance between . There is an overlap between the sensing ranges of and the nodes within it. Therefore, a node working state control algorithm can be given according to the proportion of the covered sensing area

[0036] Definition 1: Relative node coverage probability Select node , and define the relative node coverage probability as the probability that any point within the sensing range is covered and sensed by .

[0037] Definition 2: Absolute node coverage probability Assume that all nodes in the network are in the working state, and define the proportion of the overlapping sensing range of all nodes within to its own sensing range as the absolute node coverage probability of .

[0038] (2) In the formula, is 's sensing range, is the overlapping covered sensing range

[0039] The following gives , 's derivation calculation formula. As Figure 2 shown, in the three-dimensional monitoring space, the node sensing range is a sphere with radius r, the sensing range intersects with 's sensing range. Since the radii of the two spheres are the same, there is (3) According to the spherical segment volume calculation formula, we can get Covered volume Overlapped and covered volume : (4) Then The relative node coverage probability is: (5) The probability that any point within the sensing range is not sensed Therefore, The absolute node coverage probability is: is: (6) Given the minimum coverage rate of the monitoring area , if it satisfies , then the node status can be set to "sleep". To further balance the network energy consumption, select nodes with higher remaining energy to form a working node candidate set, and assume that the size of the working node candidate set is: (7) In the formula, is the volume of the monitoring area, is the control coefficient.

[0040] Definition 3. Node density: Define The node density as the number of nodes within it .

[0041] To accelerate the determination of the working status of network nodes, the three-dimensional cube monitoring area is uniformly divided into grids, each grid with a side length of L, and define nodes whose distance to the grid edge is less than r as edge nodes. Combining "Definition 1", "Definition 2", and "Definition 3", a node working status control algorithm is proposed, which is completed by the Sink node, and the specific implementation process is as Figure 3 shown.

[0042] It can be seen from the implementation process of the node working status control algorithm that this algorithm can balance the energy of network nodes and make the deployment of working nodes more uniform within the monitoring area. After the determination of the node working status, there are working nodes in the network (t represents the round), and each node sends status information such as its location and remaining energy.

[0043] 1.1.2 Adaptive clustering The Density Peak Clustering algorithm (DPC) is a new clustering analysis technique with good adaptability to any data shape. Given data points , DPC gives two definitions: local density and relative distance .

[0044] , (8) In the formula, is the Euclidean distance between and , and is the cut-off distance. DPC believes that data points with high and are cluster centers, and the remaining data points are divided into the cluster where the nearest cluster center is located.

[0045] is the only parameter that needs to be set in DPC. To better apply it to the problem of WSNs node clustering, the Adaptive Density Peak Clustering algorithm (ADPC) is proposed. ADPC adaptively adjusts according to the remaining energy of network nodes, and uses the covariance distance metric to replace the Euclidean distance metric: (9) (10) In the formula, is the proportionality coefficient, represents the remaining energy of node , and ( , , ) is the spatial coordinate of node . It can be seen from formula (9) that is closely related to the remaining energy of the network. The greater the remaining energy of network nodes, the greater is, and the fewer the number of clusters obtained, which is conducive to reducing the communication energy consumption between clusters; the smaller the remaining energy of network nodes, the smaller is, and the greater the number of clusters obtained, and the fewer the number of nodes in the cluster, which is conducive to reducing the communication energy consumption within the cluster. Using ADPC for spatial clustering analysis of WSNs nodes, c clusters are obtained, and the cluster centers are the cluster heads of the network clustering.

[0046] 1.2 Data preprocessing stage 1.2.1 Construction of the clustered data fusion BP neural network model After the network completes clustering, the data of the nodes within the cluster is processed at the cluster head. Since the nodes within the cluster are in the same monitoring area, the similarity and redundancy of the data between the nodes are relatively high. If the data of the nodes within the cluster is directly forwarded to the Sink node, it will consume a large amount of energy and shorten the node life. Therefore, a data fusion algorithm within the cluster based on deep learning is designed. The cluster head uses a BP neural network to extract the characteristic values of the data collected by the nodes within the cluster, effectively improving the data processing efficiency and reducing the node energy consumption.

[0047] The BP neural network consists of an input layer, a hidden layer, and an output layer, and has strong self-learning and non-linear mapping capabilities. The clustering of WSNs can be equivalent to a complex nervous system. The nodes within the cluster are the input layer neurons, and the hidden layer and output layer are located at the cluster head. After the BP neural network model is constructed, the Sink node trains the model parameters and feeds back the training results to the cluster head. The cluster head uses the trained BP neural network to extract the data characteristics of the nodes within the cluster, and transmits the fused data to the Sink node through a multi-hop routing path. For the c-th cluster (1 ≤ c ≤ C), there are nodes within the cluster, and the data collected by the j-th node (1 ≤ j ≤ ) is , then the input data of the input layer is . The hidden layer and output layer respectively use the positive and negative symmetric Sigmoid function ( function), the non-negative Sigmoid function ( function) and the weight matrix to perform high-dimensional mapping processing on the data.

[0048] (11) In the formula, is the weight matrix of the hidden layer, is the number of hidden layer neurons; is the weight matrix of the output layer, is the number of output neurons.

[0049] 1.2.2 Optimize the parameters of the BP neural network model with the improved zebra optimization algorithm The BP neural network is sensitive to the configuration of model parameters. The steepest descent method is a classic parameter configuration method, which has the defect of being easily trapped in the local optimum. Therefore, an improved zebra optimization algorithm (IZOA) is designed to find the optimal model parameter configuration. ZOA is a new type of computing technology proposed in 2022, which simulates the foraging and defense behaviors of zebra groups and assigns individuals the iterative evolution formulas in (12) and (13) (taking the minimization problem as an example).

[0050] (12) (13) In the formula, is the optimal solution of the population, f(·) represents the objective function, represents the individuals under attack threat, represents the maximum number of iterations. It can be seen from formulas (12) and (13) that zebras update their position information in two ways and retain excellent solutions. This update and evolution method accelerates the convergence speed of the algorithm to a certain extent, but it is prone to "premature convergence", and the method of only retaining excellent solutions in each update cannot guarantee that the algorithm converges to the global optimal solution.

[21] Therefore, an improved zebra optimization algorithm (IZOA) is proposed, and three update methods are defined respectively: small-range depth search, defense evolution of peripheral isolated individuals, and perturbation evolution of pioneer zebras.

[0051] Definition 4: Small-range depth search: Use DPC to perform cluster analysis on the zebra population to obtain categories and zebra individuals isolated from each category. Within each category, zebra individuals learn from the optimal individual in the category according to formula (14), and the optimal individual in the category learns from the optimal solution of the population (pioneer zebras) according to formula (15). After each category performs times of small-range depth search, all individuals are remixed and an iterative evolution is performed according to formula (12).

[0052] (14) (15) In the formula, and are the individual and the optimal solution of the c-th (1 ≤ c ≤ ) category respectively.

[0053] Definition 5: Defense evolution of peripheral isolated individuals: Peripheral isolated individuals are outside the population and are more vulnerable to attack. Therefore, the peripheral individuals are updated according to formula (16).

[0054] (16) In the formula, is the j-th ( ) isolated individual.

[0055] Definition 6: Perturbation evolution of pioneer zebras: For the pioneer zebra , the perturbation evolution method is defined as: (17) (18) In the formula, is a constant, , are 's maximum and minimum values. In order to analyze the IZOA population diversity, the evaluation index is defined as: (19) In the formula, Q is the population size, is the dimension of the individual . Let , and calculate the mathematical expectation of according to formula (19):

[0056] (20) Let , then there is: (21) In the formula, is 's diversity. It can be seen from formula (13) that IZOA is related to . From formulas (14)-(18), it can be known that IZOA adds a small-range depth search strategy, enabling the population to explore deeper areas; the peripheral isolated individuals adopt the extreme reverse defense evolution strategy, expanding the sample search space; the pioneer zebras adopt the perturbation evolution method, which is more conducive to the population jumping out of local extreme points, making IZOA still have a large in the later stage of the algorithm, so that IZOA has a higher convergence accuracy.

[0057] For the c sub-clusters of the WSNs network, IZOA is used to optimize the parameters of the BP neural network model. The model parameters , , , to be configured are assigned to the zebra individuals, and the individual coding x and the objective function f(x) are defined respectively as: (22) (23) In the formula, , are the actual output and the expected output of the BP neural network corresponding to the c-th sub-cluster respectively. The computational complexity of IZOA optimizing the parameters of the BP neural network model is: the computational complexity of population initialization is O(Q×a×b), the computational complexity of small-range depth search is ×( ), and the computational complexity of the peripheral isolated individual defense evolution is , the evolutionary computation complexity of the pioneer zebra perturbation is . Therefore, the total computation complexity of IZOA is

[0058] (24) 1.3 Data Routing and Transmission Phase In the data transmission phase, the cluster heads farther away from the Sink node adopt multi-hop communication to achieve data transmission. The communication distance of the node with the lowest energy consumption , for sub-clusters, if the cluster head of sub-cluster c ( ) is less than from the Sink node, then the sub-cluster data is directly transmitted to the Sink node, otherwise this cluster head is defined as a multi-hop cluster head (c is the sequence label among all cluster heads, and k is the sequence number among all multi-hop cluster heads). Let the communication path with the Sink node be , where is the ( )-th relay node in the multi-hop communication path, and . Establish the minimum multi-hop routing transmission distance model shown in Equation (25): (25) In the formula, represents the distance between . Solving Equation (25) belongs to an NP-hard problem. In this paper, the Dispersed Zebra Optimization Algorithm (DZOA) is used for solving. Assign the communication paths of cluster heads greater than from the Sink node to zebra individuals, and define the DZOA individual coding as shown in Equation (26): (26) Figure 4 gives an example of DZOA individual coding, which means there are 6 sub-clusters in the network, and 3 cluster heads are greater than from the Sink node, which are , , (the corresponding sequence numbers among all multi-hop cluster heads are , , ). The three cluster head communication paths are respectively , , .

[0059] According to the individual coding of DZOA in formula (26), the learning and evolution mechanism between individuals is redefined. Taking , as an example, it is defined that is randomly selected non-zero codes in each row of the coding to replace corresponding codes. Therefore, combining the evolutionary strategies in formulas (14) - (17) corresponding to ZOA, the individual evolutionary strategy of DZOA is given: Corresponding codes. Therefore, combining the evolutionary strategies in formulas (14) - (17) corresponding to ZOA, the individual evolutionary strategy of DZOA is given: (27) (28) In the formula, , are the maximum and minimum values, and is the proportionality coefficient. It can be seen from formulas (27) and (28) that in the initial stage of the algorithm, , , a small number of coding bits in the individuals are respectively replaced by , , ( ), which is conducive to the in-depth search of the algorithm; in the later stage of the algorithm, a large number of coding bits in the individuals are replaced, which is conducive to the accelerated convergence of the algorithm. DZOA finally obtains a multi-hop routing communication scheme for data between clusters through iterative evolution.

[0060] 2 Experimental verification 2.1 Numerical experiments of IZOA, BP neural network optimized by IZOA Table 1 Test functions

[0061] To verify the performance of the proposed IZOA, five test functions shown in Table 1 are selected for numerical experiments. The relevant parameters of IZOA are set as: , , , , . ZOA, improved honey badger algorithm (IHBA), improved PSO (IPSO), fireworks algorithm (FWA), and coati optimization algorithm (COA) are selected for comparative experiments. Each group of experiments is carried out 20 times, and the evaluation indexes are the maximum value , the minimum value , and the average value . Table 2 gives the numerical experiment results of the test functions.

[0062] Table 2 Optimization results of six algorithms for five test functions

[0063] As can be seen from Table 2, for two-dimensional and multi-modal functions 、 , the maximum, minimum, and average values of IZOA, IHBA, and IPSO are all the theoretical optimal solution "1", which indicates that these three algorithms found the global optimal solution in each experiment, while only some experiments of ZOA, FWA, and COA found the global optimal solution. For high-dimensional, multi-modal, and large number of local extreme value functions 、 、 , the convergence accuracy of IZOA, IHBA, and IPSO is significantly better than that of the other three algorithms, and the convergence accuracy of IZOA is higher. In particular, for functions 、 , the minimum experimental value of IZOA can reach the level of 10 -19 、10 -11 , while the minimum experimental values of IHBA and IPSO can only reach the levels of 10 -16 、10 -8 and 10 -15 、10 -7 . The numerical experimental results of the test functions show that compared with the other five algorithms, IZOA maintains the diversity of the population samples by designing a small-range depth search, peripheral isolated individual defense evolution, and pioneer zebra perturbation evolution mechanisms, so that the algorithm has higher global convergence accuracy and better optimization results.

[0064] To verify the generalization performance of the proposed BP neural network optimized by IZOA (IZOA-BP), five regression datasets in UCI are selected. Each dataset is divided into a training set and a test set according to a ratio of 7:3, and the five datasets are respectively predicted and analyzed by using a BP neural network and a BP neural network optimized by an improved particle swarm algorithm (IPSO-BP). The evaluation indexes are set as root mean square error ( ), mean absolute error ( ), and coefficient of determination ( ). Among them, the smaller the value, the more accurate the prediction result, the smaller the value, the better the stability of the model, the closer to 1, the better the fitting degree of the model. Table 3 shows the comparison results of the prediction evaluation indexes of the three models for the five datasets.

[0065] Table 3 Comparison results of prediction evaluation indexes of three models for five datasets

[0066] As can be seen from Table 3, for the 5 data sets, the , , three indicators of IZOA-BP are better than those of IPSO-BP and significantly better than those of BP. This shows that optimizing and configuring with IZOA parameters can effectively improve the generalization ability of the model and make the performance of IZOA-BP better.

[0067] 2.2 WSNs Earthquake Disaster Data Fusion Experiment Build an NS-2 WSNs simulation platform and import the topographic map of the area to be monitored to verify the performance of the proposed 3D WSNs data fusion algorithm (IZOA-DL). In a 3D monitoring space with a side length of , randomly and densely deploy 1200 sensor nodes, and place 1 Sink node at the coordinate . Use IZOA-DL to collect, fuse, and transmit network data. Figure 5 Figure a in Figure 5 shows the network clustering result in a certain round, and

[0068] Figure b in Figure 5 shows the path planning result of data communication between cluster heads among clusters.

[0069] 2.2.1 Network Lifetime The energy of network nodes is limited. As the number of working rounds increases, the number of "dead nodes" increases continuously, thus affecting the monitoring performance of the entire WSNs network. When the energy of all nodes in the entire network is exhausted, the network reaches its maximum life cycle. Figure 6 is the curve of the number of network nodes and the network lifetime change for IZOA-DL, IFCRA, POFCA, and IFWABP algorithms, Figure 6 is the curve graph of the number of surviving nodes of the four algorithms changing with the increase of working rounds (the initial number of deployed nodes is 1200).

[0070] From Figure 6It can be seen that as the number of network deployment nodes increases, the network lifetime also increases. However, when the number of network nodes reaches about 800, the network lifetimes of the IFCRA, POFCA, and IFWABP algorithms do not change much with the increase in the number of nodes. In particular, when the number of network nodes exceeds 1200, the network lifetimes of the three algorithms decrease with the increase in the number of nodes, while the network lifetime of the IZOA-DL algorithm increases with the increase in the number of nodes. From Figure 7 It can be seen that for WSNs initially deployed with 1200 nodes, node deaths in the IFCRA, POFCA, and IFWABP algorithms start at the 792nd, 854th, and 919th rounds respectively, while node deaths in the IZOA-DL algorithm start at the 1574th round. When the working rounds reach the 1042nd, 1182nd, and 1243rd rounds, all nodes in the POFCA, IFCRA, and IFWABP algorithms die, while all nodes in the IZOA-DL algorithm die at the 2115th round. The network lifetime is increased by 102.98%, 78.93%, and 70.15% compared to the POFCA, IFCRA, and IFWABP algorithms respectively, and the decline rate of the curve of the number of surviving nodes in the IZOA-DL algorithm is significantly smaller than that of the other three algorithms. This is because, for the IFCRA, POFCA, and IFWABP algorithms, the dense deployment of working nodes leads to a sharp increase in network data traffic, which exacerbates the energy consumption of nodes. The IZOA-DL algorithm adopts a node working state control strategy, which minimizes the number of working nodes while ensuring the requirements of monitoring coverage, enabling nodes to switch reasonably between the "working" and "sleeping" states, which is more conducive to extending the network lifetime.

[0071] 2.2.2 Network remaining energy and network energy consumption balance Combined with the change curves of the number of network nodes and the network lifetime, assuming an initial deployment of 800 nodes, Figure 8 The change curves of the network remaining energy and the working rounds of the IZOA-DL, IFCRA, POFCA, and IFWABP algorithms are given. From Figure 8It can be seen that under the same round, the remaining energy of the IZOA-DL algorithm network is much higher than that of the other three algorithms, and the decline rate of the remaining energy of the IZOA-DL algorithm network is the slowest. To further analyze the balance of network energy consumption, the IZOA-DL, IFCRA, POFCA, and IFWABP algorithms were independently run 20 times. Assuming that the network fails when 75% of the nodes in the network die, Table 4 gives the comparison results of the maximum, minimum, average, and standard deviation of the total network energy consumption of the four algorithms. It can be seen from Table 4 that all the evaluation indicators of the IZOA-DL algorithm are smaller than those of the other three algorithms, indicating that the IZOA-DL algorithm has more balanced energy consumption. This is because the IZOA-DL algorithm introduces an adaptive clustering strategy, and the network adaptively adjusts the cluster size according to the remaining energy, which is beneficial to balancing the energy consumption of intra-cluster communication and inter-cluster communication. Moreover, the IZOA-DL algorithm uses DZOA to optimize and solve the optimal path for inter-cluster communication, further reducing the network energy consumption.

[0072] Table 4 Comparison of Evaluation Indicators of Total Network Energy Consumption

[0073] 2.2.3 Amount of Data Received by Sink Node The data collected by nodes within a local range of WSNs has a certain degree of redundancy. As the number of working rounds increases, the total amount of network data communication increases, and the redundant data also increases accordingly. Figure 9 The curve of the total number of data packets received by the sink node versus the number of working rounds is given. From Figure 9 it can be seen that under the same round, the total number of data packets of the IZOA-DL algorithm is less than that of the IFWABP algorithm, and significantly less than that of the IFCRA and POFCA algorithms. This is because the IZOA-DL algorithm uses the BP neural network optimized by IZOA to perform deep learning on the data collected by nodes, efficiently extracts data features, effectively filters the redundant data between node data, and also well alleviates data communication congestion and reduces the probability of data retransmission, thereby further reducing the total amount of network data.

[0074] Focusing on the problem of 3D WSNs seismic disaster data fusion, the present invention proposes a dynamic fusion algorithm for 3D WSNs seismic disaster data that combines and improves the zebra optimization algorithm and deep learning. The main work lies in proposing a node working state control strategy, designing an adaptive density peak clustering algorithm, improving the zebra optimization algorithm and applying it to the optimization of BP neural network parameters and multi-hop routing paths. The experimental results show that the proposed algorithm has more uniform clustering, lower network energy consumption and total network data volume, and effectively extends the network survival period. Next, research can be carried out on the data collection of cluster head nodes by mobile sink nodes such as unmanned aerial vehicles to further reduce network energy consumption.

[0075] The present invention also provides a system for obtaining a multi-cluster-head optimal multi-hop routing path scheme, including a node determination module, a clustering module, a fusion module, a model construction module, and a solution module, wherein: The node determination module: is used to analyze the three-dimensional space coverage redundancy of nodes in a wireless sensor network by using a node working state control algorithm, and determine the working nodes in the wireless sensor network; The clustering module: is used to perform spatial clustering analysis on the working nodes in the wireless sensor network through an adaptive density peak clustering algorithm, and realize the adaptive dynamic clustering of the wireless sensor network; The fusion module: for the clustered wireless sensor network, establish a clustering data fusion model based on deep learning, and use an improved zebra optimization algorithm to optimize the configuration of the clustering data fusion model parameters, and fuse the data of the working nodes within the cluster; The model construction module: based on the data of the working nodes within the fused cluster, establish a multi-cluster-head routing communication model; The solution module: solve the multi-cluster-head routing communication model to obtain the multi-cluster-head optimal multi-hop routing path.

[0076] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0077] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for realizing the functions specified in Figure 1 one or more flows and / or Figure 1 blocks or multiple blocks.

[0078] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device realizes the functions in the process Figure 1One or more processes and / or blocks Figure 1 The functions specified in one or more blocks.

[0079] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one Figure 1 One or more processes and / or blocks Figure 1 The steps of the functions specified in one or more blocks.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than limit the scope of its protection. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: after reading the present invention, those skilled in the art can still make various changes, modifications or equivalent replacements to the specific implementation manners of the invention, but these changes, modifications or equivalent replacements are all within the scope of the claims of the invention pending approval.

Claims

1. A method for obtaining a multi-cluster head optimal multi-hop routing path scheme, characterized in that The steps include: Analyze the three-dimensional space coverage redundancy of nodes in the wireless sensor network using the node working state control algorithm to determine the working nodes in the wireless sensor network; Conduct spatial clustering analysis on the working nodes in the wireless sensor network through the adaptive density peak clustering algorithm to achieve adaptive dynamic clustering of the wireless sensor network; For the clustered wireless sensor network, establish a clustered data fusion model based on deep learning, and use the improved zebra optimization algorithm to optimize the configuration of the parameters of the clustered data fusion model to fuse the data of the working nodes within the cluster; Based on the data of the working nodes within the cluster after fusion, establish a multi-cluster head routing communication model; Solve the multi-cluster head routing communication model to obtain the optimal multi-hop routing path of the multi-cluster head.

2. The method for obtaining a multi-cluster head optimal multi-hop routing path scheme according to claim 1, characterized in that The determination steps of the working nodes in the wireless sensor network are as follows: Uniformly divide the three-dimensional space network of the wireless sensor to determine the candidate set of working nodes and the set of network edge nodes; First, determine the working state of the nodes in the candidate set of working nodes, and then determine the working state of the nodes in the set of network edge nodes; The steps for determining the working state of the nodes in the candidate set of working nodes are as follows: Compare the absolute node coverage probability of the node with the maximum node density in each grid in the candidate set of working nodes with the lowest coverage rate of the given monitoring area; If the absolute node coverage probability of the node is less than the lowest coverage rate of the given monitoring area, set the state of the node to "working", and if the absolute node coverage probability of the node is greater than or equal to the lowest coverage rate of the given monitoring area, set the state of the node to "sleep"; After the determination is completed, update the node state in the grid, and then determine the node with the maximum density in the new grid; The determination method of the working state of the nodes in the set of network edge nodes is the same as that of the nodes in the candidate set of working nodes.

3. The method for obtaining a multi-cluster-head optimal multi-hop routing path scheme according to claim 2, characterized in that The determination condition for switching the working state determination of the nodes in the candidate set of working nodes and the set of network edge nodes is: The grid coverage rate of the grid where the new node with the maximum density is located after updating is less than the lowest coverage rate of the given monitoring area.

4. A method for obtaining a multi-cluster-head optimal multi-hop routing path scheme according to claim 1, characterized in that, The adaptive density peak clustering algorithm is an improvement based on the density peak clustering algorithm, including adaptively adjusting the truncation distance parameter according to the remaining energy of the network nodes and using the covariance distance to replace the traditional distance metric; The expression of the truncation distance is: In the formula, is the cut-off distance, λ is the proportionality coefficient, is representing the remaining energy of the node.

5. The method for obtaining a multi-cluster-head optimal multi-hop routing path scheme according to claim 1, wherein The clustered data fusion model is constructed based on the BP neural network. After the construction of the BP neural network model, the Sink node trains the model parameters and feeds back the training results to the cluster head. The cluster head uses the trained BP neural network to extract the data features of the nodes within the cluster and transmits the fused data to the Sink node through the multi-hop routing path; The cluster head is the center of each cluster after adaptive dynamic clustering through the adaptive density peak clustering algorithm.

6. The method for obtaining a multi-cluster-head optimal multi-hop routing path scheme according to claim 1, wherein The improved zebra optimization algorithm adds a small-range deep search strategy, an extreme reverse defense evolution strategy for peripheral isolated individuals, and a perturbation evolution method for pioneer zebras on the basis of the zebra optimization algorithm; Assign the parameters of the clustered data fusion model to the zebra individuals of the improved zebra optimization algorithm and define the objective function to complete the parameter optimization.

7. The method for obtaining a multi-cluster-head optimal multi-hop routing path scheme according to claim 1, wherein The multi-cluster head routing communication model is as follows: In the formula, represents the distance between and represents the distance between and represents the distance between and In , c is the sequence number among all cluster heads, and k is the sequence number among all multi-hop cluster heads; The multi-cluster head routing communication model is solved by the discrete zebra optimization algorithm.

8. A system for obtaining a multi-cluster head optimal multi-hop routing path scheme, characterized in that, It includes a node determination module, a clustering module, a fusion module, a model construction module, and a solution module, where: Node determination module: It is used to analyze the three-dimensional space coverage redundancy of nodes in the wireless sensor network by using the node working state control algorithm, and determine the working nodes in the wireless sensor network; Clustering module: It is used to perform spatial clustering analysis on the working nodes in the wireless sensor network through the adaptive density peak clustering algorithm to achieve the adaptive dynamic clustering of the wireless sensor network; Fusion module: For the clustered wireless sensor network, a clustered data fusion model based on deep learning is established, and the improved zebra optimization algorithm is used to optimize the configuration of the parameters of the clustered data fusion model to fuse the data of the working nodes within the cluster; Model construction module: Based on the fused data of the working nodes within the cluster, a multi-cluster head routing communication model is established; Solution module: Solve the multi-cluster head routing communication model to obtain the best multi-hop routing path of the multi-cluster head.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1-7.

Images