A spider web-like hierarchical clustering network parameter optimization method

Abstract

The present invention provides a method for optimizing network parameters of a spider web-like hierarchical clustering network, and belongs to the technical field of network parameter optimization. Since the anti-destruction capability of the spider web-like hierarchical clustering network model is affected by the network deployment parameters, it cannot be reasonably optimized. Therefore, the present invention uses a statistical method to analyze the network evolution process caused by changes in network parameters, obtains the degree of influence of different network parameters and different values ​​of network parameters on network performance, and uses the degree of influence as the basis for network model parameter optimization calculation, uses network anti-destruction and the average residual energy of the entire network as optimization objective functions, and uses a multi-objective optimization algorithm to optimize network parameters, thereby providing an effective method for reasonably optimizing the network parameters of the spider web-like hierarchical clustering network.

Description

Technical Field

[0001] The invention relates to a spider web-like hierarchical clustering network parameter optimization method, belonging to the technical field of network parameter optimization, and in particular to a method for optimizing network performance through different network parameters and different values ​​of the parameters. Background Art

[0002] As the monitoring area increases, the problem of wireless sensor network technology's anti-destruction ability becomes more prominent. The artificial spider web topology based on bionics is very similar to the wireless sensor network topology. The bio-inspired spider web hierarchical clustered network model has extremely high network anti-destruction ability. However, whether the network can work long-term and stably is affected by the network deployment parameters, and the energy consumption balance within and between clusters is affected by the division of clusters.

[0003] At present, there is no systematic method for optimizing the network deployment parameters of the biomimetic spider web topology, which are generally set manually based on experience. In order to improve the network's anti-destruction ability and extend the network's life, researchers have optimized the network in three main aspects: optimizing the clustering process, optimizing cluster head election, and optimizing path selection. However, the current optimization schemes have failed to reasonably optimize the network deployment parameters. Summary of the invention

[0004] The purpose of the present invention is to provide a method for optimizing network parameters of a spider-web-like hierarchical clustering network, so as to solve the problem that the network parameters of the spider-web-like hierarchical clustering network are difficult to be reasonably optimized.

[0005] In order to achieve the above object, the present invention provides a spider web-like hierarchical clustering network parameter optimization method, comprising the following steps:

[0006] 1) The spider web-like hierarchical clustering network is simulated under different network parameter combinations to obtain the data of the whole network's invulnerability and the average residual energy of the whole network; the network parameters of the spider web-like hierarchical clustering network include the number of network layers L, the number of cluster head replacement rounds m, the intra-cluster hierarchical radius division level Gcr, the number of sector areas z and the number of sensor nodes n; the intra-cluster hierarchical radius division level Gcr is used to characterize the intra-cluster hierarchical radius, and each cluster is centered on the cluster head and is divided according to the intra-cluster hierarchical radius;

[0007] 2) Perform normality test and variance homogeneity test on the network invulnerability data and the network average residual energy data. If the network invulnerability data and the network average residual energy data obey the normal distribution and satisfy the variance homogeneity, the network parameters are taken as fixed factors and the network invulnerability is taken as the dependent variable to perform the inter-subject effect test. The inter-subject effect test results corresponding to the network parameters and the network invulnerability are obtained. i; The network parameters are taken as fixed factors, and the average residual energy of the whole network is taken as the dependent variable to conduct the inter-subject effect test, and the inter-subject effect test results corresponding to the network parameters and the average residual energy of the whole network are obtained. i , i=1,…,5;

[0008] 3) Normalize the network parameters and the corresponding network invulnerability data respectively, and establish a network invulnerability model f1(x) based on the normalized data; the network invulnerability model f1(x) is used to represent the corresponding relationship between the network parameters and the network invulnerability; according to the inter-subject effect test result p i Establish the constraints of the whole network invulnerability model f1(x);

[0009] The network parameters and the corresponding average residual energy data of the whole network are normalized respectively, and the average residual energy model f2(x) of the whole network is established according to the normalized data; the average residual energy model f2(x) of the whole network is used to express the corresponding relationship between the network parameters and the average residual energy of the whole network; according to the inter-subject effect test result q i Establish the constraints of the average residual energy model f2(x) for the entire network;

[0010] 4) The network-wide invulnerability data and the network-wide average residual energy data are used as the input of the genetic algorithm, and the maximum difference between f1(x) and f2(x) is used as the crowding degree of the individuals in the non-dominated layer of the genetic algorithm. The population is iterated according to the crowding degree, and the Pareto optimal frontier of the network-wide invulnerability model and the network-wide average residual energy model is obtained at the end of the iteration;

[0011] 5) The intersection of the Pareto optimal frontier and the straight line f1(x)=f2(x) is taken as the optimal solution equilibrium point, the network parameters corresponding to the optimal solution equilibrium point are calculated, and the spider web hierarchical clustering network is optimized according to the network parameters.

[0012] The present invention conducts simulation experiments on spider web-like hierarchical clustering networks under different network parameter combinations to obtain network-wide indestructibility data and network-wide average residual energy data, and performs normality test and variance homogeneity test on the network-wide indestructibility data and network-wide average residual energy data. On the basis that the network-wide indestructibility data and network-wide average residual energy data obey normal distribution and satisfy variance homogeneity, the network parameters are used as fixed factors, and the network-wide indestructibility data and network-wide average residual energy data are used as dependent variables to perform inter-subject effect detection and obtain the results of inter-subject effect detection. After normalizing the network parameter combination and the network-wide indestructibility data and network-wide average residual energy data, a network-wide indestructibility model and a network-wide average residual energy model are established, and according to the results of the inter-subject effect detection, the constraints of the network-wide indestructibility model and the network-wide average residual energy model are established. The whole network indestructibility data and the whole network average residual energy data are used as the input of the genetic algorithm. In the iterative process, the difference between the whole network indestructibility model and the whole network average residual energy model is used as the congestion degree. The population iteration is performed according to the congestion degree. At the end of the iteration, the Pareto optimal frontier of the whole network indestructibility model and the whole network average residual energy model is obtained. The intersection of the Pareto optimal frontier and the straight line f1(x)=f2(x) is used as the optimal solution equilibrium point. According to the whole network indestructibility data and the whole network average residual energy data corresponding to the optimal solution equilibrium point, new network parameters are obtained, and the spider web-like hierarchical clustering network is optimized according to the network parameters. By adopting the present invention, the spider web-like hierarchical clustering network can be optimized according to the network parameters, so that the whole network indestructibility and the whole network average residual energy of the spider web-like hierarchical clustering network are optimal.

[0013] Furthermore, in the above method, the spider-web-like hierarchical clustering network is established by the following steps:

[0014] S1, imitate the circular spider web topology structure, divide the area of ​​the circular network monitoring area, and n sensor nodes are randomly distributed in the circular network monitoring area;

[0015] S2, according to the number of network layers L and the number of sector areas z, the circular network monitoring area is divided into multiple clusters of equal area;

[0016] S3, randomly select the cluster heads for the first communication, and after each elected cluster head performs several rounds of communication, the next cluster head election is performed, and the number of cluster head elections does not exceed the number of cluster head replacement rounds m;

[0017] S4. The intra-cluster layer radius is divided into levels Gcr according to the intra-cluster layer radius, and the intra-cluster network communication is performed according to the intra-cluster layer radius.

[0018] The invention provides a specific method for establishing a spider web-like hierarchical clustering network, which is convenient for obtaining a spider web-like hierarchical clustering network for simulation experiments.

[0019] Further, in the above method, according to p i Calculate the weight a corresponding to L, m, Gcr, z and n i , then according to a i And the corresponding network parameters get the constraint condition g1(x) of the whole network invulnerability model f1(x): According to q i Calculate the weight b corresponding to L, m, Gcr, z and n i , then according to b i Obtain the constraint condition g2(x) of the average residual energy model f1(x) of the entire network;

[0020]

[0021]

[0022]

[0023]

[0024] In the formula, a i is the weight corresponding to the ith network parameter and the corresponding network invulnerability, p i is the inter-subject effect test result corresponding to the ith network parameter and the corresponding network invulnerability, b i is the weight corresponding to the ith network parameter and the corresponding residual energy of the whole network balance, q i is the inter-subject effect test result corresponding to the ith network parameter and the corresponding average residual energy of the whole network, x i is the i-th network parameter, i=1, 2,…, 5.

[0025] It provides a simple method to establish the constraints of the model and reduce the amount of calculation.

[0026] Furthermore, in the above method, the genetic algorithm described in step 4) comprises the following steps:

[0027] 4.1. Generate a parent population based on the input of the genetic algorithm, and obtain a child population through selection, crossover and mutation operations;

[0028] 4.2. Starting from the second generation, the parent population and the child population are merged, and the non-dominated levels of the individuals in the merged population are sorted, and the individuals with high non-dominated levels are selected for population iteration;

[0029] In the sorting process, whether the solution of the individual in the population is a feasible solution is determined according to the constraints g1(x) and g2(x). The non-dominated level of the individual corresponding to the feasible solution is higher than the non-dominated level of the individual corresponding to the infeasible solution. If they are both feasible solutions, the greater the crowding degree of the individual, the higher the non-dominated level. If they are both infeasible solutions, the smaller the constraint violation value of the individual, the higher the non-dominated level. The constraint violation value is used to characterize the degree to which the individual solution violates the constraint conditions.

[0030] Furthermore, in the above method, in step 2), the normality test of the network-wide indestructibility data and the network-wide average remaining energy data is performed through KS test, histogram and QQ graph. If the result of the KS test on the network-wide indestructibility data is greater than the first set value, the histogram of the network-wide indestructibility data conforms to the normal distribution, and the network-wide indestructibility data coincides with the first theoretical straight line in the QQ graph, then it is considered that the network-wide indestructibility data obeys the normal distribution; if the result of the KS test on the network-wide average remaining energy data is greater than the first set value, the histogram of the network-wide average remaining energy data conforms to the normal distribution, and the network-wide average remaining energy data coincides with the second theoretical straight line in the QQ graph, then it is considered that the network-wide average remaining energy data obeys the normal distribution; the first theoretical straight line is obtained according to the standard deviation and mean of the network-wide indestructibility data; the second theoretical straight line is obtained according to the standard deviation and mean of the network-wide average remaining energy data.

[0031] Through KS test, histogram and QQ diagram, the normality test of the whole network invulnerability data and the whole network average residual energy data is carried out with higher accuracy.

[0032] Further, in the above method, in step 2), when performing a variance homogeneity test on the whole network invulnerability data, a variance homogeneity test result based on the median is used for judgment, and if the variance homogeneity test result based on the median is greater than a second set value, it is considered that the whole network invulnerability data meets the variance homogeneity;

[0033] When performing a variance homogeneity test on the average remaining energy data of the entire network, a variance homogeneity test result based on the median is used for judgment. If the variance homogeneity test result based on the median is greater than a second set value, it is considered that the average remaining energy data of the entire network meets the variance homogeneity.

[0034] The median-based variance homogeneity test is used, and the test results are more robust.

[0035] Further, in the above method, in step 2), if the invulnerability data of the whole network does not obey the normal distribution and does not satisfy the homogeneity of variance, the invulnerability data of the whole network is logarithmically transformed, and if the invulnerability data of the whole network after the logarithmic transformation obeys the normal distribution and satisfies the homogeneity of variance, it is considered that the invulnerability data of the whole network obeys the normal distribution and satisfies the homogeneity of variance;

[0036] If the average remaining energy data of the whole network does not obey the normal distribution and does not satisfy the variance homogeneity, the average remaining energy data of the whole network is logarithmically transformed. If the average remaining energy data of the whole network after logarithmic transformation obeys the normal distribution and satisfies the variance homogeneity, it is considered that the average remaining energy data of the whole network obeys the normal distribution and satisfies the variance homogeneity.

[0037] When the network-wide invulnerability data and the network-wide average residual energy data do not obey the normal distribution and do not satisfy the variance homogeneity, the data is transformed by the logarithmic transformation method, and the normality test and variance homogeneity test of the transformed data information are performed. If the network-wide invulnerability data and the network-wide average residual energy data after logarithmic transformation obey the normal distribution and satisfy the variance homogeneity, it is considered that the network-wide invulnerability data and the network-wide average residual energy both obey the normal distribution and satisfy the variance homogeneity, thereby improving the data utilization rate.

[0038] Furthermore, in the above method, data fitting is performed on the normalized network parameters and the corresponding network-wide indestructibility data to obtain the network-wide indestructibility model f1(x); data fitting is performed on the normalized network parameters and the corresponding network-wide average residual energy data to obtain the network-wide average residual energy model f2(x).

[0039] The data fitting method is used to establish the corresponding relationship between network parameters and the whole network's invulnerability as the whole network's invulnerability model, and the corresponding relationship between network parameters and the whole network's average residual energy is established as the whole network's average residual energy model, with small calculation amount.

[0040] Furthermore, in the above method, in step 1), when the simulation experiment is performed on the spider-web-like hierarchical clustering network under different network parameter combinations, only one network parameter changes between different network parameter combinations.

[0041] Between different network parameter combinations, only one network parameter and its value is changed, which ensures that the whole network's invulnerability and the whole network's average residual energy under different network parameter combinations can be obtained while reducing the amount of calculation.

[0042] Furthermore, in the above method, in step 1), a simulation experiment is performed on the spider web hierarchical clustering network under different network parameter combinations. If the number of node deaths in the spider web hierarchical clustering network exceeds a set percentage of the number of sensor nodes n, the simulation experiment is terminated.

[0043] When conducting simulation experiments, if the number of dead nodes in the spider web hierarchical clustering network exceeds a certain proportion of the total number, it is considered that the remaining number of nodes is insufficient to represent the performance of the entire network. At this time, the simulation experiment is terminated directly to ensure the accuracy of subsequent data processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1A flowchart of constructing a spider-web-like hierarchical clustering network model in an embodiment of the method of the present invention;

[0045] Figure 2 It is a flowchart of the spider web-like hierarchical clustering network parameter optimization method in the method embodiment of the present invention;

[0046] Figure 3 A schematic diagram of different parameter combinations in an embodiment of the method of the present invention;

[0047] Figure 4 A histogram of the network-wide invulnerability data in an embodiment of the method of the present invention;

[0048] Figure 5 It is a QQ graph of the whole network invulnerability data in the embodiment of the method of the present invention;

[0049] Figure 6 A histogram of the average residual energy data of the entire network in the embodiment of the method of the present invention;

[0050] Figure 7 It is a QQ graph of the average residual energy data of the whole network in the embodiment of the method of the present invention;

[0051] Figure 8 The data model of the whole network invulnerability in the embodiment of the method of the present invention;

[0052] Fig. 9 The average residual energy data model of the whole network in the embodiment of the method of the present invention;

[0053] Fig.10 Schematic diagram of the Pareto optimal frontier and equilibrium line in an embodiment of the method of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0055] Method Example:

[0056] The present invention discloses a method for optimizing network parameters of a spider web-like hierarchical clustering network. The method uses a statistical method to analyze the network evolution process caused by changes in network parameters, obtains the degree of influence of different network model parameters and different values ​​of network parameters on network performance, and uses the degree of influence as the basis for network model parameter optimization calculation. The method uses network anti-destruction performance and the average residual energy of the entire network as optimization objective functions, and uses a multi-objective optimization algorithm to optimize network parameters, thereby solving the problem that the parameters of a spider web-like hierarchical clustering network cannot be reasonably optimized.

[0057] In order to solve the problem of optimizing the parameters of the spider web-like hierarchical clustering network, the present invention constructs a spider web-like hierarchical clustering network model with uniformly divided cluster areas. The construction process is as follows: Figure 1 As shown, the following steps are included:

[0058] S1. Network initialization: Imitate the circular spider web topology and divide the circular network monitoring area into regions. n sensor nodes are randomly distributed in the monitoring area. The larger the cluster area, the more randomly distributed nodes there are. The energy of each sensor node is consistent, all of which are 0.5J.

[0059] S2. Cluster division: Set layers according to the different distances between sensor nodes and base stations. Each layer forms L concentric rings with the base station. A set of uniform radial lines centered on the base station divides the network into z sector areas. The common area surrounded by adjacent concentric rings and a single sector area constitutes a cluster area. The L and z values ​​jointly determine the number of cluster areas in the monitoring area. The more cluster areas there are, the smaller the cluster area is and the fewer nodes there are in the cluster.

[0060] In order to obtain clusters of the same area, the network model is layered as follows:

[0061]

[0062] Where L represents the number of network layers, R1, R2, R3, …, RL represent the distances from the 1st, 2nd, 3rd, …, Lth layers to the base station respectively, and s represents the area of ​​the monitoring area.

[0063] S3, cluster head election: Randomly select the cluster heads for the first communication. After each elected cluster head performs several rounds of communication, the next cluster head election is performed. The number of cluster head elections does not exceed the set number of cluster head replacement rounds m. The network's invulnerability can directly reflect the network's service performance, and the network's average residual energy reflects the node's survivability. Taking the network's invulnerability and the network's average residual energy into consideration can more comprehensively evaluate the network performance. Therefore, starting from the second time, cluster heads are elected based on the principle of the highest network's invulnerability and the highest network's average residual energy.

[0064] In order to quantify the network-wide invulnerability and the network-wide average residual energy, the network-wide invulnerability and the network-wide average residual energy are defined as follows:

[0065] (1) Network survivability

[0066] Definition 1: For a network with N nodes, if there is u0(k min ) is of length k min The number of equivalent shortest paths between node i and node j is em ij , equivalent shortest path number em ij Calculated by the following formula:

[0067]

[0068] In the formula, u0(k min ) is the number of shortest paths between node i and node j in an N-node network, k min is the length of the shortest path, also known as the number of hops, k min ≤N-1, N is the number of network nodes, u1(k min ) is the distance between nodes in the N-node fully connected network is no greater than k min The number of paths.

[0069] Obviously, 0 ij ≤1, if and only if there is a direct edge between the two nodes, em ij = 1. For any i and j, if em ij =1, then it is a fully connected network.

[0070] Assuming that the WSN network (Wireless Sensor Networks) consists of multiple clusters, the invulnerability of each cluster is equal to the equivalent shortest path number inv of the cluster. The equivalent shortest path number inv of the cluster area is calculated by the following formula:

[0071]

[0072] In the formula, is the sum of the equivalent shortest paths in the entire network, em ij is the number of equivalent shortest paths between node i and node j, N(N-1) / 2 is the number of node pairs that can be connected in an N-node network. This function actually represents the average number of equivalent shortest paths in the cluster area network.

[0073] The average equivalent shortest path number inv of a non-fully connected network is greater than 0 and less than 1. The larger inv is, the more compact the network structure is, the higher the inv is, and the stronger the invability is. The average equivalent shortest path number inv of a fully connected network is equal to 1, and the invability is the strongest.

[0074] The WSN network is composed of several clusters, and the invulnerability of the entire WSN network can be measured by the weighted sum of the invulnerability of each cluster.

[0075] Definition 2: Assuming that the WSN network is composed of h clusters, the invulnerability INV of the WSN network is calculated by the following formula:

[0076]

[0077] In the formula, |V v | is the number of nodes in the vth cluster in the WSN network, |V| is the number of nodes in the entire WSN network, inv v is the invulnerability of the vth cluster, and h is the number of clusters in the WSN network. ​

[0078] If each cluster in a WSN network is a fully connected network and there is a path between each cluster and the sink node, the invulnerability of the WSN network is the largest and the invulnerability is 1.

[0079] (2) Average residual energy of the entire network

[0080] According to the energy consumption model, the remaining energy of all nodes in the network is calculated, and the average remaining energy of the surviving nodes is Calculated by the following formula:

[0081]

[0082] Where sur is the number of surviving nodes, E i is the remaining energy of the i-th surviving node.

[0083] S4, intra-cluster networking communication: Intra-cluster nodes communicate in a multi-hop manner. To reduce the redundancy of networking, it is necessary to set a suitable intra-cluster hierarchical radius. For this purpose, the intra-cluster hierarchical radius is divided into levels Gcr (Grade communication radius). The intra-cluster hierarchical radius is changed by changing the size of Gcr. According to the intra-cluster hierarchical radius, the cluster head is networked with the first-layer nodes, the first-layer nodes are internally networked, and the other layer nodes are networked with the first-layer nodes. After all nodes in the cluster are networked, intra-cluster communication is carried out.

[0084] The intra-cluster networking communication process is as follows:

[0085] S4.1. Network initialization: The cluster head node sends networking information to all nodes in the cluster. At the same time, it sends the location information of all nodes in the cluster to the nodes in the cluster, and sends the layer number of each node to the node.

[0086] S4.2, Cluster head networking: Based on the location of each node and the layered radius within the cluster, the cluster head node establishes communication links with the first layer nodes.

[0087] S4.3. Networking of first-layer nodes: The first-layer nodes send networking information based on the layered radius within the cluster. If the child nodes at the same layer receive this information, they select the two nearest neighboring nodes as the adjacent nodes at the same layer based on the location information and record them in the neighboring nodes of the local routing table. If the child nodes at the next layer receive this information, they regard the node sending the information as their own upper-layer communication node and record the information in the adjacent inner layer of their own local routing table.

[0088] S4.4. According to the above method, the 2nd, 3rd, 4th, ..., Lth layer nodes are networked respectively.

[0089] After the intra-cluster networking is completed, intra-cluster communication can be divided into the following two situations:

[0090] (1) After the cluster head is generated, nodes that are farther away from the cluster head will give priority to upper-layer communication nodes that are closer to the cluster head for information transmission.

[0091] (2) If the preferred node dies due to energy exhaustion, the adjacent node on the same layer is selected to establish contact and transmit information.

[0092] S5. Inter-cluster communication: The cluster head performs data fusion on the collected intra-cluster information and communicates with the base station in a multi-hop manner.

[0093] The spider web-like hierarchical clustering network parameter optimization method of the present invention is implemented in two parts. One part is to establish a network parameter and evaluation index model by statistical methods, and the other part is to optimize the network parameter and evaluation index model by using NSGA-II. Figure 2 As shown, the specific process is as follows:

[0094] 1. Statistical methods for modeling network parameters and evaluation indicators

[0095] 1) Perform a normality test on the network-wide invulnerability data and the network-wide average residual energy data, obtain the P value based on the KS (Kolmogorov–Smirnov) test, and draw the histogram and QQ graph of the network-wide invulnerability data and the network-wide average residual energy data. If P>0.05, the histogram distribution conforms to the normal distribution curve, and the actual value in the QQ graph basically coincides with the theoretical straight line, then the data obeys the normal distribution, otherwise it does not obey.

[0096] The Levene variance homogeneity test method is used to test the variance homogeneity of the network-wide invulnerability data and the network-wide average residual energy data. According to the P value obtained from the variance homogeneity test, it is judged whether the network-wide invulnerability data and the network-wide average residual energy data meet the variance homogeneity. If P>0.05, the data meets the variance homogeneity, otherwise it does not meet the variance homogeneity. The study found that the variance homogeneity test results based on the median are more robust, so the variance homogeneity test results based on the median are used to judge the variance homogeneity.

[0097] If the network-wide invulnerability data and the network-wide average remaining energy data do not obey the normal distribution and cannot meet the variance homogeneity, the data is logarithmically transformed and then the normal distribution test and variance homogeneity test are performed. If the network-wide invulnerability data and the network-wide average remaining energy data obey the normal distribution and meet the variance homogeneity, a multi-factor variance analysis is performed.

[0098] 2) Multi-factor variance analysis was performed using parameter combinations and the corresponding network-wide invulnerability and network-wide average residual energy. The number of network layers L, the number of cluster head replacement rounds m, the level of intra-cluster layered radius division Gcr, the number of sector areas z, and the number of sensor nodes n were used as fixed factors, and the network-wide invulnerability and network-wide average residual energy were used as dependent variables to conduct inter-subject effect tests. If the result of the inter-subject effect test is P<0.05, it means that the fixed factor has a significant effect on the network-wide invulnerability or the network-wide average residual energy, otherwise, it is not significant.

[0099] The inter-subject effect test was conducted with the network-wide invulnerability as the dependent variable, and the inter-subject effect test result P values ​​corresponding to the fixed factors L, m, Gcr, z, and n were obtained, which are p i (i=1,…,5); the average residual energy of the whole network is used as the dependent variable to test the inter-subject effect, and the P values ​​of the inter-subject effect test results corresponding to the fixed factors L, m, Gcr, z and n are obtained, which are q i (i=1, ..., 5). By sorting according to the size of the P value, we can intuitively see the influence of the fixed factors L, m, Gcr, z and n on the invulnerability of the whole network and the average residual energy of the whole network.

[0100] 2. NSGA-II optimizes network parameters and evaluation index models

[0101] Using MATLAB2020b, the network parameter combination, as well as the corresponding network-wide invulnerability and network-wide average residual energy are normalized, and the logistic regression algorithm is selected to construct the network parameter model. The network-wide invulnerability data and network-wide average residual energy data are modeled using the glmfit function in MATLAB, with the parameters set to "binomial", "link" and "logit" to obtain functions f1(x) and f2(x). Among them, function f1(x) corresponds to the network-wide invulnerability model, and function f2(x) corresponds to the network-wide average residual energy model. x1, x2, x3, x4 and x5 correspond to network parameters L, m, Gcr, z and n, respectively.

[0102] According to the influence of network parameters on the invulnerability of the whole network and the average residual energy of the whole network, the ranges of variables x1, x2, x3, x4 and x5 are set.

[0103] According to the multi-factor variance analysis, the p corresponding to the invulnerability of the whole network is obtained. i , and q corresponding to the average residual energy of the entire network i , according to p i and q i , calculate the weight a corresponding to x1, x2, x3, x4 and x5 i and b i, and use the weights to obtain the inequality constraints g1(x) and g2(x).

[0104]

[0105]

[0106]

[0107]

[0108] For a solution x i (i=1,2,…,5), which satisfies the constraints g1(x) and g2(x), then it is called a feasible solution; if it does not satisfy the constraints, it is called an infeasible solution. The constraint violation value is generally used to quantitatively describe a solution x. i The degree of constraint violation. The constraint violation value is defined as CV(x). For a feasible solution x i , CV(x i ) is greater than 0, for infeasible solutions, CV(x i ) is 0. The constraint violation value CV(x) is expressed by the following formula:

[0109]

[0110] Where J is the number of inequality constraints, <g j (x)> means if g j (x)≤0, then <g j (x)>=0, otherwise, <g j (x)>=|g j (x)|.

[0111] The NSGA-II (non-dominated sorting multi-objective optimization genetic algorithm) algorithm is used to optimize the network's invulnerability and the average residual energy of the network, including the following steps:

[0112] (1) The network's invulnerability and the average residual energy of the network are used as inputs to randomly generate an initial population of size P = 60. After non-dominated sorting, the first generation of offspring population is obtained through the three basic operations of genetic algorithm: selection, crossover, and mutation.

[0113] (2) Starting from the second generation, the parent population is merged with the child population, and a fast non-dominated sort is performed to construct a non-dominated solution set with different levels. The non-dominated solution sets with different levels are sorted by congestion degree, and individuals with relatively high non-dominated levels are selected to form a new parent population with a population size of 2P.

[0114] The individuals corresponding to feasible solutions have higher non-dominated levels than those corresponding to infeasible solutions. If there are multiple individuals whose solutions are all feasible solutions, the crowding degree of the corresponding individuals is calculated according to the objective function f1(x) and the objective function f2(x). The individuals with large crowding degree have higher non-dominated levels. If there are multiple individuals whose solutions are all infeasible solutions, the individuals are judged according to the constraint violation value under the constraint violation condition. The smaller the constraint violation value, the higher the non-dominated level.

[0115] (3) Through the basic operations of the genetic algorithm, a new offspring population with a population size of P = 60 is generated, and so on, until the conditions for the end of the program are met.

[0116] Using the NSGA-II algorithm, the Pareto optimal frontier of the whole network's invulnerability and average residual energy can be output after the program ends. In order to obtain the optimal balance between the whole network's invulnerability and the average residual energy, the intersection of the Pareto optimal frontier and the straight line f1(x)=f2(x) is taken as the optimal solution equilibrium point. The network parameters L′, m′, Gcr′, z′ and n′ corresponding to the optimal solution equilibrium point are calculated using the inverse normalization formula.

[0117] The present invention is further described by the following examples:

[0118] The simulation test was carried out based on the network parameters corresponding to the optimal balance point obtained by network parameter optimization. The test was implemented on PyCharm2019.3.3. The sensor nodes were randomly deployed in a circular monitoring area with a radius of 114m (equivalent to a square area with a side length of 200m), forming a spider web-like hierarchical clustering network. The initial energy of the sensor node was set to 0.5J, the data packet length was set to 4000bits, and the RF energy consumption coefficient E elec Set to 50nJ / bits, the free space channel model signal amplifier power consumption coefficient ε fs Set to 10pJ·bit -1 ·m -2 , multipath fading channel model signal amplifier power consumption coefficient ε mp Set to 0.0013pJ·bit -1 ·m -4 .

[0119] The number of network layers L of the spider web-like hierarchical clustering network is set to {3, 4, 5, 6}, the number of cluster head replacement rounds m is set to {5, 10, 15, 20}, the level of hierarchical radius division Gcr within the cluster is set to {3, 4, 5, 6}, the number of sector areas z is set to {4, 5, 6, 7, 8}, and the number of nodes n is set to {500, 600, 700, 800, 900, 1000, 1100}. Set the initial values ​​of L, m, Gcr, z and n to 3, 10, 3, 4, 1000, change the values ​​of the parameters in turn, and adjust one of the network parameters and the corresponding network parameter values ​​each time to obtain the following: Figure 3 There are 20 parameter combinations shown in the figure. The program of the spider web layered clustering network model runs three times under different parameter combinations. When the number of dead nodes reaches 50%, the program stops running. During the running process, the data of the invulnerability of the whole network and the average residual energy data of the whole network are collected.

[0120] The normality test of the network-wide invulnerability data and the network-wide average remaining energy data was performed, and according to the KS test results, the single-sample KS normality test results were obtained as shown in Table 1. The P value of the normality test of the network-wide invulnerability data was 0.054, P>0.05; the P value of the normality test of the network-wide average remaining energy data was 0.2, P>0.05.

[0121] Table 1 Single sample KS normality test results

[0122]

[0123] At the same time, a histogram and QQ graph are drawn based on the network-wide survivability data, such as Figure 4 and Figure 5 As shown in Figure 2, a histogram and a QQ graph are drawn based on the average residual energy data of the entire network, as shown in Figure 2 Figure 6 and Figure 7 As shown. Figure 4 and Figure 6 It can be seen that the network invulnerability data and the network average residual energy data approximately follow the normal distribution. The QQ graph reflects the degree of conformity between the actual distribution of the network invulnerability data and the network average residual energy data and the theoretical distribution. Figure 5 and Figure 7 It can be seen that the network invulnerability data and the network average remaining energy data are basically consistent with the theoretical straight line. Based on the above judgment results, it can be concluded that the network invulnerability data and the network average remaining energy data are both subject to normal distribution.

[0124] Variance homogeneity test: The Levene variance homogeneity test method is used to perform variance homogeneity tests on the network-wide invulnerability data and the network-wide average residual energy data, and the Levene equality test results of the error variance shown in Table 2 below are obtained.

[0125] Table 2 Results of Levene's equality test of error variance

[0126]

[0127]

[0128] It can be concluded from Table 2 that the variance homogeneity test based on the median of the whole network's invulnerability data is performed, and the test result P value is 0.979, P>0.05. Therefore, the variance homogeneity of the whole network's invulnerability data meets the variance homogeneity; the variance homogeneity test based on the median of the whole network's average residual energy data is performed, and the test result P value is 0.863, P>0.05. Therefore, the average residual energy data of the whole network meets the variance homogeneity.

[0129] On the basis that the data of the whole network invulnerability and the whole network average remaining energy obey the normal distribution and meet the variance homogeneity, a multi-factor variance analysis is conducted. The number of layers L of the WSN network, the number of clusters m in the WSN network, the level of layered radius division Gcr within the cluster, the number of sector areas z in the WSN network and the number of sensor nodes n in the monitoring area are taken as fixed factors, and the whole network invulnerability and the whole network average remaining energy are taken as dependent variables, and the inter-subject effect test is conducted.

[0130] The whole network indestructibility is used as the dependent variable to conduct the inter-subject effect test, and the inter-subject effect test table shown in Table 3 below is obtained.

[0131] Table 3. Test table of inter-subject effect of dependent variable as whole network invulnerability

[0132] source Type III Sum of Squares Degrees of Freedom Mean Square F Significant P Modified Model 0.010 19 0.001 14.925 0.000 intercept 0.000 1 0.000 4.988 0.031 L 2.025E-5 3 6.751E-6 0.201 0.895 m 4.838E-5 3 1.613E-5 0.480 0.698 Gcr 0.000 3 8.327E-5 2.478 0.075 z 0.003 4 0.001 23.279 0.000 n 0.001 6 0.000 4.879 0.001 error 0.001 40 3.361E-5 total 0.038 60 Corrected Total 0.011 59

[0133] As can be seen from Table 3, the P values ​​corresponding to L, m and Gcr are 0.895, 0.698 and 0.075, respectively, which are greater than the significance level (0.05), indicating that the different values ​​of L, m and Gcr have no significant impact on the invulnerability of the entire network. The P values ​​corresponding to z and n are 0 and 0.001, respectively, which are both less than the significance level (0.05), indicating that different values ​​of z and n have a significant impact on the invulnerability of the entire network. The order of the impact of network parameters on the invulnerability of the entire network is: z, n, Gcr, m, L.

[0134] The average residual energy of the entire network is used as the dependent variable to conduct an inter-subject effect test, and the inter-subject effect test table shown in Table 4 below is obtained.

[0135] Table 4. Test table of inter-subject effects with the dependent variable being the average residual energy of the entire network

[0136] source Type III Sum of Squares Degrees of Freedom Mean Square F Significant P Modified Model 0.019 19 0.001 6.001 0.000 intercept 0.005 1 0.005 29.138 0.000 L 0.005 3 0.002 9.329 0.000 m 0.001 3 0.000 1.315 0.283 Gcr 0.003 3 0.001 6.014 0.002 z 0.003 4 0.001 3.912 0.009 n 0.004 6 0.001 3.830 0.004 error 0.007 40 0.000 total 4.923 60 Corrected Total 0.026 59

[0137] As can be seen from Table 4, the P value corresponding to m is 0.283, which is greater than the significance level (0.05), indicating that different values ​​of m have no significant effect on the average residual energy of the entire network. The P values ​​corresponding to L, Gcr, z and n are 0.000, 0.002, 0.009 and 0.004, respectively, which are all less than the significance level (0.05), indicating that different values ​​of L, Gcr, z and n have a significant effect on the average residual energy of the entire network. The order of network parameters affecting the average residual energy of the entire network is: L, Gcr, n, z, m.

[0138] Using MATLAB2020b, the network parameter combination, as well as the corresponding network-wide indestructibility data and network-wide average residual energy data are normalized, and the logistic regression algorithm is selected to construct the network parameter model to obtain the network-wide indestructibility model and the network-wide average residual energy model.

[0139] The whole network invulnerability model is as follows Figure 8 As shown, it is represented by the function f1(x):

[0140]

[0141] z1=-0.2219-0.3436x1-0.0494x2+2.5330x3+4.3366x4-2.4000x5

[0142] The average residual energy model of the whole network is as follows: Fig. 9 As shown, it is represented by the function f2(x):

[0143]

[0144] z2=0.9185-2.4985x1-1.0584x2-1.0833x3-1.2555x4+0.5419x5

[0145] In the formula, the ranges of variables x1, x2, x3, x4 and x5 are: x1∈[0,1], x2∈[0,1], x3∈[0,1], x4∈[0,1], x5∈[0,1] respectively.

[0146] According to the multi-factor variance analysis, the p corresponding to the invulnerability of the whole network is obtained. i , respectively 0.895, 0.698, 0.075, 0 and 0.001; we also get the p corresponding to the average residual energy of the whole network i , respectively 0, 0.283, 0.002, 0.009 and 0.004. According to p i and q i , calculate the weight a corresponding to x1, x2, x3, x4 and x5 i and b i, and use the weights to obtain the inequality constraints g1(x) and g2(x).

[0147] g1(x)=0.464x1+0.582x2+0.955x3+x4+0.999x5≥0.194

[0148] g2(x)=x1+0.050x2+0.993x3+0.970x4+0.987x5≥0.017

[0149] The NSGA-II algorithm is used to optimize the network's invulnerability and the average residual energy of the network, including the following steps:

[0150] (1) The network's invulnerability and the average residual energy of the network are used as inputs to randomly generate an initial population of size P = 60. After non-dominated sorting, the first generation of offspring population is obtained through the three basic operations of genetic algorithm: selection, crossover, and mutation.

[0151] (2) Starting from the second generation, the parent population and the child population are merged and quickly non-dominated sorted. At the same time, the crowding degree of individuals in each non-dominated layer is calculated according to the function f1(x) and the function f2(x). According to the non-dominated relationship and the crowding degree of the individual, the individuals with relatively large crowding degree are selected to form a new parent population with a population size of 2P.

[0152] (3) Through the basic operations of the genetic algorithm, a new offspring population with a population size of P = 60 is generated, and so on, until the conditions for the end of the program are met.

[0153] After the program is finished, the Pareto optimal frontier of the whole network's invulnerability and the whole network's average residual energy is obtained, and the optimal solution equilibrium point is obtained according to the intersection of the Pareto optimal frontier and the straight line f1(x)=f2(x). Fig.10 As shown, the horizontal axis is the average residual energy of the whole network, the vertical axis is the invulnerability of the whole network, the solid line in the figure is the balance line of f1(x)=f2(x), and the curve is the Pareto optimal frontier. Therefore, the intersection of the balance line and the Pareto optimal frontier is taken as the optimal solution equilibrium point. The optimal solution equilibrium point is (0.0852, 0.0883), which means that when the invulnerability of the whole network is high and the average residual energy of the whole network is the largest, it also meets the Pareto optimality. At this time, the value of the invulnerability of the whole network is 0.0883, and the value of the average residual energy of the whole network is 0.0852. The invulnerability of the whole network and the average residual energy of the whole network are denormalized, and the corresponding network parameters are obtained as follows: L′=6, m′=20, Gcr′=4, z′=4 and n′=1100. The spider-like hierarchical clustering network can be optimized by adjusting the parameter combination of L′=6, m′=20, Gcr′=4, z′=4 and n′=1100.

[0154] The present invention uses a statistical method to analyze the network evolution process caused by changes in network parameters, obtains the degree of influence of different network parameters and different values ​​of network parameters on network performance, and uses the degree of influence as the basis for network model parameter optimization calculation. The network indestructibility and the average residual energy of the entire network are used as optimization objective functions, and a multi-objective optimization algorithm is used to optimize the network parameters, thereby providing an effective method for reasonably optimizing the network parameters of the spider web-like hierarchical clustering network.

Claims

1. A spider web-like hierarchical clustering network parameter optimization method, characterized in that: The steps include: 1) The spider web-like hierarchical clustering network was simulated under different network parameter combinations to obtain the invulnerability data of the whole network and the average residual energy data of the whole network; the network parameters of the spider web-like hierarchical clustering network include the number of network layers L, the number of cluster head replacement rounds m, the intra-cluster hierarchical radius division level Gcr, the number of sector areas z and the number of sensor nodes n; the intra-cluster hierarchical radius division level Gcr is used to characterize the intra-cluster hierarchical radius, each cluster is centered on the cluster head, and the clusters are divided according to the intra-cluster hierarchical radius; the invulnerability of the whole network is calculated by the following formula: Where INV is the invulnerability of the entire network, is the number of nodes in the vth cluster in the network, is the number of nodes in the entire network, is the invulnerability of the vth cluster, h is the number of clusters in the network; 2) Perform normality test and variance homogeneity test on the network invulnerability data and the network average residual energy data. If the network invulnerability data and the network average residual energy data are both normally distributed and meet the variance homogeneity, the network parameters are taken as fixed factors and the network invulnerability is taken as the dependent variable to perform the inter-subject effect test. The inter-subject effect test results corresponding to the network parameters and the network invulnerability are obtained. ; The network parameters were taken as fixed factors and the average residual energy of the whole network was taken as the dependent variable to conduct the inter-subject effect test, and the inter-subject effect test results corresponding to the network parameters and the average residual energy of the whole network were obtained. , i=1,…,5; 3) Normalize the network parameters and the corresponding network-wide invulnerability data respectively, and establish a network-wide invulnerability model based on the normalized data ; The whole network invulnerability model Used to express the corresponding relationship between network parameters and the invulnerability of the entire network; according to the inter-subject effect test results Establishing a network-wide survivability model Constraints of Normalize the network parameters and the corresponding average residual energy data of the whole network respectively, and establish the average residual energy model of the whole network based on the normalized data ; The average residual energy model of the whole network Used to express the correspondence between network parameters and the average residual energy of the whole network; according to the inter-subject effect test results Establishing the average residual energy model of the whole network Constraints of 4) The network-wide invulnerability data and the network-wide average residual energy data are used as the input of the genetic algorithm. and The maximum difference is taken as the crowding degree of individuals in the non-dominated layer in the genetic algorithm, and the population is iterated according to the crowding degree. At the end of the iteration, the Pareto optimal frontier of the whole network invulnerability model and the whole network average residual energy model is obtained; 5) Compare the Pareto optimal frontier with the straight line The intersection of is taken as the optimal solution equilibrium point, the network parameters corresponding to the optimal solution equilibrium point are calculated, and the spider web hierarchical clustering network is optimized according to the network parameters.

2. The spider web-like hierarchical clustering network parameter optimization method according to claim 1, characterized in that: The spider web-like hierarchical clustering network is established by the following steps: S1, imitate the circular spider web topology structure, divide the area of ​​the circular network monitoring area, and n sensor nodes are randomly distributed in the circular network monitoring area; S2, according to the number of network layers L and the number of sector areas z, the circular network monitoring area is divided into multiple clusters of equal area; S3, randomly select the cluster heads for the first communication, and after each elected cluster head performs several rounds of communication, the next cluster head election is performed, and the number of cluster head elections does not exceed the number of cluster head replacement rounds m; S4. The intra-cluster layer radius is divided into levels Gcr according to the intra-cluster layer radius, and the intra-cluster network communication is performed according to the intra-cluster layer radius.

3. The spider web-like hierarchical clustering network parameter optimization method according to claim 1, characterized in that: according to calculate L , m , Gcr , z and n The corresponding weight , then according to And the corresponding network parameters to obtain the whole network invulnerability model Constraints :according to calculate L , m , Gcr , z and n The corresponding weight , then according to Get the average residual energy model of the whole network Constraints ; In the formula, is the weight corresponding to the ith network parameter and the corresponding network invulnerability, is the inter-subject effect test result corresponding to the ith network parameter and the corresponding network invulnerability, is the weight corresponding to the ith network parameter and the corresponding residual energy of the whole network balance, is the inter-subject effect test result corresponding to the ith network parameter and the corresponding average residual energy of the whole network, is the i-th network parameter, .

4. The spider web-like hierarchical clustering network parameter optimization method according to claim 3, characterized in that: The genetic algorithm described in step 4) comprises the following steps: 4.

1. Generate a parent population based on the input of the genetic algorithm, and obtain a child population through selection, crossover and mutation operations; 4.

2. Starting from the second generation, the parent population and the child population are merged, and the non-dominated levels of the individuals in the merged population are sorted, and the individuals with high non-dominated levels are selected for population iteration; During the sorting process, according to the constraints and Determine whether the solution of an individual in the population is a feasible solution. The non-dominated level of the individual corresponding to the feasible solution is higher than the non-dominated level of the individual corresponding to the infeasible solution. If both are feasible solutions, the greater the crowding degree of the individual, the higher the non-dominated level. If both are infeasible solutions, the smaller the constraint violation value of the individual, the higher the non-dominated level. The constraint violation value is used to characterize the degree to which the individual solution violates the constraint condition.

5. The spider web-like hierarchical clustering network parameter optimization method according to claim 1, characterized in that: In step 2), the normality test of the whole network invulnerability data and the whole network average remaining energy data is performed through KS test, histogram and QQ graph. If the result of KS test on the whole network invulnerability data is greater than the first set value, the histogram of the whole network invulnerability data conforms to the normal distribution, and the whole network invulnerability data coincides with the first theoretical straight line in the QQ graph, it is considered that the whole network invulnerability data obeys the normal distribution; if the result of KS test on the whole network average remaining energy data is greater than the first set value, the histogram of the whole network average remaining energy data conforms to the normal distribution, and the whole network average remaining energy data coincides with the second theoretical straight line in the QQ graph, it is considered that the whole network average remaining energy data obeys the normal distribution; the first theoretical straight line is obtained according to the standard deviation and mean of the whole network invulnerability data; the second theoretical straight line is obtained according to the standard deviation and mean of the whole network average remaining energy data.

6. The spider web-like hierarchical clustering network parameter optimization method according to claim 5, characterized in that: In step 2), when performing a variance homogeneity test on the invulnerability data of the entire network, a variance homogeneity test result based on the median is used for judgment. If the variance homogeneity test result based on the median is greater than a second set value, it is considered that the invulnerability data of the entire network meets the variance homogeneity. When performing a variance homogeneity test on the average remaining energy data of the entire network, a variance homogeneity test result based on the median is used for judgment. If the variance homogeneity test result based on the median is greater than a second set value, it is considered that the average remaining energy data of the entire network meets the variance homogeneity.

7. The spider web-like hierarchical clustering network parameter optimization method according to claim 6, characterized in that: In step 2), if the invulnerability data of the whole network does not obey the normal distribution and does not satisfy the homogeneity of variance, the invulnerability data of the whole network is logarithmically transformed. If the invulnerability data of the whole network after logarithmic transformation obeys the normal distribution and satisfies the homogeneity of variance, it is considered that the invulnerability data of the whole network obeys the normal distribution and satisfies the homogeneity of variance; If the average remaining energy data of the whole network does not obey the normal distribution and does not satisfy the variance homogeneity, the average remaining energy data of the whole network is logarithmically transformed. If the average remaining energy data of the whole network after logarithmic transformation obeys the normal distribution and satisfies the variance homogeneity, it is considered that the average remaining energy data of the whole network obeys the normal distribution and satisfies the variance homogeneity.

8. The spider web-like hierarchical clustering network parameter optimization method according to claim 1, characterized in that: Perform data fitting on the normalized network parameters and the corresponding network-wide invulnerability data to obtain the network-wide invulnerability model ; Perform data fitting on the normalized network parameters and the corresponding average residual energy data of the entire network to obtain the average residual energy model of the entire network .

9. The spider web-like hierarchical clustering network parameter optimization method according to claim 1, characterized in that: In step 1), when the simulation experiment is conducted on the spider-web-like hierarchical clustering network under different network parameter combinations, only one network parameter changes between different network parameter combinations.

10. The spider web-like hierarchical clustering network parameter optimization method according to claim 9, characterized in that: In step 1), a simulation experiment is conducted on the spider web-like hierarchical clustering network under different network parameter combinations. If the number of node deaths in the spider web-like hierarchical clustering network exceeds a set percentage of the number of sensor nodes n, the simulation experiment is terminated.

Images