Wireless sensor network energy efficiency competition clustering method based on fuzzy width learning

Abstract

The invention relates to the field of wireless sensor network clustering optimization, in particular to a wireless sensor network energy efficiency competition clustering method based on fuzzy width learning. The method comprises the following steps: collecting historical node information of the wireless sensor network, and preprocessing the historical node information to generate a historical node data set; training and testing the fuzzy width learning model according to the historical node data set to obtain a trained model; nodes of the wireless sensor network and corresponding node information are generated, the node information is preprocessed and then input into the trained model, and competitiveness values of the nodes are obtained; obtaining an optimal cluster head number; selecting a cluster head node according to the competitiveness value of the node, and setting other nodes as common nodes; and according to the cluster head nodes and the common nodes, constructing a plurality of clusters in the wireless sensor network for data transmission. The method is used for dynamically selecting proper cluster head nodes and optimizing the clustering process so as to meet the real-time and adaptive requirements of the wireless sensor network.

Description

Technical Field

[0001] The present invention relates to the field of wireless sensor network clustering optimization, and more specifically, to a wireless sensor network energy efficiency competition clustering method based on fuzzy width learning. Background Art

[0002] With the rapid development of Internet of Things technology, Wireless Sensor Network (WSN) as the core technology has been widely used in many fields such as environmental monitoring, military reconnaissance, medical care, smart home, etc. Wireless sensor network consists of a large number of sensor nodes with perception, computing and communication capabilities. These sensor nodes are distributed in the monitoring area and complete data collection, processing and transmission tasks through collaboration. The collected data will be reported to the base station BS (BaseStation). Through the base station BS, data communication between end users and wireless sensor networks can be realized. However, since sensor nodes are usually powered by batteries, energy resources are limited and difficult to replenish. How to effectively use the energy of sensor nodes to extend the network life has become one of the core issues in the study of wireless sensor networks.

[0003] In order to solve the problem of limited energy and avoid the huge energy consumption caused by direct communication between each sensor node and the base station BS, the cluster routing method, which is a cluster-based transmission protocol, is currently being studied and widely used. The cluster routing method divides the wireless sensor network into multiple clusters, each cluster consists of a cluster head node and several ordinary nodes, where the cluster head node CH (Cluster Head) is responsible for aggregating the data collected by ordinary nodes in the cluster and forwarding it to the base station BS. This method can effectively reduce the communication overhead between nodes, thereby reducing energy consumption. In order to achieve the above effect, how to reasonably select the cluster head node of each cluster is a necessary study in this neighborhood. In the prior art, a method for selecting cluster head node CH based on LEACH low-power adaptive clustering hierarchical routing protocol is disclosed, but there are defects such as randomness and uncertainty in the selection of cluster head node CH. Low-energy nodes may be selected as cluster head nodes CH, resulting in premature energy exhaustion; a method for dynamically selecting cluster head nodes CH based on fuzzy logic system is disclosed, but it relies too much on expert experience and the effect is not good in dynamic environment; a heuristic method based on chaotic ant colony is disclosed for clustering process, but it is easy to encounter performance bottlenecks on large-scale problems; in addition, there are other deep learning complex methods that cannot meet the current real-time requirements of wireless sensor networks due to excessive time overhead; therefore, how to achieve reasonable cluster head selection and cluster group construction in complex dynamic environment is still a challenge in wireless sensor network problems. Summary of the invention

[0004] The present invention aims to overcome at least one defect (shortcoming) of the above-mentioned prior art and provide a wireless sensor network energy-efficient competitive clustering method based on fuzzy width learning, which is used to dynamically select suitable cluster head nodes and optimize the clustering process to meet the real-time and adaptability requirements of the wireless sensor network.

[0005] According to a first aspect of the present application, a wireless sensor network energy efficiency competitive clustering method based on fuzzy width learning is provided, the method comprising:

[0006] Collecting some historical node information of the wireless sensor network, and preprocessing some of the historical node information to generate a historical node data set;

[0007] Training and testing the fuzzy width learning model according to the historical node data set to obtain a trained fuzzy width learning model;

[0008] Generate a number of nodes and corresponding node information of the wireless sensor network, pre-process the node information and input it into the trained fuzzy width learning model, and calculate and obtain the competitiveness value of each node corresponding to the node information through the fuzzy width learning model;

[0009] Acquire the properties of the wireless sensor network, and acquire the optimal number of cluster heads of the wireless sensor network according to the properties of the wireless sensor network;

[0010] According to the competitiveness value of each node, a number of cluster head nodes are selected, wherein the number of the selected cluster head nodes is equal to the number of the optimal cluster heads; nodes that are not selected as cluster head nodes are set as ordinary nodes;

[0011] According to the plurality of cluster head nodes and the plurality of common nodes, a plurality of cluster groups in the wireless sensor network are constructed for data transmission.

[0012] It is understandable that the introduction of the fuzzy width learning model calculates the competitiveness value of each node in the wireless sensor network, and by comparing the competitiveness value of each node, selects the cluster head node suitable for the wireless sensor network, and obtains the optimal number of cluster heads according to the properties of the wireless sensor network, which can make the transmission work of the entire wireless sensor network more efficient and orderly, and at the same time optimize the clustering process of the wireless sensor network to meet the real-time and adaptability requirements of the wireless sensor network.

[0013] Optionally, preprocessing the historical node information to generate a historical node data set includes:

[0014] Performing data normalization processing on the historical node information to obtain historical node information of the same dimension;

[0015] Performing missing value processing on a number of historical node information of the same dimension to obtain a number of complete historical node information; wherein the missing value processing is to obtain node information of which competitiveness value is missing in the number of node information of the same dimension, and inputting the node information of which competitiveness value is missing into the fuzzy logic system for processing to obtain the corresponding competitiveness value, and completing the information of the node of which competitiveness value is missing according to the obtained competitiveness value;

[0016] The data format conversion process is performed on the complete historical node information to obtain a historical node data set.

[0017] It is understandable that data normalization, missing value processing and data format conversion processing of historical node information can provide a complete and normalized data set for subsequent training of the fuzzy width learning model, improve the efficiency of the fuzzy width learning model in data processing and obtain as much training information as possible.

[0018] Optionally, the competitiveness value depends on influencing factors, which include one or more of the relative residual energy Energy_Re(i) of the node, the relative distance Dist_BS(i) from the node to the base station BS, the neighbor node density Nei_Dens(i) of the node and the historical cluster head state CH_State(i) of the node.

[0019] It is understandable that the fuzzy width learning model needs to be trained and tested so that the trained fuzzy width learning model can obtain the competitiveness value of the node based on the input node information. The present application provides four influencing factors of the competitiveness value. When the fuzzy width learning model is trained and tested, the influencing factors can be used as influencing data for learning the method of calculating the competitiveness value, so that the method of calculating the competitiveness value learned by the fuzzy width learning model is more accurate and the competitiveness value obtained is reliable.

[0020] Optionally, the relative residual energy Energy_Re(i) of the node is obtained by the following formula:

[0021]

[0022] Energy res (i) represents the remaining energy of node i, Energy init (i) represents the initial energy of node i;

[0023] and / or,

[0024] The relative distance Dist_BS(i) from the node to the base station BS is obtained by the following formula:

[0025]

[0026] Wherein, d1, d2, d3, d4 represent the distances from the four boundary corners in the wireless sensor network scene to the base station BS, max(.) is the maximum value function, indicating the selection of the farthest distance from the base station BS among d1, d2, d3, f4; Dist BS (S i ) represents the Euclidean distance from node i to the base station BS, which is obtained by the following calculation formula:

[0027]

[0028] where x i ,y i Respectively represent the horizontal and vertical coordinates of node i, x BS ,y BS Respectively represent the horizontal and vertical coordinates of the base station BS;

[0029] and / or,

[0030] The neighbor node density Nei_Dens(i) of the node i is specifically obtained by the following formula:

[0031]

[0032] Among them, N w is the total number of nodes in the wireless sensor network; Neighbor(i) is the number of neighbor nodes of the node i, which is obtained by the following formula:

[0033]

[0034] Wherein, j represents other nodes j in the wireless sensor network except the node i, d ij represents the distance between the node i and the other node j, R t represents the maximum radius distance of the node i to transmit relevant information, δ(.) is the indicator function, when d ij ≤R, δ=1; otherwise, δ=0;

[0035] and / or,

[0036] The historical cluster head state CH_State(i) of the node is obtained by the following formula:

[0037]

[0038] Wherein, r represents the number of times the node i is continuously selected as the cluster head node.

[0039] It can be understood that the present application introduces the relative residual energy of a node, the relative distance from a node to a base station BS, the density of neighboring nodes of a node, and the method for obtaining the relative residual energy of a node, and provides formulas and methodologies for obtaining influencing factors; wherein the historical cluster head state of a node is used as one of the influencing factors for calculating the competitiveness value, and the real-time energy state of each node can be fully considered, so that the obtained competitiveness value can better reflect the most real state of a node, and the cluster head node selected later can be more accurate.

[0040] Optionally, the fuzzy width learning model includes an input layer, a fuzzy feature layer, a feature enhancement layer and an output layer; the fuzzy feature layer includes several fuzzy subsystems;

[0041] The training and testing of the fuzzy width learning system network model according to the historical node data set includes:

[0042] Dividing the historical node data set into a training set and a test set;

[0043] Setting the number Q of fuzzy subsystems and the number K of fuzzy rules;

[0044] Initializing coefficient weights of the fuzzy width learning model;

[0045] Inputting the training set into the fuzzy width learning model through the input layer;

[0046] In the fuzzy feature layer, an intermediate matrix Z is generated by the fuzzy subsystems of the fuzzy width learning model and the training set. n and the feature node matrix F n , where n is the total number of training samples in the training set;

[0047] The intermediate matrix Z n Each node feature value in is input into the feature enhancement layer, and several enhanced nodes H are obtained by enhanced feature processing. m , according to the characteristic node matrix F n and the plurality of enhanced nodes H m Obtain a target value Y, and obtain a weight matrix W based on the target value Y and pseudo-inverse fast calculation;

[0048] In the output layer, preliminary parameters of the fuzzy width learning model are obtained according to the weight matrix W;

[0049] Input the test set into the fuzzy width learning model to calculate the error, wherein the error includes the error rate ERR and the mean absolute percentage error MAPE;

[0050] After optimizing the number of fuzzy subsystems Q and the number of fuzzy rules K according to the error rate ERR and the mean absolute percentage error MAPE, the fuzzy width learning model is iteratively trained, the preliminary parameters are optimized to obtain the optimal model parameters, and the training of the fuzzy width learning model is completed.

[0051] It is understandable that the fuzzy width learning model is trained and tested, and the parameters of the fuzzy width learning model are continuously iterated and optimized by calculating the relevant errors until a certain threshold is reached, and the training and testing of the model are terminated to obtain a trained fuzzy width learning model. This can provide a clear methodology for training the fuzzy width learning model, making the entire training process clearer and more organized.

[0052] Optionally, each training sample in the training set is N×M dimensional data, where N represents the number of groups of the training samples and M represents the dimension of feature information of the training samples, which is specifically expressed as:

[0053] X=(x1,x2,...,x s ,...,x N ) T ∈R N×M

[0054] where x s =(x s1 ,x s2 ,…,x sm ,…,x sM ), s=1,2,…,N;

[0055] And / or, the setting of the number of fuzzy subsystems Q and the number of fuzzy rules K specifically includes:

[0056] In the qth fuzzy subsystem, set K q The fuzzy rules conform to the following representation:

[0057] If x s1 is and x s2 is … and x sM is

[0058] but

[0059] Where k = 1, 2, ..., K q , x sm represents one of the training samples in the training set, represents the preset fuzzy set in the fuzzy subsystem, is the adjustable parameter of the model rule, uqg Represents the training sample g in the training set for the fuzzy rule K q The membership degree of is the intermediate matrix Z n The output of the qth fuzzy subsystem, where u qg It can be expressed as:

[0060]

[0061] Among them, c q 、c k represents the cluster center corresponding to the fuzzy rule;

[0062] Then the training sample x s The output Z of the qth fuzzy subsystem sq It is expressed as:

[0063]

[0064] in, The output Z of the qth fuzzy subsystem is sq A single intermediate value The weight of

[0065] Then the training set is the intermediate matrix Z of the combined output of Q fuzzy subsystems n It is expressed as:

[0066] Z n =(Z1,Z2,…,Z n )

[0067] The defuzzified output F of the fuzzy feature layer sq It is expressed as:

[0068]

[0069] in, is an adjustable parameter;

[0070] The qth fuzzy subsystem defuzzifies the output F q for:

[0071]

[0072] in, diag{.} is a diagonal matrix;

[0073] The output F of the fuzzy feature layer n It is expressed as:

[0074]

[0075] And / or, the intermediate matrix Z n Each node feature value in is input into the feature enhancement layer, and several enhanced nodes H are obtained by enhanced feature processing. m ,include:

[0076] The intermediate matrix Z n Each feature node value Z in I The value H mapped to the enhanced node J :

[0077]

[0078] Among them, ξ J (.) is the mapping function, is a random weight, is the bias term;

[0079] Then the number of enhanced nodes H of the m features m The output is represented as:

[0080] H m =(H1,H2,…,H m )

[0081] And / or, according to the characteristic node matrix F n and the plurality of enhanced nodes H m Get the target value Y, including:

[0082] The target value Y output by the output layer is:

[0083]

[0084] Among them, W f is the weight coefficient from the fuzzy feature layer to the output layer, W h is the weight coefficient from the feature enhancement layer to the output layer;

[0085] And / or, the weight matrix W is obtained by fast calculation based on the target value Y and pseudo-inverse, specifically according to the following formula:

[0086] W=(Bω,H m ) + Y

[0087] Among them, (Bω,H m ) + =((Bω,H m ) T (Bω,H m )) -1 (Bω,H m ) T ;

[0088] And / or, the calculation formula of the mean absolute percentage error MAPE is expressed as:

[0089]

[0090] Among them, y v is the true value of the vth intermediate target of the target value Y, for y v The predicted value of

[0091] And / or, the calculation formula of the error rate ERR is expressed as:

[0092]

[0093] Among them, E baseline is the benchmark error value of the fuzzy width learning model, E m0del Represents the error value of the current training process of the fuzzy width learning model.

[0094] It is understandable that the calculation formulas of the relevant steps in the training or testing process shown above can provide an accurate calculation method for the entire training process, making the calculation or processing in the training and testing of the fuzzy width model more clear and organized.

[0095] Optionally, the attributes of the wireless sensor network include the total number of nodes in the wireless sensor network, the range of the wireless sensor network, and the average distance from each of the nodes in the wireless sensor network to a base station BS;

[0096] The obtaining the optimal number of cluster heads of the wireless sensor network according to the property of the wireless sensor network includes:

[0097] The optimal number of cluster heads k opt The specific formula is as follows:

[0098]

[0099] Among them, N w is the total number of nodes in the wireless sensor network, P is the range of the wireless sensor network, d toBS is the average distance from each of the nodes in the wireless sensor network to the base station BS;

[0100] The average distance from each node in the wireless sensor network to the base station BS is obtained by the following formula:

[0101]

[0102] It is understandable that the number of cluster heads required for different types of wireless sensor networks, including the total number of nodes, range size and attribute information of each node, will be different. Too few cluster heads cannot complete the transmission problem of the network well, and too many cluster heads may waste related resources. Therefore, it is necessary to obtain the optimal number of cluster heads according to the attributes of each wireless sensor network, so that the transmission work of this wireless sensor network can be distributed more reasonably and the transmission efficiency can be improved.

[0103] Optionally, the selecting a number of cluster head nodes according to the competitiveness value of each node includes:

[0104] Sort the competitiveness values ​​of all the nodes from high to low, select the nodes with high competitiveness as cluster head nodes, and the remaining nodes as ordinary nodes;

[0105] and / or,

[0106] The constructing a plurality of cluster groups in the wireless sensor network according to the plurality of cluster head nodes and the plurality of common nodes for data transmission includes:

[0107] The cluster group is constructed by a second cluster head strategy, a secondary cluster head strategy and a minimum energy consumption criterion strategy;

[0108] The second cluster head strategy is to select a node with the largest competitiveness value other than the cluster head node from the constructed cluster group to serve as the second cluster head, and the second cluster head assists the cluster head node in transmission work;

[0109] The secondary cluster head strategy is to select a node in the cluster group that can replace the cluster head node as the secondary cluster head node. When the cluster head node fails or cannot work normally, the secondary cluster head node serves as the cluster head node to complete the work of the cluster head node.

[0110] The minimum energy consumption criterion strategy is to calculate the indirect energy consumption E of data transmitted indirectly from the ordinary node to the base station BS through the cluster head node when the ordinary node in the wireless sensor network transmits data. idt , and calculate the direct energy consumption E of data transmitted directly from the ordinary node to the base station BS dt , and compare the indirect energy consumption E through the cost function idt and the direct energy consumption E dt The size of , selects the path with the lowest energy consumption as the transmission path.

[0111] It is understandable that selecting cluster head nodes based on the competitiveness value of the nodes can truly reflect the status and transmission capacity of the nodes, thereby selecting nodes that are more suitable for cluster heads, ensuring that the complex transmission tasks that the cluster heads are responsible for can be completed smoothly, so that the transmission work of the entire wireless sensor network can also be completed smoothly; to construct cluster groups in wireless sensor networks, it is necessary to follow the second cluster head strategy, the deputy cluster head strategy and the minimum energy consumption criterion strategy; the second cluster head strategy can ensure that the work of the cluster head nodes is completed smoothly and efficiently, and reduce the workload of the cluster head nodes; the deputy cluster head strategy can ensure rapid response when a cluster head node fails, reducing the impact of the failure on the entire transmission work; the minimum energy consumption criterion strategy can ensure that the energy consumption of the entire transmission work is reduced, making the entire transmission work more economical and reliable.

[0112] Optionally, the indirect energy consumption and / or direct energy consumption is affected by the following parameters:

[0113] The distance a from the common node to the cluster head node is: E idt ∝a;

[0114] The distance b from the cluster head node to the base station BS is: idt ∝b;

[0115] The distance c from the common node to the base station BS is: dt ∝c;

[0116] Then the indirect energy consumption E idt The specific formula is as follows:

[0117] E idt =l·[(1+μ)E elec +∈ fs (a 2 +μb 2 )]

[0118] Wherein, μ represents the preset data aggregation rate;

[0119] Among them, E elec represents the energy consumption coefficient of the path, ∈ fs represents the power amplification factor of the existing free space model;

[0120] The direct energy consumption E dt The specific formula is as follows:

[0121] E dt = l·(E elec +∈ fs c 2 )

[0122] Then the indirect energy consumption E idt and the direct energy consumption Edt The energy consumption difference △E is obtained by the following formula:

[0123] △E=E idt -E dt = l·[μE elec +ε fs (a 2 +μb 2 -c 2 )]

[0124] When △E>0, it means E dt The value is smaller, and the ordinary node selects data to be directly transmitted from the ordinary node to the base station BS as the transmission path for data transmission; when △E<0, it means that E idt The value is smaller, and the common node data is indirectly transmitted from the common node through the cluster head node to the base station BS as a transmission path for data transmission.

[0125] It is understandable that selecting a suitable path to transmit information can save energy consumption used for transmission, thereby reducing transmission consumption in the entire wireless sensor network. Optionally, the method further includes:

[0126] After the cluster group is built, the base station BS sends a confirmation message BS_info message to each node in the wireless sensor network, including the ID of the cluster head node, the coordinates of the cluster head node and the cluster group to which the cluster head node belongs; after receiving the confirmation message, each node starts working and enters the data collection and transmission stage;

[0127] In the data collection and transmission stage, the cluster head node aggregates the collected data; wherein, the data aggregation adopts the IA data aggregation model, and the cluster head node aggregates and forwards the received data packet length according to the data aggregation rate, and the specific calculation formula is as follows:

[0128] L agg =L r +μ×L r ×N p

[0129] Among them, L agg Indicates the length of the data packet after aggregation, L r represents the length of the received data packet, μ represents the data aggregation rate, N p Indicates the number of packets received.

[0130] It is understandable that a reliable method is provided for cluster head node data transmission, data redundancy is reduced, the information accessed by the entire transmission work is made more flexible and streamlined, and the usability of the wireless sensor network memory is improved.

[0131] Based on any of the above aspects, an embodiment of the present application provides a wireless sensor network energy efficiency competitive clustering method based on fuzzy width learning, which collects some historical node information of the wireless sensor network, pre-processes some of the historical node information to generate a historical node data set; trains and tests a fuzzy width learning model according to the historical node data set to obtain a trained fuzzy width learning model; generates some nodes and corresponding node information of the wireless sensor network, pre-processes some of the node information and inputs them into the trained fuzzy width learning model, and calculates and obtains the competitiveness value of each node corresponding to the node information through the fuzzy width learning model; obtains the attributes of the wireless sensor network, and obtains the optimal number of cluster heads of the wireless sensor network according to the attributes of the wireless sensor network; selects some cluster head nodes according to the competitiveness value of each node, wherein the number of the selected cluster head nodes is equal to the optimal number of cluster heads; nodes that are not selected as cluster head nodes are set as ordinary nodes; according to some of the cluster head nodes and some of the ordinary nodes, constructs some cluster groups in the wireless sensor network for data transmission; compared with the prior art, the present invention has the following advantages and technical effects:

[0132] 1) The present invention provides a competitive clustering method based on a fuzzy width learning model that can accurately and quickly process the clustering process in a complex and changeable environment of a wireless sensor network, thereby solving the defect of unreasonable cluster head decision-making;

[0133] 2) The fuzzy width learning model proposed in the present invention is used to calculate the competitiveness of each node, which combines the characteristic information processing capability of the fuzzy system and the fast training capability of the width learning system;

[0134] 3) The present invention adaptively determines the best cluster head node in the current situation according to the dynamic changes of the network environment, effectively improving the energy efficiency of the network; in addition, the "second cluster head" and "secondary cluster head" strategies are added to achieve the balance of network energy efficiency and provide a fault-tolerant mechanism;

[0135] 4) The present invention proposes a wireless sensor network energy efficiency competitive clustering method based on fuzzy width learning. It is verified through simulation experiments in multiple environments that it has good advantages in energy efficiency, network life and network stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0136] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0137] Figure 1 A flow chart of a wireless sensor network energy-efficient competitive clustering method based on fuzzy width learning provided in this embodiment.

[0138] Figure 2 A flow chart of a method for generating a historical node data set provided in this embodiment.

[0139] Figure 3 A flow chart of a method for training and testing a fuzzy width learning model provided in this embodiment.

[0140] Figure 4 A diagram showing the average remaining energy of nodes in different embodiments of the present invention and the comparison method provided for this embodiment.

[0141] Figure 5 A diagram showing fluctuations in the average remaining energy performance of nodes in different embodiments of the methods provided in this embodiment. DETAILED DESCRIPTION

[0142] The drawings of this application are only used for illustrative purposes and should not be construed as limiting the present application. In order to better illustrate the following embodiments, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product; it is understandable to those skilled in the art that some well-known structures and their descriptions in the drawings may be omitted.

[0143] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0144] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0145] With the rapid development of Internet of Things technology, wireless sensor networks, as its core technology, have been widely used in many fields. Wireless sensor networks are composed of a large number of sensor nodes with perception, computing and communication capabilities. These nodes are distributed in the monitoring area and collaborate to complete data collection, processing and transmission tasks. The data is finally reported to the base station BS. However, the energy of sensor nodes is limited and difficult to replenish. How to effectively use energy to extend the life of the network has become a core research issue.

[0146] In order to solve the problem of limited energy, cluster routing methods are widely used. This method divides the wireless sensor network into multiple clusters. Each cluster consists of a cluster head node CH and several ordinary nodes. The cluster head node CH is responsible for aggregating and forwarding the data in the cluster to the base station BS, thereby reducing communication overhead and energy consumption. Reasonable selection of cluster head nodes is the key, but the existing technology has defects: the method based on the LEACH low-power adaptive clustering hierarchical routing protocol has randomness and uncertainty; the fuzzy logic system relies on expert experience and has poor dynamic effect; the chaotic ant colony method has performance bottlenecks on large-scale problems; the deep learning complex method has a large time overhead and is difficult to meet real-time requirements. Therefore, how to achieve reasonable cluster head selection and cluster group construction in a complex dynamic environment is still a major challenge facing wireless sensor networks.

[0147] This embodiment provides a technical solution that can solve the above-mentioned problem. The specific implementation methods of this application are described in detail below in conjunction with the accompanying drawings.

[0148] like Figure 1 As shown, this embodiment provides a wireless sensor network energy efficiency competition clustering method based on fuzzy width learning, which may include the following steps:

[0149] S110, collecting some historical node information of the wireless sensor network, and preprocessing some of the historical node information to generate a historical node data set;

[0150] In this embodiment, historical node information includes several historical nodes in the historical transmission work of the wireless sensor network, and the historical nodes also include corresponding historical node information, and the historical node information includes the node's ID unique identity number, the node's location information in the wireless sensor network, the node's energy information, and the node's status information; it can be understood that each node has a corresponding node device to complete the corresponding transmission work, and the historical node information is stored in the node device for call or use.

[0151] Specifically, Figure 2 As shown, the preprocessing of the historical node information to generate a historical node data set includes the following steps:

[0152] S111, performing data normalization processing on the plurality of historical node information to obtain a plurality of historical node information of the same dimension;

[0153] In this embodiment, the data normalization process refers to normalizing all historical node information to make it distributed in (0,1) so that historical node information of different dimensions has the same dimension, which is convenient for subsequent processing;

[0154] S112, performing missing value processing on the historical node information of the same dimension to obtain a number of complete historical node information; wherein the missing value processing is to obtain node information with missing competitiveness values ​​in the node information of the same dimension, and input the node information with missing competitiveness values ​​into the fuzzy logic system for processing to obtain corresponding competitiveness values, and complete the information of the node with missing competitiveness values ​​according to the obtained competitiveness values;

[0155] In this embodiment, the fuzzy logic system mainly includes four parts: a fuzzification processor, a fuzzy inference engine, a fuzzy rule base and a defuzzification processor;

[0156] The fuzzification processor is used to perform fuzzification processing on the input node information, and convert the input node information into the membership value in the fuzzy set in the fuzzy logic system. The commonly used membership functions include triangle membership function, trapezoidal membership function, Gaussian membership function and Sigmoid membership function.

[0157] The fuzzy inference engine is responsible for inferring and calculating the membership value generated by the fuzzification processor according to the input node information based on the fuzzy rule base, and adopts a specific inference method to transform the input fuzzy set into the output fuzzy set. Commonly used inference methods include Mamdani fuzzy inference method and Takagi-Sugeno fuzzy inference method.

[0158] The fuzzy rule base is used to store the knowledge base of the fuzzy logic system. The rules are in the form of "if-then" statements, which describe the impact of different input variable combinations on the output;

[0159] The function of the defuzzification processor is to convert the output fuzzy set of the fuzzy inference engine into a specific precise value to achieve the defuzzification function. Commonly used methods include the center method, the maximum membership method and the weighted average method;

[0160] S113, performing data format conversion processing on some of the complete historical node information to obtain a historical node data set.

[0161] In this embodiment, the data format conversion process refers to converting the data format into (X, Y)=(RE, DB, ND, CS, CV) according to the relationship between the historical node information.

[0162] S120, training and testing the fuzzy width learning model according to the historical node data set to obtain a trained fuzzy width learning model;

[0163] Specifically, the competitiveness value depends on the influencing factors, which include one or more of the relative residual energy Energy_Re(i) of the node, the relative distance Dist_BS(i) from the node to the base station BS, the neighbor node density Nei_Dens(i) of the node and the historical cluster head state CH_State(i) of the node.

[0164] Specifically, the relative remaining energy Energy_Re(i) of the node is obtained by the following formula:

[0165]

[0166] Among themEngrgy res (i) represents the remaining energy of node i, Engrgy init (i) represents the initial energy of the node;

[0167] It can be understood that the more residual energy a node has, the greater the energy it can use for data transmission, the more suitable it is for the role of cluster head node CH for data collection, and the more competitive it is for the cluster head node;

[0168] And / or, the relative distance Dist_BS(i) from the node to the base station BS is obtained by the following formula:

[0169]

[0170] Wherein, d1, d2, d3, d4 represent the distances from the four boundary corners in the wireless sensor network to the base station BS, max(.) is the maximum value function, indicating the selection of the farthest distance from the base station BS among d1, d2, d3, d4; Dist BS (S i ) represents the Euclidean distance from the node i to the base station BS, which is specifically obtained by the following calculation formula:

[0171]

[0172] where x i ,y i Respectively represent the horizontal and vertical coordinates of node i, x BS ,y BS Respectively represent the horizontal and vertical coordinates of the base station BS;

[0173] It is understandable that the closer the node is to the base station BS, the less energy consumption is required for transmission, the more data forwarding tasks it can undertake, and the greater competitiveness of the cluster head node. Although the coverage of the distant node may be larger, the energy required for data transmission is also increased accordingly;

[0174] And / or, the neighbor node density Nei_Dens(i) of the node is specifically obtained by the following formula:

[0175]

[0176] Among them, N w is represented by the total number of nodes in the wireless sensor network, and Neighbor(i) represents the number of neighbor nodes of the node i, which is specifically obtained by the following formula:

[0177]

[0178] Wherein, j represents other nodes j in the wireless sensor network except the node i, d ij represents the distance between the node i and the other node j, R t represents the maximum radius distance of the node i to transmit relevant information, δ(.) is the indicator function, when d ij ≤R t When , δ=1; otherwise, δ=0;

[0179] It is understandable that a node i is within the radius R tThe ADV message is broadcasted within the range, including the sending node ID; when the neighboring node receives the broadcast message, it returns the confirmation message ACK message, including the receiving node ID, node location and node remaining energy; after receiving the confirmation message, node i stores the information in the adjacency table; the adjacency table of the node is updated in each round; the number of neighbor nodes of node i can be calculated through the adjacency table of node i; it is understandable that the greater the density of neighbor nodes of the node, the better it is to collect and transmit data, the lower the transmission and aggregation energy consumption, and the node is more suitable to serve as a cluster head node;

[0180] And / or, the historical cluster head state CH_State(i) of the node is obtained by the following formula:

[0181]

[0182] Wherein, r represents the number of times the node i is continuously selected as the cluster head node;

[0183] It is understandable that if a node serves as a cluster head node for many consecutive times, the node energy will be consumed rapidly. Excessive consumption will cause the node to die, affecting the stable operation of the wireless sensor network. The historical status of the node is recorded and the cluster head node is replaced regularly. For the nodes that did not serve as cluster head nodes in the first two rounds, the original competitiveness is maintained; for the nodes that served as cluster head nodes in one round in the first two rounds, their competitiveness is reduced; for the nodes that served as cluster head nodes continuously in the first two rounds, their competitiveness is greatly reduced.

[0184] Specifically, Figure 3 As shown, the fuzzy width learning model includes an input layer, a fuzzy feature layer, a feature enhancement layer and an output layer; the fuzzy feature layer includes several fuzzy subsystems;

[0185] In this embodiment, the input layer inputs the preprocessed node information; the purpose of the fuzzy feature layer is to extract the feature information of the data sample, and the FCM fuzzy c-means clustering algorithm (Fuzzy C-means algorithm) is used to fuzzy process the input data sample; the feature enhancement layer expands and enhances the feature information output by the fuzzy feature layer through nonlinear transformation to extract higher-dimensional feature representation; the calculation result of the output layer is the competitiveness value CV (Competitiveness value) of the node, which is used as a competitiveness indicator to measure the node to become a cluster head node;

[0186] The training and testing of the fuzzy width learning system network model according to the historical node data set includes the following steps:

[0187] S121, dividing the historical node data set into a training set and a test set;

[0188] Preferably, the ratio of training set to test set is 4:1.

[0189] S122, setting the number of fuzzy subsystems Q and the number of fuzzy rules K;

[0190] In this embodiment, the fuzzy feature layer includes Q fuzzy subsystems, K represents the number of fuzzy rules in each fuzzy subsystem, and reasonable setting of the values ​​of Q and K has an important impact on the performance and training effect of the fuzzy width learning model;

[0191] S123, initializing the coefficient weights of the fuzzy width learning model;

[0192] S124, inputting the training set into the fuzzy width learning model through the input layer;

[0193] S125, in the fuzzy feature layer, an intermediate matrix Z is generated through the fuzzy subsystems of the fuzzy width learning model and the training set. n and the feature node matrix F n , where n is the total number of training samples in the training set;

[0194] S126, the intermediate matrix Z n Each node feature value in is input into the feature enhancement layer, and several enhanced nodes H are obtained by enhanced feature processing. m , according to the characteristic node matrix F n and the plurality of enhanced nodes H m Obtain a target value Y, and obtain a weight matrix W based on the target value Y and pseudo-inverse fast calculation;

[0195] S127. In the output layer, obtaining preliminary parameters of the fuzzy width learning model according to the weight matrix W;

[0196] S128, inputting the test set into the fuzzy width learning model, calculating the error, wherein the error includes the error rate ERR and the mean absolute percentage error MAPE;

[0197] S129. After optimizing the number of fuzzy subsystems Q and the number of fuzzy rules K according to the error rate ERR and the mean absolute percentage error MAPE, the fuzzy width learning model is iteratively trained to optimize the preliminary parameters to obtain the optimal model parameters, thereby completing the training of the fuzzy width learning model.

[0198] In this embodiment, the fuzzy width learning model is trained multiple times, and the values ​​of Q and K are grid searched within a certain range in order to find the most suitable and optimal model.

[0199] Specifically, each training sample in the training set is N×M dimensional data, where N represents the number of groups of the training samples and M represents the dimension of feature information of the training samples, which is specifically expressed as:

[0200] X=(x1,x2,...,x s ,...,x N ) T ∈R N×M

[0201] where x s =(x s1 ,x s2 ,...,x sm ,...,x sM ), s=1,2,…,N;

[0202] And / or, the setting of the number of fuzzy subsystems Q and the number of fuzzy rules K specifically includes:

[0203] In the qth fuzzy subsystem, set K q The fuzzy rules conform to the following representation:

[0204] If x s1 is and x s2 is … and x sM is

[0205] but

[0206] Where k = 1, 2, ..., K q , x sM represents one of the training samples in the training set, represents the preset fuzzy set in the fuzzy subsystem, is the adjustable parameter of the model rule, u qg Represents the training sample g in the training set for the fuzzy rule K q The membership degree of is the intermediate matrix Z n The output of the qth fuzzy subsystem, where u qg It can be expressed as:

[0207]

[0208] Among them, c q 、c k represents the cluster center corresponding to the fuzzy rule;

[0209] Then the training sample xs The output Z of the qth fuzzy subsystem sq It is expressed as:

[0210]

[0211] in, The output Z of the qth fuzzy subsystem is sq A single intermediate value The weight of the training set is the intermediate matrix Z of the combined output of the Q fuzzy subsystems n It is expressed as:

[0212] Z n =(Z1,Z2,…,Z n )

[0213] The defuzzified output F of the fuzzy feature layer sq It is expressed as:

[0214]

[0215] in, is an adjustable parameter;

[0216] The qth fuzzy subsystem defuzzifies the output F q for:

[0217]

[0218] in, diag{.} is a diagonal matrix;

[0219] The output F of the fuzzy feature layer n It is expressed as:

[0220]

[0221] And / or, the intermediate matrix Z n Each node feature value in is input into the feature enhancement layer, and several enhanced nodes H are obtained by enhanced feature processing. m ,include:

[0222] The intermediate matrix Z n Each feature node value Z in I The value H mapped to the enhanced node J :

[0223]

[0224] Among them, ξ J (.) is the mapping function, is a random weight, is the bias term;

[0225] Then the number of enhanced nodes H of the m features m The output is represented as:

[0226] H m =(H1,H2,…,H m )

[0227] And / or, according to the characteristic node matrix F n and the plurality of enhanced nodes H m Get the target value Y, including:

[0228] The target value Y output by the output layer is:

[0229]

[0230] Among them, W f is the weight coefficient from the fuzzy feature layer to the output layer, W h is the weight coefficient from the feature enhancement layer to the output layer;

[0231] And / or, the weight matrix W is obtained by fast calculation based on the target value Y and pseudo-inverse, specifically according to the following formula:

[0232] W=(Bω,H m ) + Y

[0233] Among them, (Bω,H m ) + =((Bω,H m ) T (Bω,H m )) -1 (Bω,H m ) T ;

[0234] And / or, the calculation formula of the mean absolute percentage error MAPE is expressed as:

[0235]

[0236] Among them, y v is the true value of the vth intermediate target of the target value Y, for y v The predicted value of

[0237] And / or, the calculation formula of the error rate ERR is expressed as:

[0238]

[0239] Among them, E baselineis the benchmark error value of the fuzzy width learning model, E model Represents the error value of the current training process of the fuzzy width learning model.

[0240] S130, generating a plurality of nodes and corresponding node information of the wireless sensor network, pre-processing the plurality of node information and inputting the pre-processed node information into the trained fuzzy width learning model, and calculating and obtaining the competitiveness value of the node corresponding to each node information through the fuzzy width learning model;

[0241] S140, acquiring properties of the wireless sensor network, and acquiring an optimal number of cluster heads of the wireless sensor network according to the properties of the wireless sensor network;

[0242] Specifically, the attributes of the wireless sensor network include the total number of nodes in the wireless sensor network, the range of the wireless sensor network, and the average distance from each of the nodes in the wireless sensor network to a base station BS;

[0243] The obtaining the optimal number of cluster heads of the wireless sensor network according to the property of the wireless sensor network includes:

[0244] The optimal number of cluster heads k opt The specific formula is as follows:

[0245]

[0246] Among them, N w is the total number of nodes in the wireless sensor network, P is the range of the wireless sensor network, d toBS is the average distance from each of the nodes in the wireless sensor network to the base station BS;

[0247] The average distance from each node in the wireless sensor network to the base station BS is obtained by the following formula:

[0248]

[0249] In this embodiment, the optimal number of cluster heads refers to the most appropriate number of cluster heads in the current wireless sensor network environment. Too many cluster heads will lead to additional communication consumption, and too few cluster heads will lead to excessive burden on cluster heads and rapid energy consumption. To determine the optimal number of cluster heads, it is necessary to calculate according to the scale, node density and energy consumption of the wireless sensor network;

[0250] S150, selecting a number of cluster head nodes according to the competitiveness value of each of the nodes, wherein the number of the selected cluster head nodes is equal to the number of the optimal cluster heads; the nodes not selected as cluster head nodes are set as ordinary nodes;

[0251] Specifically, the selecting of several cluster head nodes according to the competitiveness value of each node includes:

[0252] Sort the competitiveness values ​​of all the nodes from high to low, select the nodes with high competitiveness as cluster head nodes, and the remaining nodes as ordinary nodes;

[0253] And / or, constructing a plurality of cluster groups in the wireless sensor network for data transmission based on the plurality of cluster head nodes and the plurality of common nodes, including:

[0254] The cluster group is constructed by a second cluster head strategy, a secondary cluster head strategy and a minimum energy consumption criterion strategy;

[0255] The second cluster head strategy is to select a node with the largest competitiveness value other than the cluster head node from the constructed cluster group to serve as the second cluster head, and the second cluster head assists the cluster head node in transmission work;

[0256] In this embodiment, the second cluster head strategy can effectively balance the energy consumption within the cluster;

[0257] The secondary cluster head strategy is to select a node in the cluster group that can replace the cluster head node as the secondary cluster head node. When the cluster head node fails or cannot work normally, the secondary cluster head node serves as the cluster head node to complete the work of the cluster head node.

[0258] In this embodiment, the secondary cluster head strategy provides a fault-tolerant mechanism for the wireless sensor network to avoid the impact of single-point cluster head failure on network operation;

[0259] The minimum energy consumption criterion strategy is to calculate the indirect energy consumption E of data transmitted indirectly from the ordinary node to the base station BS through the cluster head node when the ordinary node in the wireless sensor network transmits data. idt , and calculate the direct energy consumption E of data transmitted directly from the ordinary node to the base station BS dt , and compare the indirect energy consumption E through the cost function idt and the direct energy consumption E dt The size of , selects the path with the lowest energy consumption as the transmission path;

[0260] Specifically, the indirect energy consumption and / or direct energy consumption is affected by the following parameters:

[0261] The distance a from the common node to the cluster head node is: E idt ∝a;

[0262] The distance b from the cluster head node to the base station BS is: idt ∝b;

[0263] The distance c from the common node to the base station BS is: dt ∝c;

[0264] Then the indirect energy consumption E idt The specific formula is as follows:

[0265] E idt =l·[(1+μ)E elec +∈ fs (a 2 +μb 2 )]

[0266] Wherein, μ represents the preset data aggregation rate;

[0267] Among them, E elec represents the energy consumption coefficient of the path, ∈ fs represents the power amplification factor of the existing free space model;

[0268] The direct energy consumption E dt The specific formula is as follows:

[0269] E dt = l·(E elec +∈ fs c 2 )

[0270] Then the indirect energy consumption E idt and the direct energy consumption E dt The energy consumption difference △E is obtained by the following formula:

[0271] △E=E idt -E dt = l·[μE elec +ε fs (a 2 +μb 2 -c 2 )]

[0272] When △E>0, it means E dt The value is smaller, and the ordinary node selects data to be directly transmitted from the ordinary node to the base station BS as the transmission path for data transmission; when △E<0, it means that E idt The value is smaller, and the common node data is indirectly transmitted from the common node through the cluster head node to the base station BS as a transmission path for data transmission.

[0273] In this embodiment, the minimum energy consumption criterion strategy can further improve network performance and life cycle by optimizing the transmission path and reducing the overall energy consumption;

[0274] Parameter a means that the farther the ordinary node is from the cluster head node, the greater the energy consumed by indirect data transmission through the cluster head, so the non-cluster head node preferentially joins the cluster close to it;

[0275] Parameter b indicates that the farther the cluster head node is from the base station BS, the more energy the cluster head node needs to consume to transmit data packets to the base station BS;

[0276] Parameter c indicates that the closer the common node is to the base station BS, the less energy is consumed in transmitting data to the base station BS, and direct transmission to the base station BS can be considered.

[0277] S160: construct a plurality of cluster groups in the wireless sensor network according to the plurality of cluster head nodes and the plurality of common nodes for data transmission.

[0278] Specifically, the method further includes:

[0279] After the cluster group is built, the base station BS sends a confirmation message BS_info message to each node in the wireless sensor network, including the ID of the cluster head node, the coordinates of the cluster head node and the cluster group to which the cluster head node belongs; after receiving the confirmation message, each node starts working and enters the data collection and transmission stage;

[0280] In the data collection and transmission stage, the cluster head node aggregates the collected data; wherein, the data aggregation adopts the IA data aggregation model, and the cluster head node aggregates and forwards the received data packet length according to the data aggregation rate, and the specific calculation formula is as follows:

[0281] L agg =L r +μ×L r ×N p

[0282] Among them, L agg Indicates the length of the data packet after aggregation, L r represents the length of the received data packet, μ represents the data aggregation rate, N p Indicates the number of packets received.

[0283] In this embodiment, the above operation can take into account that the data sensed by the adjacent nodes have similarity or overlap, and the cluster head node aggregates the collected data to reduce the redundancy between the data.

[0284] Based on the above methods and steps, this application completes the simulation experiment of Example 1, which specifically includes:

[0285] In Example 1, 5000 historical nodes and corresponding historical node information are collected for training and testing the fuzzy width learning model. Subsequently, 200 sensor nodes are set up in a simulated wireless sensor network environment, randomly distributed in a 100m*100m scenario. Each node initially has the same initial energy of 0.5J, and after deployment, the position cannot be moved and the energy cannot be replenished; preferably, as shown in Table 1 of the fuzzy logic rules for competing cluster head nodes in this method.

[0286] Table 1 Fuzzy logic rules for competing cluster head nodes

[0287]

[0288]

[0289] If a node has more remaining energy, is closer to the base station BS, has a higher density of surrounding nodes, and has not served as a cluster head for multiple consecutive times, then the node's competitiveness in serving as a cluster head will be correspondingly greater.

[0290] In the node competitiveness calculation, the characteristic information obtained from the node information is used as an influencing factor, and a fuzzy width learning model is used to obtain the competitiveness value of the node as a cluster head node;

[0291] Table 2 shows the cluster head competitiveness values ​​calculated based on the fuzzy width learning model in the simulation experiment of Example 1.

[0292] Table 2 Computational node cluster head competitiveness values

[0293]

[0294] Preferably, in this embodiment 1, the optimal number of cluster heads can be obtained as 14 from the environment size of the simulation network being 100m*100m;

[0295] In the present embodiment 1, the optimal number of cluster heads is 14. Due to the second cluster head strategy, the number of generated cluster groups is only 7, and each cluster group includes 1 cluster head and 1 second cluster head.

[0296] The present application also includes Example 2. Compared with Example 1, Example 2 performs a simulation experiment on a network environment with randomly distributed energy;

[0297] The present application also includes Example 3. Compared with Example 1, Example 3 performs a simulation experiment in a network environment with a small amount of initial energy;

[0298] The present application also includes Example 4. Compared with Example 1, Example 3 performs a simulation experiment on a network environment of a mobile node situation;

[0299] In Example 2 and Example 3, the simulation parameter configuration is different from that of the Base Scenario, reflecting different performances in multiple scenarios, one is the uneven energy distribution, and the other is the small initial energy situation. The experiment uses Matlab simulation parameters to simulate the process of the real wireless sensor network environment, collects all node information and inputs it into the fuzzy width learning model, and optimizes the entire clustering process. In order to compare and analyze the applicability of dynamic nodes, Example 4 conducts simulation experiments in the mobile node situation.

[0300] Figure 4 The average residual energy performance of nodes in different embodiments of the present invention and the comparative method is shown; Figure 5 The fluctuation of the average residual energy performance of nodes in different embodiments of each method is shown, where FBLSC represents the present invention. The experimental results show that compared with other existing methods, the present invention has better performance fluctuation in different environments and stronger adaptability. Compared with the huge time overhead generated by complex methods, the present invention can greatly reduce the training time of the model through the fast training ability of the fuzzy width learning model, and ensure good accuracy. At the same time, it has incremental characteristics and shows good scalability.

[0301] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the technical solution of the present invention, and are not intended to limit the specific implementation methods of the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the claims of the present invention shall be included in the protection scope of the claims of the present invention.

Claims

1. A wireless sensor network energy efficiency competitive clustering method based on fuzzy width learning, characterized in that: The method comprises: Collecting some historical node information of the wireless sensor network, and preprocessing some of the historical node information to generate a historical node data set; Training and testing the fuzzy width learning model according to the historical node data set to obtain a trained fuzzy width learning model; Generate a number of nodes and corresponding node information of the wireless sensor network, pre-process the node information and input it into the trained fuzzy width learning model, and calculate and obtain the competitiveness value of each node corresponding to the node information through the fuzzy width learning model; Acquire the properties of the wireless sensor network, and acquire the optimal number of cluster heads of the wireless sensor network according to the properties of the wireless sensor network; According to the competitiveness value of each node, a number of cluster head nodes are selected, wherein the number of the selected cluster head nodes is equal to the number of the optimal cluster heads; nodes that are not selected as cluster head nodes are set as ordinary nodes; According to the plurality of cluster head nodes and the plurality of common nodes, a plurality of cluster groups in the wireless sensor network are constructed for data transmission.

2. According to claim 1, a wireless sensor network energy efficiency competitive clustering method based on fuzzy width learning is characterized in that: The preprocessing of the historical node information to generate a historical node data set includes: Performing data normalization processing on the historical node information to obtain historical node information of the same dimension; Performing missing value processing on a number of historical node information of the same dimension to obtain a number of complete historical node information; wherein the missing value processing is to obtain node information of which competitiveness value is missing in the number of node information of the same dimension, and inputting the node information of which competitiveness value is missing into the fuzzy logic system for processing to obtain the corresponding competitiveness value, and completing the information of the node of which competitiveness value is missing according to the obtained competitiveness value; The data format conversion process is performed on the complete historical node information to obtain a historical node data set.

3. According to claim 1, a wireless sensor network energy efficiency competitive clustering method based on fuzzy width learning is characterized in that: The competitiveness value depends on influencing factors, which include one or more of the relative remaining energy Energy_Re(i) of the node, the relative distance Dist_BS(i) from the node to the base station BS, the neighbor node density Nei_Dens(i) of the node and the historical cluster head state CH_State(i) of the node.

4. According to claim 3, a wireless sensor network energy efficiency competitive clustering method based on fuzzy width learning is characterized in that: The relative remaining energy Energy_Re(i) of the node is obtained by the following formula: Energy res (i) represents the remaining energy of node i, Energy init (i) represents the initial energy of node i; and / or, The relative distance Dist_BS(i) from the node to the base station BS is obtained by the following formula: Wherein, d1, d2, d3, d4 represent the distances from the four boundary corners in the wireless sensor network scene to the base station BS, max(.) is the maximum value function, indicating the selection of the farthest distance from the base station BS among d1, d2, d3, d4; Dist BS (S i ) represents the Euclidean distance from node i to the base station BS, which is obtained by the following calculation formula: where x i ,y i Respectively represent the horizontal and vertical coordinates of node i, x BS ,y BS Respectively represent the horizontal and vertical coordinates of the base station BS; And / or, the neighbor node density Nei_Dens(i) of the node i is specifically obtained by the following formula: Among them, N w is the total number of nodes in the wireless sensor network; Neighbor(i) represents the number of neighbor nodes of the node i, which is specifically obtained by the following formula: Wherein, j represents other nodes j in the wireless sensor network except the node i, d ij represents the distance between the node i and the other node j, R t represents the maximum radius distance of the node i to transmit relevant information, δ(.) is the indicator function, when d ij ≤R t When , δ=1; otherwise, δ=0; And / or, the historical cluster head state CH_State(i) of the node is obtained by the following formula: Wherein, r represents the number of times the node i is continuously selected as the cluster head node.

5. According to claim 1, a wireless sensor network energy efficiency competitive clustering method based on fuzzy width learning is characterized in that: The fuzzy width learning model includes an input layer, a fuzzy feature layer, a feature enhancement layer and an output layer; the fuzzy feature layer includes several fuzzy subsystems; The training and testing of the fuzzy width learning system network model according to the historical node data set includes: Dividing the historical node data set into a training set and a test set; Setting the number Q of fuzzy subsystems and the number K of fuzzy rules; Initializing coefficient weights of the fuzzy width learning model; Inputting the training set into the fuzzy width learning model through the input layer; In the fuzzy feature layer, an intermediate matrix Z is generated by the fuzzy subsystems of the fuzzy width learning model and the training set. n and the feature node matrix F n , where n is the total number of training samples in the training set; The intermediate matrix Z n Each node feature value in is input into the feature enhancement layer, and several enhanced nodes H are obtained by enhanced feature processing. m , according to the characteristic node matrix F n and the plurality of enhanced nodes H m Obtain a target value Y, and obtain a weight matrix W based on the target value Y and pseudo-inverse fast calculation; In the output layer, preliminary parameters of the fuzzy width learning model are obtained according to the weight matrix W; Input the test set into the fuzzy width learning model to calculate the error, wherein the error includes the error rate ERR and the mean absolute percentage error MAPE; After optimizing the number of fuzzy subsystems Q and the number of fuzzy rules K according to the error rate ERR and the mean absolute percentage error MAPE, the fuzzy width learning model is iteratively trained, the preliminary parameters are optimized to obtain the optimal model parameters, and the training of the fuzzy width learning model is completed.

6. According to claim 5, a wireless sensor network energy efficiency competitive clustering method based on fuzzy width learning is characterized in that: Each training sample in the training set is N×M dimensional data, where N represents the number of groups of the training samples and M represents the dimension of the feature information of the training samples, which is specifically expressed as: X=(x1,x2,…,x s ,…,x N ) T ∈R N×M where x s =(x s1 ,x s2 ,...,x sm ,...,x sM ), s=1,2,…,N; And / or, the setting of the number of fuzzy subsystems Q and the number of fuzzy rules K specifically includes: In the qth fuzzy subsystem, set K q The fuzzy rules conform to the following representation: if but Where k = 1, 2, ..., K q , x sm represents one of the training samples in the training set, represents the preset fuzzy set in the fuzzy subsystem, is the adjustable parameter of the model rule, u qg Represents the training sample g in the training set for the fuzzy rule K q The membership degree of is the intermediate matrix Z n The output of the qth fuzzy subsystem, where u qg It can be expressed as: Among them, c q 、c k represents the cluster center corresponding to the fuzzy rule; Then the training sample x s The output Z of the qth fuzzy subsystem sq It is expressed as: in, The output Z of the qth fuzzy subsystem is sq A single intermediate value The weight of Then the training set is the intermediate matrix Z of the combined output of Q fuzzy subsystems n It is expressed as: Z n =(Z1,Z2,…,Z n ) The defuzzified output F of the fuzzy feature layer sq It is expressed as: in, is an adjustable parameter; The qth fuzzy subsystem defuzzifies the output F q for: in, diag{.} is a diagonal matrix; The output F of the fuzzy feature layer n It is expressed as: And / or, the intermediate matrix Z n Each node feature value in is input into the feature enhancement layer, and several enhanced nodes H are obtained by enhanced feature processing. m ,include: The intermediate matrix Z n Each feature node value Z in I The value H mapped to the enhanced node J : Among them, ξ J (.) is the mapping function, is a random weight, is the bias term; Then the number of enhanced nodes H of the m features m The output is represented as: H m =(H1,H2,…,H m ) And / or, according to the characteristic node matrix F n and the plurality of enhanced nodes H m Get the target value Y, including: The target value Y output by the output layer is: Among them, W f is the weight coefficient from the fuzzy feature layer to the output layer, W h is the weight coefficient from the feature enhancement layer to the output layer; And / or, the weight matrix W is obtained by fast calculation based on the target value Y and pseudo-inverse, specifically according to the following formula: W=(Bω,H m ) + Y Among them, (Bω, H m ) + =((Bω,H m ) T (Bω,H m )) -1 (Bω,H m ) T ; And / or, the calculation formula of the mean absolute percentage error MAPE is expressed as: Among them, y v is the true value of the vth intermediate target of the target value Y, for y v The predicted value of And / or, the calculation formula of the error rate ERR is expressed as: Among them, E baseline is the benchmark error value of the fuzzy width learning model, E model Represents the error value of the current training process of the fuzzy width learning model.

7. According to claim 1, a wireless sensor network energy efficiency competitive clustering method based on fuzzy width learning is characterized in that: The attributes of the wireless sensor network include the total number of nodes in the wireless sensor network, the range of the wireless sensor network, and the average distance from each of the nodes in the wireless sensor network to a base station BS; The obtaining the optimal number of cluster heads of the wireless sensor network according to the property of the wireless sensor network includes: The optimal number of cluster heads k opt The specific formula is as follows: Among them, N w is the total number of nodes in the wireless sensor network, P is the range of the wireless sensor network, d toBS is the average distance from each of the nodes in the wireless sensor network to the base station BS; The average distance from each node in the wireless sensor network to the base station BS is obtained by the following formula:

8. A wireless sensor network energy efficiency competitive clustering method based on fuzzy width learning according to any one of claims 1 to 7, characterized in that: The step of selecting a number of cluster head nodes according to the competitiveness value of each node includes: Sort the competitiveness values ​​of all the nodes from high to low, select the nodes with high competitiveness as cluster head nodes, and the remaining nodes as ordinary nodes; and / or, The constructing a plurality of cluster groups in the wireless sensor network according to the plurality of cluster head nodes and the plurality of common nodes for data transmission includes: The cluster group is constructed by a second cluster head strategy, a secondary cluster head strategy and a minimum energy consumption criterion strategy; The second cluster head strategy is to select a node with the largest competitiveness value other than the cluster head node from the constructed cluster group to serve as the second cluster head, and the second cluster head assists the cluster head node in transmission work; The secondary cluster head strategy is to select a node in the cluster group that can replace the cluster head node as the secondary cluster head node. When the cluster head node fails or cannot work normally, the secondary cluster head node serves as the cluster head node to complete the work of the cluster head node. The minimum energy consumption criterion strategy is to calculate the indirect energy consumption E of data transmitted indirectly from the ordinary node to the base station BS through the cluster head node when the ordinary node in the wireless sensor network transmits data. idt , and calculate the direct energy consumption E of data transmitted directly from the ordinary node to the base station BS dt , and compare the indirect energy consumption E through the cost function idt and the direct energy consumption E dt The size of , selects the path with the lowest energy consumption as the transmission path.

9. The method for energy-efficient competitive clustering of wireless sensor networks based on fuzzy width learning according to claim 8, characterized in that: The indirect energy consumption and / or direct energy consumption are influenced by the following parameters: The distance a from the common node to the cluster head node is: E idt ∝a; The distance b from the cluster head node to the base station BS is: idt ∝b; The distance c from the common node to the base station BS is: dt ∝c; Then the indirect energy consumption E idt The specific formula is as follows: And idt =l·[(1+μ)E elec +∈ fs (to 2 +μb 2 )] Wherein, μ represents the preset data aggregation rate; Among them, E elec represents the energy consumption coefficient of the path, ∈ fs represents the power amplification factor of the existing free space model; The direct energy consumption E dt The specific formula is as follows: AND dt =l·(E elec +∈ fs c 2 ) Then the indirect energy consumption E idt and the direct energy consumption E dt The energy consumption difference △E is obtained by the following formula: △E=E idt -E dt =l·[μE elec +e fs (a 2 +μb 2 -c 2 )] When △E>0, it means E dt The value is smaller, and the ordinary node selects data to be directly transmitted from the ordinary node to the base station BS as the transmission path for data transmission; when △E<0, it means that E idt The value is smaller, and the common node data is indirectly transmitted from the common node through the cluster head node to the base station BS as a transmission path for data transmission.

10. A wireless sensor network energy efficiency competitive clustering method based on fuzzy width learning according to any one of claims 1 to 7, characterized in that: The method further comprises: After the cluster group is built, the base station BS sends a confirmation message BS_info message to each node in the wireless sensor network, including the ID of the cluster head node, the coordinates of the cluster head node and the cluster group to which the cluster head node belongs; after receiving the confirmation message, each node starts working and enters the data collection and transmission stage; In the data collection and transmission stage, the cluster head node aggregates the collected data; wherein, the data aggregation adopts the IA data aggregation model, and the cluster head node aggregates and forwards the received data packet length according to the data aggregation rate, and the specific calculation formula is as follows: L agg =L r +μ×L r ×N p Among them, L agg Indicates the length of the data packet after aggregation, L r represents the length of the received data packet, μ represents the data aggregation rate, N p Indicates the number of packets received.

Images