A method, device and medium for intelligent and low-power WSN data fusion

Abstract

The present application discloses an intelligent low-power WSN data fusion method, device and medium, which relates to the field of data processing. The method uses chaotic mapping to initialize the initial fireworks of the weight matrix and the initial fireworks of the threshold matrix of the BP neural network, so that the initial position distribution of the population is more uniform; the training error is calculated according to the training sample, the initial weight matrix and the initial threshold matrix; whether the training error is qualified is determined; if so, the initial weight matrix and the initial threshold matrix are used as the current weight matrix and the current threshold matrix of the BP neural network in the cluster head for data fusion; if not, the fireworks of the weight matrix and the fireworks of the threshold matrix are updated by FWA, and the step of calculating the training error according to the training sample, the initial weight matrix and the initial threshold matrix is ​​entered. It can be seen that the method uses FWA to update the fireworks of the weight matrix and the threshold matrix to enrich the diversity of the population, and therefore, it can solve the current shortcomings of being prone to local optimality and slow speed.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to an intelligent low-power WSN data fusion method, device and medium. Background Art

[0002] Wireless Sensor Network (WSN) is an intelligent private network system composed of a large number of low-power, multi-functional, low-cost sensor nodes deployed in the monitoring area. The nodes use multi-hop self-organizing wireless communication to transmit information, collaborate to complete specific functions, collect various information in the monitoring area in real time and transmit it to the observer. WSN usually adopts high-density deployment. The data collected by adjacent nodes has a lot of redundancy. If a single node sends data to the aggregation node alone, it will waste network bandwidth, consume more energy, and reduce the efficiency of message collection. Therefore, WSN requires some nodes to have data fusion function to fully utilize the node's own storage and computing capabilities.

[0003] The current WSN data fusion method uses approximate fusion algorithm for data fusion, which is easy to fall into local optimum and has low accuracy.

[0004] It can be seen that how to solve the problem that WSN data fusion easily falls into local optimality is an urgent problem to be solved by those skilled in the art. Summary of the invention

[0005] The purpose of this application is to provide an intelligent low-power WSN data fusion method, device and medium to solve the problem that WSN data fusion is prone to fall into local optimality and improve the accuracy of data fusion.

[0006] In order to solve the above technical problems, the present application provides an intelligent low-power WSN data fusion method, including:

[0007] Initializing the initial fireworks of the weight matrix and the initial fireworks of the threshold matrix of the BP neural network using chaotic mapping to obtain the initial weight matrix and the initial threshold matrix;

[0008] Obtain training samples, and calculate training errors based on the training samples, the initial weight matrix, and the initial threshold matrix;

[0009] Determine whether the training error is qualified;

[0010] If yes, the initial weight matrix and initial threshold matrix are used as the current weight matrix and current threshold matrix of the BP neural network in the cluster head for data fusion;

[0011] If not, use FWA to update the fireworks of the weight matrix and the fireworks of the threshold matrix, update the initial weight matrix and the initial threshold matrix according to the fireworks of the weight matrix and the fireworks of the threshold matrix, and enter the step of calculating the training error according to the training samples, the initial weight matrix and the initial threshold matrix.

[0012] Preferably, if the training error is unqualified, the method further includes:

[0013] Add one to the number of iterations and determine whether the number of iterations reaches the iteration threshold;

[0014] If it is reached, the step of using the initial weight matrix and the initial threshold matrix as the current weight matrix and the current threshold matrix of the BP neural network in the cluster head for data fusion is entered;

[0015] If not reached, the step of using FWA to update the fireworks of the weight matrix and the fireworks of the threshold matrix is ​​entered.

[0016] Preferably, the fireworks of updating the weight matrix and the fireworks of updating the threshold matrix using FWA include:

[0017] The explosion sparks of the weight matrix and the explosion sparks of the threshold matrix are obtained through FWA;

[0018] The fireworks of the weight matrix are updated using the initial fireworks of the weight matrix and the explosion sparks of the weight matrix, and the fireworks of the threshold matrix are updated using the initial fireworks of the threshold matrix and the explosion sparks of the threshold matrix.

[0019] Preferably, after obtaining the explosion sparks of the weight matrix and the explosion sparks of the threshold matrix through FWA, and before updating the fireworks of the weight matrix using the initial fireworks of the weight matrix and the explosion sparks of the weight matrix, the method further includes:

[0020] Displace the explosion spark of the weight matrix and the explosion spark of the threshold matrix.

[0021] Preferably, before using the initial fireworks of the weight matrix and the explosion sparks of the weight matrix to update the fireworks of the weight matrix, and after shifting the explosion sparks of the weight matrix and the explosion sparks of the threshold matrix, the method further includes:

[0022] The initial fireworks of the weight matrix and the initial fireworks of the threshold matrix are processed by Gaussian mutation to obtain Gaussian mutation fireworks of the weight matrix and Gaussian mutation fireworks of the threshold matrix;

[0023] Correspondingly, using the initial fireworks of the weight matrix and the explosion sparks of the weight matrix to update the fireworks of the weight matrix, and using the initial fireworks of the threshold matrix and the explosion sparks of the threshold matrix to update the fireworks of the threshold matrix includes:

[0024] The fireworks of the weight matrix are updated using the initial fireworks of the weight matrix, the explosion sparks of the weight matrix, and the Gaussian mutation fireworks of the weight matrix, and the fireworks of the threshold matrix are updated using the initial fireworks of the threshold matrix, the explosion sparks of the threshold matrix, and the Gaussian mutation fireworks of the threshold matrix.

[0025] Preferably, before using the initial fireworks of the weight matrix, the explosion sparks of the weight matrix and the Gaussian variation fireworks of the weight matrix to update the fireworks of the weight matrix, and after obtaining the Gaussian variation fireworks of the weight matrix and the Gaussian variation fireworks of the threshold matrix, the method further includes:

[0026] Determine whether there are weight matrices and threshold matrices beyond the feasible domain in the explosion sparks of the weight matrix and the explosion sparks of the threshold matrix;

[0027] If it exists, the weight matrix and threshold matrix beyond the feasible domain are mapped into the feasible domain through modulo operation.

[0028] Preferably, using the initial weight matrix and the initial threshold matrix as the current weight matrix and the current threshold matrix of the BP neural network in the cluster head for data fusion includes:

[0029] Receive monitoring data sent by each member node in the cluster;

[0030] Extracting characteristic values ​​of the monitoring data to obtain characteristic values ​​of the monitoring data;

[0031] The initial weight matrix and the initial threshold matrix are used as the current weight matrix and the current threshold matrix of the BP neural network in the cluster head, and the BP neural network is called to calculate the eigenvalues ​​to obtain the data fusion result.

[0032] In order to solve the above technical problems, the present application also provides an intelligent low-power WSN data fusion device, including:

[0033] An initialization module, used for initializing the initial fireworks of the weight matrix and the initial fireworks of the threshold matrix of the BP neural network by using chaotic mapping to obtain an initial weight matrix and an initial threshold matrix;

[0034] An acquisition module is used to acquire training samples and calculate training errors based on the training samples, an initial weight matrix and an initial threshold matrix;

[0035] The judgment module is used to judge whether the training error is qualified; if so, the fusion module is triggered; if not, the update module is triggered;

[0036] A fusion module is used to fuse the initial weight matrix and the initial threshold matrix as the current weight matrix and the current threshold matrix of the BP neural network in the cluster head;

[0037] The updating module is used to update the fireworks of the weight matrix and the fireworks of the threshold matrix by using FWA, update the initial weight matrix and the initial threshold matrix according to the fireworks of the weight matrix and the fireworks of the threshold matrix, and trigger the acquisition module.

[0038] In order to solve the above technical problems, the present application also provides an intelligent low-power WSN data fusion device, including:

[0039] Memory for storing computer programs;

[0040] The processor is used to implement the steps of the above-mentioned intelligent low-power WSN data fusion method when executing the computer program.

[0041] In order to solve the above technical problems, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above intelligent low-power WSN data fusion method are implemented.

[0042] The present application provides an intelligent low-power WSN data fusion method, which uses chaotic mapping to initialize the initial fireworks of the weight matrix of the BP neural network and the initial fireworks of the threshold matrix to obtain the initial weight matrix and the initial threshold matrix; obtain training samples, and calculate the training error based on the training samples, the initial weight matrix and the initial threshold matrix; judge whether the training error is qualified; if so, use the initial weight matrix and the initial threshold matrix as the current weight matrix and the current threshold matrix of the BP neural network in the cluster head for data fusion; if not, use FWA to update the fireworks of the weight matrix and the fireworks of the threshold matrix, update the initial weight matrix and the initial threshold matrix according to the fireworks of the weight matrix and the fireworks of the threshold matrix, and enter the step of calculating the training error according to the training samples, the initial weight matrix and the initial threshold matrix. It can be seen that the method uses chaotic mapping to initialize the initial fireworks of the weight matrix and the initial fireworks of the threshold matrix of the BP neural network, so that the initial position distribution of the population is more uniform. In addition, the method uses FWA to update the fireworks of the weight matrix and the fireworks of the threshold matrix to enrich the diversity of the population, so it can solve the current shortcomings of being prone to local optimality and slow speed.

[0043] In addition, the present application also provides an intelligent low-power WSN data fusion device and medium, which have the same beneficial effects as the above-mentioned intelligent low-power WSN data fusion method. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. 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 paying any creative work.

[0045] Figure 1 A flowchart of an intelligent low-power WSN data fusion method provided by this application;

[0046] Figure 2 A structural diagram of an intelligent low-power WSN data fusion device provided in this application;

[0047] Figure 3 This is a structural diagram of another intelligent low-power WSN data fusion device provided in this application. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions 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 are within the scope of protection of this application.

[0049] The core of this application is to provide an intelligent low-power WSN data fusion method, device and medium, which are used to solve the problem that WSN data fusion is prone to fall into local optimality and improve the accuracy of data fusion.

[0050] In order to enable those skilled in the art to better understand the present application, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0051] Figure 1 A flowchart of an intelligent low-power WSN data fusion method provided in this application, such as Figure 1 As shown, the method includes:

[0052] S1: Initializing the initial fireworks of the weight matrix and the initial fireworks of the threshold matrix of the BP neural network using chaotic mapping to obtain the initial weight matrix and the initial threshold matrix.

[0053] In the Fireworks Algorithm (FWA), fireworks are used as feasible answers to the solution space in the optimization problem. The explosion process of fireworks producing a certain number of sparks is the process of searching for the neighborhood in the solution space. The fitness function values ​​of fireworks and sparks can be screened through multiple iterations to quickly find the optimal solution for fireworks. In this embodiment, first, the basic parameters of the fireworks algorithm should be initialized. Assume that the population size is M, the search space dimension is D, and the current position of the i-th firework in the search space is X. i =(X i1 , X i2 , …, X ij , …, X iD)(1≤i≤M, 1≤j≤D), where X i The specific calculation of is shown in formula (1):

[0054] X i =rand(X max -X min )+X min (1)

[0055] In formula (1), [X min , X max ] is the boundary of the search space, and rand is a random number uniformly distributed between [0, 1].

[0056] In FWA, the initial distribution of the fireworks population has a great influence on the optimal solution. Specifically, the more uniform the initial population distribution in the feasible domain, the greater the probability of finding the optimal solution. Therefore, in order to improve the uniformity of the distribution of the initial state of the fireworks population, this embodiment introduces a chaotic mapping to initialize the fireworks population. Common chaotic mapping models include Logistic mapping, Hemon mapping, and Tent mapping. Among them, the Tent mapping has a simple structure, good chaotic performance, more uniform distribution, and good ergodicity. For details, see formula (2):

[0057]

[0058] In formula (2), H j is the chaotic variable, b is the system parameter, when H j When ∈(0, 1) and b∈(0, 1), the system is in a chaotic state.

[0059] In the initial stage of the fireworks population, the chaotic sequence generated by the Tent chaotic graph is used to replace the random variable rand in formula (1), then X ij+1 The calculation of is shown in formula (3):

[0060] X ij+1 =X min +H j ·(X max -X min ) (3)

[0061] Among them, X ij+1 represents the position of firework i in the j+1 dimension. According to formula (3), since H j Evenly distributed on (0, 1), so X determined by formula (3) ij+1 The positions are more evenly distributed in the feasible region.

[0062] S2: Obtain training samples and calculate training errors based on the training samples, the initial weight matrix and the initial threshold matrix.

[0063] For the WSN network model, for the convenience of discussion, this embodiment assumes that the WSN has the following characteristics:

[0064] (1) Sensor nodes are randomly distributed in the monitoring area, the deployment locations are static or slightly moving, and all nodes have unique IDs;

[0065] (2) The initial energy of sensor nodes is the same and cannot be replenished, and the communication and processing capabilities of all sensor nodes are the same;

[0066] (3) There is only one sink node in the WSN, which is deployed in the monitoring area;

[0067] (4) The energy of the aggregation node is sufficient and can be replenished;

[0068] (5) Any sensor node can determine its own location by the strength of its transmission power.

[0069] Among them, assumptions (1)-(4) are the traditional settings of the network model in WSN. Since cluster heads may be replaced, member nodes may be changed or die in WSN, which will affect the topology of WSN, assumption (5) can be used to enable sensor nodes to obtain their own location information to reduce the impact on WSN. Therefore, assumption (5) is crucial.

[0070] The topological structure of the BP neural network includes an input layer, a hidden layer, and an output layer, which is a three-layer network model. It can be understood that for a multi-layer neural network, the output of the previous layer is the input of the next layer. The specific working process of the BP neural network includes two stages: the first stage is the forward propagation of the signal from the input layer through the hidden layer to the output layer; the second stage is the reverse propagation of the error, from the output layer through the hidden layer to the input layer, which requires constantly changing the weight matrix and threshold matrix from the hidden layer to the output layer and from the input layer to the hidden layer. Among them, the hidden layer adopts the tangent double-bend transfer function, as shown in formula (4), and the output layer can choose a linear function as the operation function, as shown in formula (5):

[0071]

[0072] f 2 (x) = kx + l (5)

[0073] In formulas (4) and (5), x is the function variable, k is the weight matrix, and l is the threshold matrix.

[0074] In a specific implementation, the training sample is the input data of the BP neural network input layer. After the training sample is input, the output value of the BP neural network can be calculated according to the training sample, the initial weight matrix and the initial threshold matrix, and the calculated output value is compared with the output value predicted according to the previous sample to calculate the training error.

[0075] S3: Determine whether the training error is qualified; if so, proceed to S4; if not, proceed to S5.

[0076] After the training error is calculated, it can be determined whether the training error is qualified according to whether the training error meets the threshold condition. Specifically, if the training error is not less than the preset threshold, it can be determined that the training error is unqualified, that is, the initial weight matrix and initial threshold matrix corresponding to the training error are not the optimal solution. At this time, step S5 is entered, and the fireworks of the weight matrix and the fireworks of the threshold matrix are updated by FWA, and the optimal solution of the weight matrix and the threshold matrix is ​​found through multiple iterations; and if the training error is less than the preset threshold, it can be determined that the training error is qualified, that is, the initial weight matrix and the initial threshold matrix corresponding to the training error are the optimal solution. At this time, step S4 is entered, and the initial weight matrix and the initial threshold matrix are used as the current weight matrix and the current threshold matrix of the BP neural network in the cluster head for data fusion. It can be understood that the preset threshold can be set according to the actual situation. In order to ensure the accuracy of the optimal solution obtained, the preset threshold should be as small as possible.

[0077] S4: The initial weight matrix and the initial threshold matrix are used as the current weight matrix and the current threshold matrix of the BP neural network in the cluster head for data fusion.

[0078] S5: Use FWA to update the fireworks of the weight matrix and the fireworks of the threshold matrix, update the initial weight matrix and the initial threshold matrix according to the fireworks of the weight matrix and the fireworks of the threshold matrix, and enter S2.

[0079] When the training error is unqualified, the fireworks of the weight matrix and the fireworks of the threshold matrix are updated using FWA. Specifically, the explosion process of FWA follows the law of nature, that is, after each firework explodes, a certain number of explosion sparks will be generated within a certain range. Therefore, the number of explosion sparks needs to be determined in each iteration to enrich the diversity of the population. It is understandable that high-quality fireworks will produce more sparks, closer distances, and better adaptability, while low-quality fireworks will produce fewer sparks, longer distances, and poorer adaptability. The calculation formula for generating explosion sparks is shown in formula (6):

[0080]

[0081] Among them, S i is the number of explosion sparks generated by firework i, S is the parameter that limits the number of explosion sparks, and fmax is the worst adaptation value of fireworks in the current situation, f(X i ) is the adaptation value of firework i, and ε is the smallest positive constant to avoid the denominator of formula (6) being 0.

[0082] In order to prevent good quality fireworks from producing too many explosion sparks and poor quality fireworks from producing too few explosion sparks, the number of explosion sparks can be controlled. The control equation is shown in formula (7):

[0083]

[0084] In formula (7), α and β represent constants that limit the overall range, m is a constant, and round is a rounding function.

[0085] After obtaining the number of explosion sparks of the fireworks, it is necessary to further determine the explosion amplitude of the fireworks. In FWA, fireworks with better fitness values ​​have smaller explosion amplitudes, while fireworks with worse fitness values ​​have larger explosion amplitudes. The smaller the fitness value, the better the fitness value, and the larger the fitness value, the worse the fitness value. Therefore, in the optimization process of FWA, the explosion range of fireworks with poor fitness values ​​can be reduced by controlling the explosion amplitude, thereby improving the convergence speed of FWA and accelerating the speed of finding the optimal value of fireworks. The explosion amplitude R of firework i i The calculation formula is shown in formula (8):

[0086]

[0087] Among them, R is the limit parameter of explosion amplitude, f min It represents the optimal fitness value of fireworks in the current situation. After the explosion, the feasible domain space includes fireworks and explosion sparks, which means that the update of the fireworks population is completed.

[0088] This embodiment provides an intelligent low-power WSN data fusion method, which uses chaotic mapping to initialize the initial fireworks of the weight matrix of the BP neural network and the initial fireworks of the threshold matrix to obtain the initial weight matrix and the initial threshold matrix; obtain training samples, and calculate the training error based on the training samples, the initial weight matrix and the initial threshold matrix; judge whether the training error is qualified; if so, use the initial weight matrix and the initial threshold matrix as the current weight matrix and the current threshold matrix of the BP neural network in the cluster head for data fusion; if not, use FWA to update the fireworks of the weight matrix and the fireworks of the threshold matrix, update the initial weight matrix and the initial threshold matrix according to the fireworks of the weight matrix and the fireworks of the threshold matrix, and enter the step of calculating the training error according to the training samples, the initial weight matrix and the initial threshold matrix. It can be seen that the method uses chaotic mapping to initialize the initial fireworks of the weight matrix and the initial fireworks of the threshold matrix of the BP neural network, so that the initial position distribution of the population is more uniform. In addition, the method uses FWA to update the fireworks of the weight matrix and the fireworks of the threshold matrix to enrich the diversity of the population, so that it can solve the current shortcomings of being prone to local optimality and slow speed.

[0089] On the basis of the above embodiment, if the training error fails multiple times, in order to reduce the number of iterations, this embodiment determines whether to terminate the iteration according to the number of iterations. This step includes:

[0090] Add one to the number of iterations and determine whether the number of iterations reaches the iteration threshold;

[0091] If reached, proceed to step S4;

[0092] If not reached, go to step S5.

[0093] Specifically, the number of iterations will be recorded when judging whether the training error is qualified. After each iteration process, i.e., the process of judging whether the training error is qualified, the recorded number of iterations will be increased by one, and each time the training error is judged to be unqualified, the recorded number of iterations will be compared with the preset iteration threshold. If the number of iterations does not reach the iteration threshold, step S5 is entered to continue iteration to find the optimal solution of the weight matrix and the threshold matrix; if the number of iterations reaches the iteration threshold, the iteration is terminated, and step S4 is entered to use the initial weight matrix and initial threshold matrix corresponding to the current training error as the current weight matrix and current threshold matrix of the BP neural network in the cluster head for data fusion. It should be noted that the iteration threshold is a positive integer, and the value can be taken according to the actual situation in the specific implementation, and this embodiment does not limit this.

[0094] On the basis of the above embodiment, in order to enrich the population diversity, this embodiment updates the fireworks of the weight matrix and the fireworks of the threshold matrix through FWA. This step includes:

[0095] The explosion sparks of the weight matrix and the explosion sparks of the threshold matrix are obtained through FWA;

[0096] The fireworks of the weight matrix are updated using the initial fireworks of the weight matrix and the explosion sparks of the weight matrix, and the fireworks of the threshold matrix are updated using the initial fireworks of the threshold matrix and the explosion sparks of the threshold matrix.

[0097] In FWA, the explosion sparks corresponding to the fireworks can be obtained through the explosion process of fireworks. At this time, the feasible domain space of the weight matrix includes the initial fireworks and explosion sparks of the weight matrix, and the feasible domain space of the threshold matrix includes the initial fireworks and explosion sparks of the threshold matrix. It is necessary to select the individuals with the best fitness value from the feasible domain space through the selection operation to enter the next iteration, and the remaining individuals are kept M-1 by the roulette rule to enter the next iteration, that is, the population size of the fireworks after each update is always kept at M. It should be noted that in this embodiment, the individuals with the best fitness value selected by the selection operation are the initial weight matrix and the initial threshold matrix.

[0098] In FWA, the Euclidean distance is usually used to measure the distance between two objects. The sum of the Euclidean distances from firework i to other fireworks is Z(X i ) The calculation formula is shown in formula (9):

[0099]

[0100] In the calculation of Z(X i ), the roulette rule can be used to select the remaining individuals for the next iteration, and the probability of firework i being selected is P(X i ) is calculated by referring to formula (10):

[0101]

[0102] According to formula (10), if the current individual is far away from other sparks, the probability of the individual being selected will increase, thus ensuring the diversity of the fireworks population.

[0103] In this embodiment, in each iteration, the fireworks of the weight matrix and the fireworks of the threshold matrix are updated through FWA to enrich the diversity of the fireworks of the weight matrix and the fireworks of the threshold matrix.

[0104] Based on the above embodiment, after determining the number and amplitude of the explosion sparks, it is also necessary to displace the fireworks so that the explosion sparks are distributed within a determined range. This step includes:

[0105] Displace the explosion spark of the weight matrix and the explosion spark of the threshold matrix.

[0106] The displacement operation of FWA requires displacement in each dimension so that it is evenly distributed within the explosion range. The displacement equation is shown in formula (11):

[0107]

[0108] in, represents the position of firework i in the kth dimension after the shift operation, represents the position of firework i in the kth dimension, rand(0,R i ) represents a random number uniformly distributed within the explosion amplitude Ri. At this time, each firework has a corresponding number of explosion sparks, explosion amplitude and random displacement, thus ensuring the diversity of the firework population.

[0109] On the basis of the above embodiment, in order to further enrich the diversity of fireworks of weight matrix and fireworks of threshold matrix and avoid the obtained weight matrix and threshold matrix falling into local optimum, this embodiment performs Gaussian mutation on part of weight matrix and part of threshold matrix based on FWA. This step includes:

[0110] The initial fireworks of the weight matrix and the initial fireworks of the threshold matrix are processed by Gaussian mutation to obtain Gaussian mutation fireworks of the weight matrix and Gaussian mutation fireworks of the threshold matrix;

[0111] Correspondingly, using the initial fireworks of the weight matrix and the explosion sparks of the weight matrix to update the fireworks of the weight matrix, and using the initial fireworks of the threshold matrix and the explosion sparks of the threshold matrix to update the fireworks of the threshold matrix includes:

[0112] The fireworks of the weight matrix are updated using the initial fireworks of the weight matrix, the explosion sparks of the weight matrix, and the Gaussian mutation fireworks of the weight matrix, and the fireworks of the threshold matrix are updated using the initial fireworks of the threshold matrix, the explosion sparks of the threshold matrix, and the Gaussian mutation fireworks of the threshold matrix.

[0113] The operating equation of Gaussian variation is shown in formula (12):

[0114]

[0115] In formula (12), represents the position of the spark generated by firework i in the k-dimensional space after Gaussian mutation, and Gaussian (1,1) represents a Gaussian distribution with a mean and variance of 1. In FWA, Gaussian mutation can be performed on some fireworks in the fireworks population, so that the feasible domain space includes the initial fireworks, explosion sparks and Gaussian variation fireworks, that is, the feasible domain space of the weight matrix includes the initial fireworks, explosion sparks and Gaussian variation fireworks of the weight matrix, and the feasible domain space of the threshold matrix includes the initial fireworks, explosion sparks and Gaussian variation fireworks of the threshold matrix, thereby ensuring the diversity of the fireworks population. At this time, the weight matrix with the best fitness value can be selected from the initial fireworks, explosion sparks and Gaussian variation fireworks of the weight matrix through the selection operation in the above embodiment to enter the next iteration, and the remaining individuals are kept M-1 by using the roulette rule to enter the next iteration, so as to keep the population size of the fireworks of the updated weight matrix still M. It can be understood that the selection rule of the threshold matrix is ​​the same as that of the weight matrix, and this embodiment will not be repeated.

[0116] This embodiment performs Gaussian mutation on part of the weight matrix and part of the threshold matrix. Since Gaussian mutation has good global search capability, it can effectively prevent the iterative process from falling into a local optimum.

[0117] On the basis of the above embodiment, since fireworks distributed near the boundary of the feasible domain may produce explosion sparks beyond the feasible domain when they explode, in order to ensure that all explosion sparks are within the feasible domain, this embodiment maps the explosion sparks beyond the feasible domain to the feasible domain. This step includes:

[0118] Determine whether there are weight matrices and threshold matrices beyond the feasible domain in the explosion sparks of the weight matrix and the explosion sparks of the threshold matrix;

[0119] If it exists, the weight matrix and threshold matrix beyond the feasible domain are mapped into the feasible domain through modulo operation.

[0120] Specifically, the calculation formula of the modulo operation is shown in formula (13):

[0121]

[0122] In formula (13), X k Indicates that firework i crosses the k-th dimension boundary and is mapped to the position of the feasible domain, and Respectively represent the lower limit and upper limit of the k-dimensional feasible domain, and % represents the modulo operation. In a specific implementation, the weight matrix and threshold matrix beyond the feasible domain can be mapped into the feasible domain according to formula (13) as mapping fireworks, so that the feasible domain space includes initial fireworks, explosion sparks, Gaussian variation fireworks and mapping fireworks.

[0123] On the basis of the above embodiment, in order to improve the accuracy of data fusion, this embodiment will pre-process the monitoring data sent by each member node before performing data fusion. This step includes:

[0124] Receive monitoring data sent by each member node in the cluster;

[0125] Extracting characteristic values ​​of the monitoring data to obtain characteristic values ​​of the monitoring data;

[0126] The initial weight matrix and the initial threshold matrix are used as the current weight matrix and the current threshold matrix of the BP neural network in the cluster head, and the BP neural network is called to calculate the eigenvalues ​​to obtain the data fusion result.

[0127] For a BP neural network within a cluster, its input layer is located in each member node in the cluster and only performs data transmission, which is used to send the monitoring data collected by each member node in the cluster to the cluster head node, while its hidden layer and output layer are both located in the cluster head node and can be used for data transmission and data processing.

[0128] Specifically, the entire WSN should be clustered first, and each member node in each cluster will send the collected monitoring data to the cluster head node in the cluster to which it belongs; then the cluster head node and the member nodes in the cluster will use the BP neural network to perform two-level data fusion, which is specifically manifested as follows: the first-level neural network pre-processes the monitoring data sent by each member node in the cluster according to the function of the hidden layer, that is, extracts the feature values ​​of the monitoring data; the second-level neural network normalizes the pre-processed monitoring data by the cluster head node, and the normalized data is optimized through the functions of the hidden layer and output layer of the second-level neural network, and the data fusion result can be obtained at this time.

[0129] This embodiment uses the initial weight matrix and the initial threshold matrix as the current weight matrix and the current threshold matrix of the BP neural network for data fusion. Data fusion is used to reduce the amount of data transmission from the cluster head node to the aggregation node, effectively preventing increased energy consumption, increased delay and data loss caused by excessive data transmission. At the same time, since the current weight matrix and the current threshold matrix are obtained through multiple iterations based on FWA, the accuracy of data fusion is effectively improved.

[0130] In the above embodiments, the intelligent low-power WSN data fusion method is described in detail, and the present application also provides an embodiment corresponding to the intelligent low-power WSN data fusion device. It should be noted that the present application describes the embodiments of the device part from two perspectives, one is based on the functional module perspective, and the other is based on the hardware perspective.

[0131] Figure 2 The structure diagram of an intelligent low-power WSN data fusion device provided in this application is as follows: Figure 2 As shown, the device comprises:

[0132] An initialization module 10 is used to initialize the initial fireworks of the weight matrix and the initial fireworks of the threshold matrix of the BP neural network by using chaotic mapping to obtain an initial weight matrix and an initial threshold matrix;

[0133] An acquisition module 11 is used to acquire training samples and calculate training errors based on the training samples, the initial weight matrix and the initial threshold matrix;

[0134] The judging module 12 is used to judge whether the training error is qualified; if so, the fusion module 13 is triggered; if not, the updating module 14 is triggered;

[0135] A fusion module 13 is used to perform data fusion using the initial weight matrix and the initial threshold matrix as the current weight matrix and the current threshold matrix of the BP neural network in the cluster head;

[0136] The updating module 14 is used to update the fireworks of the weight matrix and the fireworks of the threshold matrix by using FWA, update the initial weight matrix and the initial threshold matrix according to the fireworks of the weight matrix and the fireworks of the threshold matrix, and trigger the obtaining module 11.

[0137] Since the embodiments of the apparatus part correspond to the embodiments of the method part, please refer to the description of the embodiments of the method part for the embodiments of the apparatus part, which will not be repeated here.

[0138] The intelligent low-power WSN data fusion device provided in this embodiment uses the initialization module to initialize the initial fireworks of the weight matrix and the initial fireworks of the threshold matrix of the BP neural network using the chaotic mapping to obtain the initial weight matrix and the initial threshold matrix; obtains the training sample through the acquisition module, and calculates the training error according to the training sample, the initial weight matrix and the initial threshold matrix; judges whether the training error is qualified through the judgment module; if so, triggers the fusion module; if not, triggers the update module; uses the initial weight matrix and the initial threshold matrix as the current weight matrix and the current threshold matrix of the BP neural network in the cluster head for data fusion through the fusion module; updates the fireworks of the weight matrix and the fireworks of the threshold matrix using FWA through the update module, updates the initial weight matrix and the initial threshold matrix according to the fireworks of the weight matrix and the fireworks of the threshold matrix, and triggers the acquisition module. It can be seen that the device uses the chaotic mapping to initialize the initial fireworks of the weight matrix and the initial fireworks of the threshold matrix of the BP neural network, so that the initial position distribution of the population is more uniform. In addition, the device uses FWA to update the fireworks of the weight matrix and the fireworks of the threshold matrix to enrich the diversity of the population, so that it can solve the current shortcomings of being easy to fall into local optimality and slow speed.

[0139] Figure 3The structure diagram of another intelligent low-power WSN data fusion device provided by this application is as follows: Figure 3 As shown, the intelligent low-power WSN data fusion device includes:

[0140] A memory 20, used for storing computer programs;

[0141] The processor 21 is used to implement the steps of the intelligent low-power WSN data fusion method mentioned in the above embodiment when executing the computer program.

[0142] The intelligent low-power WSN data fusion device provided in this embodiment may include but is not limited to a smart phone, a tablet computer, a laptop computer or a desktop computer.

[0143] Among them, the processor 21 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. The processor 21 may be implemented in at least one hardware form of a digital signal processor (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA). The processor 21 may also include a main processor and a coprocessor. The main processor is a processor for processing data in an awake state, also known as a central processing unit (CPU); the coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 21 may be integrated with a graphics processing unit (GPU), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 21 may also include an artificial intelligence (AI) processor, which is used to process computing operations related to machine learning.

[0144] The memory 20 may include one or more computer-readable storage media, which may be non-transitory. The memory 20 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In this embodiment, the memory 20 is at least used to store the following computer program 201, wherein, after the computer program is loaded and executed by the processor 21, the relevant steps of the intelligent low-power WSN data fusion method disclosed in any of the aforementioned embodiments can be implemented. In addition, the resources stored in the memory 20 may also include an operating system 202 and data 203, etc., and the storage method may be temporary storage or permanent storage. Among them, the operating system 202 may include Windows, Unix, Linux, etc. Data 203 may include but is not limited to fireworks of weight matrices and fireworks of threshold matrices, etc.

[0145] In some embodiments, the intelligent low-power WSN data fusion device may further include a display screen 22 , an input / output interface 23 , a communication interface 24 , a power supply 25 , and a communication bus 26 .

[0146] Those skilled in the art will understand that Figure 3 The structure shown in the figure does not constitute a limitation on the intelligent low-power WSN data fusion device, and may include more or fewer components than those shown in the figure.

[0147] The intelligent low-power WSN data fusion device provided in this embodiment includes a memory and a processor. When the processor executes the program stored in the memory, it can implement the above intelligent low-power WSN data fusion method, and the effect is the same as above.

[0148] Finally, the present application also provides an embodiment corresponding to a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the intelligent low-power WSN data fusion method as described in the above method embodiment are implemented.

[0149] It is understandable that if the method in the above embodiment is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0150] The computer-readable storage medium provided in this embodiment includes the above-mentioned intelligent low-power WSN data fusion method, and the effect is the same as above.

[0151] The above is a detailed introduction to the intelligent low-power WSN data fusion method, device and medium provided by the present application. The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the embodiments can refer to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can refer to the method part description. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the scope of protection of the claims of the present application.

[0152] It should also be noted that, in this specification, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

Claims

1. An intelligent low-power WSN data fusion method, characterized in that: include: Initializing the initial fireworks of the weight matrix and the initial fireworks of the threshold matrix of the BP neural network using chaotic mapping to obtain the initial weight matrix and the initial threshold matrix; Acquire a training sample, and calculate a training error based on the training sample, the initial weight matrix, and the initial threshold matrix; Determining whether the training error is qualified; If yes, the initial weight matrix and the initial threshold matrix are used as the current weight matrix and the current threshold matrix of the BP neural network in the cluster head for data fusion; If not, use FWA to update the fireworks of the weight matrix and the fireworks of the threshold matrix, update the initial weight matrix and the initial threshold matrix according to the fireworks of the weight matrix and the fireworks of the threshold matrix, and enter the step of calculating the training error according to the training samples, the initial weight matrix and the initial threshold matrix.

2. The intelligent low-power WSN data fusion method according to claim 1 is characterized in that: If the training error is unqualified, the following is also included: Increment the number of iterations by one, and determine whether the number of iterations reaches an iteration threshold; If it is reached, the step of using the initial weight matrix and the initial threshold matrix as the current weight matrix and the current threshold matrix of the BP neural network in the cluster head for data fusion is entered; If not reached, the step of using FWA to update the fireworks of the weight matrix and the fireworks of the threshold matrix is ​​entered.

3. The intelligent low-power WSN data fusion method according to claim 1 is characterized in that: The method of updating the fireworks of the weight matrix and the fireworks of the threshold matrix by using FWA includes: Obtaining the explosion spark of the weight matrix and the explosion spark of the threshold matrix through the FWA; The fireworks of the weight matrix are updated using the initial fireworks of the weight matrix and the explosion sparks of the weight matrix, and the fireworks of the threshold matrix are updated using the initial fireworks of the threshold matrix and the explosion sparks of the threshold matrix.

4. The intelligent low-power WSN data fusion method according to claim 3 is characterized in that: After obtaining the explosion sparks of the weight matrix and the explosion sparks of the threshold matrix through the FWA, and before updating the fireworks of the weight matrix using the initial fireworks of the weight matrix and the explosion sparks of the weight matrix, the method further includes: The explosion sparks of the weight matrix and the explosion sparks of the threshold matrix are shifted.

5. The intelligent low-power WSN data fusion method according to claim 4 is characterized in that: Before using the initial fireworks of the weight matrix and the explosion sparks of the weight matrix to update the fireworks of the weight matrix, and after shifting the explosion sparks of the weight matrix and the explosion sparks of the threshold matrix, the method further includes: Processing the initial fireworks of the weight matrix and the initial fireworks of the threshold matrix by Gaussian mutation to obtain Gaussian mutation fireworks of the weight matrix and Gaussian mutation fireworks of the threshold matrix; Correspondingly, the using the initial fireworks of the weight matrix and the explosion sparks of the weight matrix to update the fireworks of the weight matrix, and using the initial fireworks of the threshold matrix and the explosion sparks of the threshold matrix to update the fireworks of the threshold matrix includes: The fireworks of the weight matrix are updated using the initial fireworks of the weight matrix, the explosion sparks of the weight matrix, and the Gaussian variation fireworks of the weight matrix, and the fireworks of the threshold matrix are updated using the initial fireworks of the threshold matrix, the explosion sparks of the threshold matrix, and the Gaussian variation fireworks of the threshold matrix.

6. The intelligent low-power WSN data fusion method according to claim 5 is characterized in that: Before updating the fireworks of the weight matrix using the initial fireworks of the weight matrix, the explosion sparks of the weight matrix and the Gaussian variation fireworks of the weight matrix, and after obtaining the Gaussian variation fireworks of the weight matrix and the Gaussian variation fireworks of the threshold matrix, the method further includes: Determine whether there are weight matrices and threshold matrices that are beyond the feasible domain in the explosion sparks of the weight matrix and the explosion sparks of the threshold matrix; If so, the weight matrix and the threshold matrix that are beyond the feasible domain are mapped into the feasible domain through a modulo operation.

7. The intelligent low-power WSN data fusion method according to claim 1 is characterized in that: The data fusion of using the initial weight matrix and the initial threshold matrix as the current weight matrix and the current threshold matrix of the BP neural network in the cluster head comprises: Receive monitoring data sent by each member node in the cluster; Extracting characteristic values ​​from the monitoring data to obtain characteristic values ​​of the monitoring data; The initial weight matrix and the initial threshold matrix are used as the current weight matrix and the current threshold matrix of the BP neural network in the cluster head, and the BP neural network is called to calculate the eigenvalues ​​to obtain a data fusion result.

8. An intelligent low-power WSN data fusion device, characterized in that: include: An initialization module, used for initializing the initial fireworks of the weight matrix and the initial fireworks of the threshold matrix of the BP neural network by using chaotic mapping to obtain an initial weight matrix and an initial threshold matrix; An acquisition module, used for acquiring training samples and calculating training errors according to the training samples, the initial weight matrix and the initial threshold matrix; A judgment module is used to judge whether the training error is qualified; if so, the fusion module is triggered; if not, the update module is triggered; The fusion module is used to perform data fusion using the initial weight matrix and the initial threshold matrix as the current weight matrix and the current threshold matrix of the BP neural network in the cluster head; The updating module is used to update the fireworks of the weight matrix and the fireworks of the threshold matrix by using FWA, update the initial weight matrix and the initial threshold matrix according to the fireworks of the weight matrix and the fireworks of the threshold matrix, and trigger the acquisition module.

9. An intelligent low-power WSN data fusion device, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of the intelligent low-power WSN data fusion method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the steps of the intelligent low-power WSN data fusion method according to any one of claims 1 to 7 are implemented.

Images