Relay protection checking method based on improved wavelet threshold and kalman filter

Abstract

The application discloses a relay protection checking method based on improved wavelet threshold and Kalman filtering, and belongs to the technical field of automatic test of power systems, in particular to the following: at the moment of simulating the action of a circuit breaker, current signals in a driving coil are collected in real time; an improved wavelet threshold denoising algorithm is used, a db4 wavelet base with a compact support set and orthogonality is selected, and discrete wavelet transform is used to perform J-layer decomposition on the collected discrete current signals; a soft threshold function is used to denoise the high-frequency detail components, then inverse wavelet transform is performed to obtain the denoised current signals; the denoised current signals are optimally estimated by using Kalman filtering, and through adaptive adjustment of an observation noise covariance matrix, the action current of the circuit breaker is accurately reconstructed. The application significantly improves the optimal estimation and accurate reconstruction capability of the action current of the circuit breaker under a complex electromagnetic environment.

Description

Technical Field

[0001] This invention belongs to the field of power system automation testing technology, and particularly relates to a relay protection verification method based on improved wavelet threshold and Kalman filtering. Background Technology

[0002] Relay protection systems are among the most important secondary equipment in power systems and are a key component in ensuring the safe operation of the power system. During the construction and operation of substations, it is necessary to regularly inspect relay protection devices and circuits to ensure they are functioning properly.

[0003] With the expansion of the power grid, the periodic inspection of relay protection and safety automatic devices (such as automatic transfer switch and low-frequency low-voltage load shedding devices) is crucial. Currently, on-site verification mainly faces the following problems: Power supply reliability contradictions: Directly using real high-voltage circuit breakers for transmission tests can affect the operation of primary equipment, and frequent operations can shorten the lifespan of circuit breakers. Complex and inefficient wiring: Existing methods often require manual short-circuiting of input quantities or using multimeters to measure multiple output contacts, making multi-person coordination difficult, prone to accidental tripping of circuits, and difficult to verify complex logic (such as multiple output contacts of main transformer backup protection). Lack of refined diagnostics: Most existing analog devices can only provide simple "on / off" switching status feedback, unable to perform in-depth diagnostics based on analog quantities (current signals) on the electrical health status of analog circuits (such as coil aging and poor contact).

[0004] This relay-assisted verification device frequently operates in environments with strong electromagnetic interference and requires long-distance transmission capabilities. Furthermore, the current signal in this project is not a standard sine wave and exhibits the following characteristics: 1. Sudden abrupt change: When the main or sub-equipment issues a closing / opening command, the relay (simulated circuit breaker) coil is energized, and the current jumps instantaneously from 0 to its steady-state value; or it returns to zero instantaneously upon opening. 2. Contains abundant electromagnetic noise. 3. Inductive load characteristics: Because the simulated circuit breaker is composed of relay coils, a back electromotive force is generated at the moment of opening and closing, causing spikes or oscillating attenuation in the current waveform. This makes effective transmission of the current signal challenging. Therefore, providing an effective method for filtering and diagnosing the current signal under these operating characteristics is of great significance for the efficient operation of the relay-assisted verification device. Summary of the Invention

[0005] Purpose of the invention: The purpose of this invention is to provide a relay protection verification method based on improved wavelet thresholding and Kalman filtering. This method replaces the actual circuit breaker with wireless networking and introduces advanced filtering and feature extraction algorithms for current signals, thereby achieving precise control of the verification process and fault diagnosis.

[0006] Technical solution: This invention provides a relay protection verification method based on improved wavelet thresholding and Kalman filtering, specifically as follows:

[0007] At the moment of the simulated circuit breaker operation, the discrete current signal in the drive coil is acquired in real time;

[0008] An improved wavelet thresholding denoising algorithm is adopted. The db4 wavelet basis with compact support and orthogonality is selected to perform J-level decomposition on the acquired discrete current signal. The soft thresholding function is applied to denoise the high-frequency detail components, and then the inverse wavelet transform is performed to obtain the denoised current signal.

[0009] A Kalman filter is used to perform optimal estimation of the denoised current signal. By adaptively adjusting the observation noise covariance matrix, the circuit breaker operating current is reconstructed.

[0010] Furthermore, the acquired discrete current signal is decomposed into J-levels using discrete wavelet transform:

[0011] ;

[0012] in, Let be the approximation coefficients for the j-th layer. Here, n represents the detail coefficients of the j-th layer, which are also known as high-frequency detail components, and n represents the n-th sampling point. Represents the discrete current after wavelet transform;

[0013] Calculate the wavelet detail coefficients of the j-th layer :

[0014] ;

[0015] in, These are wavelet basis functions;

[0016] The following smoothing threshold function is used to adjust the wavelet detail coefficients. Processing:

[0017] ;

[0018] in, For the threshold, , Let N be the noise standard deviation and N be the total number of sampling points. Here are the processed wavelet detail coefficients, m is the adjustment factor, and sgn is the sign function;

[0019] use and right Perform inverse wavelet transform to obtain the denoised current signal.

[0020] Furthermore, the denoised current signal As an observation, it is introduced into the Kalman filter, specifically as follows:

[0021] The state-space equation of the Kalman filter is:

[0022] ;

[0023] in, for The estimated current at time [time]. Here, A represents the observed values ​​after wavelet denoising, and H represents the state transition equation. For process noise, To observe the noise, for The estimated current value at time;

[0024] Prior estimation: ;

[0025] Prior covariance: ;

[0026] Kalman gain: ;

[0027] in, Let be the prior state estimate at time k. This is the posterior state estimate from the previous time step. To estimate the covariance matrix a priori, Let be the posterior error covariance matrix. The process noise covariance matrix is... To observe the noise covariance matrix, Let K be the Kalman gain at time k, and T denote the matrix transpose;

[0028] Furthermore, the observation noise covariance matrix is ​​adaptively adjusted, specifically as follows:

[0029] Define residual sequence :

[0030] ;

[0031] Constructing a statistical criterion function based on residuals :

[0032] ;

[0033] in, The sliding window length of the residual sequence;

[0034] when Exceeding the preset confidence interval threshold At that time, the observation noise covariance matrix is ​​corrected by exponential decay:

[0035] ;

[0036] in, The corrected observation noise covariance matrix is ​​obtained when the exponential decay correction is performed for the first time. , The initial observation noise covariance matrix, This is the adjustment coefficient.

[0037] Furthermore, the method also includes using the CPSO-BP model to identify the state of the circuit breaker and drive circuit based on the reconstructed current, the state including: state I: normal operation, state II: inter-turn short circuit, state III: poor contact or loose connection, and state IV: mechanical jamming or delayed operation.

[0038] Furthermore, the number of input layer nodes in the BP neural network of the CPSO-BP model is... The number of hidden layer nodes is E, and the number of output layer nodes is... Then the total number of weights and thresholds that the BP neural network needs to optimize is: .

[0039] Furthermore, the chaotic particle swarm optimization (CPSO) employs the Cubic chaotic map:

[0040] A random number is generated at the beginning. Initial vector of dimension Iterative generation using the Cubic mapping formula A sequence of chaotic vectors :

[0041] ;

[0042] in, This represents the i-th chaotic vector. Indicates system control parameters, and ;

[0043] Will The range of values ​​mapped to the parameters of the BP neural network Inside, the location where the initial particle swarm is generated. :

[0044] ;

[0045] in, This represents the position of the i-th particle. express The d-th dimension, express The d-th dimension, This represents the minimum value of the parameters in a BP neural network. This represents the maximum value of the parameters in the BP neural network;

[0046] Define the fitness function of a particle. for:

[0047] ;

[0048] in, The total number of training samples, In order to target the The nth sample, the nth in the output layer The true expected output of each node for The predicted value;

[0049] In each iteration, the particle is based on its historical best position. and the global optimal position Update speed and location :

[0050] ;

[0051] ;

[0052] in, For inertial weights, and As a learning factor, and For random numbers that follow a uniform distribution, This represents the historical optimal position of the i-th particle in dimension d during the t-th iteration. Let be the globally optimal position of the i-th particle in the d-th dimension at the t-th iteration. Represents the value at the t-th iteration. , The d-th dimension represents the velocity of the i-th particle at the t-th iteration;

[0053] If the iteration gets stuck in a local optimum, then extract the current global optimum position. Chaotic mutation is performed on it using Cubic mapping:

[0054] ;

[0055] in, This represents the globally optimal position after the chaotic mutation. The radius is the perturbation radius, which decreases adaptively with the number of iterations. The fitness of a chaotic vector, randomly generated according to the Cubic mapping formula, is calculated. If the fitness is better than the current global optimum, a replacement is implemented.

[0056] A computer device includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the relay protection verification method based on improved wavelet thresholding and Kalman filtering.

[0057] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the relay protection verification method based on improved wavelet thresholding and Kalman filtering.

[0058] Beneficial effects:

[0059] 1. This invention improves the quality and fidelity of signal denoising. By adopting the db4 wavelet basis with compact support and orthogonality and a smooth adjustment threshold function, it effectively overcomes the signal distortion and constant deviation problems existing in traditional soft and hard threshold functions during denoising. While filtering out high-frequency noise, it can more completely preserve the characteristic details of the current signal at the moment of circuit breaker operation.

[0060] 2. To enhance the real-time performance and accuracy of current reconstruction, an adaptive Kalman filter is introduced. By constructing a statistical criterion function based on residuals, the covariance matrix of the observation noise is optimized. The dynamic exponential decay correction enables the algorithm to track signal changes in real time, significantly improving the optimal estimation and accurate reconstruction capability of circuit breaker operating current in complex electromagnetic environments.

[0061] 3. Improve the intelligence and accuracy of equipment status identification. Utilizing a CPSO-BP (Chaotic Particle Swarm Optimized Backpropagation Neural Network Model) constructed using reconstructed current, this method can automatically classify and identify various latent faults such as inter-turn short circuits, poor contact, and mechanical jamming. Compared to traditional verification methods, this method is no longer limited to judging a single action time or amplitude, offering a more comprehensive diagnostic dimension.

[0062] 4. The global search capability and training efficiency of the algorithm have been optimized. The Chaotic Particle Swarm Optimization (CPSO) algorithm introduces Cubic (cubic chaotic mapping) chaotic mapping for population initialization and chaotic mutation when trapped in local optima. This effectively avoids the problem of BP neural networks being prone to getting trapped in local minima, improves the convergence speed and global optimization accuracy of model parameter optimization, and ensures the stability of the verification results. Attached Figure Description

[0063] Figure 1 This is a schematic diagram of the embodiment;

[0064] Figure 2 This is a flowchart of the cascaded processing of the improved wavelet threshold and Kalman filter according to the present invention.

[0065] Figure 3 Output diagram for CPSO-BP model optimization and diagnosis;

[0066] Figure 4 This is a flowchart for fault diagnosis. Detailed Implementation

[0067] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0068] This invention mainly addresses two issues: first, how to effectively filter the current signal acquired in real time by the device in a working environment characterized by complex electromagnetic environment and high signal noise during long-distance signal transmission; and second, how to select a suitable model to perform pattern recognition and classification on the filtered current signal.

[0069] like Figure 1 As shown, the sub-device unit is responsible for sampling high-frequency current data and performing initial packaging, then transmitting it to the main device via a wireless module. The main device or host computer (tablet computer) has a built-in diagnostic module that runs a cascaded processing method based on improved wavelet thresholding and Kalman filtering to analyze the received waveform data and identify faults. Figure 2 , 3 The specific steps of the method shown are as follows:

[0070] (1) Current signal acquisition: At the moment of the simulated circuit breaker action (closing / opening), the device acquires the current signal in the drive coil in real time;

[0071] (2) Filtering based on the cascaded processing algorithm of improved wavelet threshold and Kalman filter: In view of the problems of complex electromagnetic environment, high signal noise, long-distance signal transmission and current signal characteristics, this invention proposes a cascaded processing method based on improved wavelet threshold and Kalman filter.

[0072] a. The current signal is decomposed into multiple scales using the db4 wavelet basis, and a soft threshold function is applied to denoise the high-frequency detail components to effectively filter out power supply ripple and high-frequency noise in the field.

[0073] b. Introduce the Kalman filter algorithm to make the optimal estimate of the current waveform after cleaning, and realize the accurate reconstruction of the circuit breaker operating current by observing the adaptive adjustment of the noise covariance matrix.

[0074] (3) Fault diagnosis algorithm based on CPSO-BP neural network: extract the feature vector of the filtered current waveform, input the feature vector into the parameter-optimized BP neural network model, and output the diagnosis result. The specific steps are as follows:

[0075] a. Chaotic initialization of the particle swarm, generating the initial particle swarm location. ;

[0076] b. Based on iterative conditions and a global monitoring mechanism, the optimal values ​​for individual particles and the global optimum are obtained;

[0077] c. Position the optimal particle The weights and thresholds are decoded into a BP neural network, and the network is fine-tuned using the standard error backpropagation algorithm to output the diagnostic results.

[0078] like Figure 2 As shown, addressing the first problem raised in this invention, and based on the characteristics of the current signal during device operation, this invention employs a cascaded processing method based on improved wavelet thresholding and Kalman filtering. The specific implementation method is as follows:

[0079] Discrete Wavelet Transform (DWT) based on an improved threshold function:

[0080] To preserve the edge features of the current signal at the instant of transition (i.e., the moment the circuit breaker operates) while filtering out noise, this invention employs an improved wavelet thresholding denoising algorithm. A db4 wavelet basis with compact support and orthogonality is selected, and the acquired discrete current signal is decomposed into J-level components using discrete wavelet transform to obtain the approximate coefficients of the j-th level. (Low frequency) and detail coefficient (high frequency):

[0081] ;

[0082] in, Let be the approximation coefficients for the j-th layer. Here, n represents the detail coefficients of the j-th layer, which are also known as high-frequency detail components, and n represents the n-th sampling point. Represents the discrete current after wavelet transform;

[0083] In the Wavelet detail coefficients at the level decomposition scale The calculation is as follows:

[0084] ;

[0085] in, These are wavelet basis functions;

[0086] To address the issues of constant bias in traditional soft thresholding functions and discontinuity in hard thresholding functions, this invention proposes a method for smoothly adjusting the thresholding function for wavelet detail coefficients. Process accordingly. Let the threshold be... (in The standard deviation of noise. (where N is the total number of signal sampling points, and is the adaptive signal threshold) and the improved wavelet detail coefficients. for:

[0087] ;

[0088] when hour, Setting it to 0 ensures continuity at the threshold and avoids oscillations caused by hard thresholds.

[0089] when At that time, the exponent term Rapidly increasing, making ,at this time This is through the regulating factor (m>0) achieves a smooth approximation to the hard threshold, thereby preserving the steep rising edge characteristics of the relay operating current to the greatest extent in the high signal-to-noise ratio region.

[0090] Using the processed coefficients And the original approximation coefficient Perform inverse wavelet transform to obtain the signal after preliminary denoising. This step will effectively filter out 150kHz power supply ripple and high-frequency electromagnetic noise.

[0091] Residual Adaptive Kalman Filtering

[0092] Although the signal is smoothed after wavelet denoising, it may still exhibit low-frequency random drift. A Kalman filter is introduced as the observation value. To address the filter tracking lag caused by the sudden change in current during relay operation, an adaptive update mechanism based on residuals is introduced.

[0093] State-space equations:

[0094] ;

[0095] in, for Current estimate at time step for Current estimate at time step Here, A represents the observed values ​​after wavelet denoising, and H represents the state transition equation. For process noise, To observe noise.

[0096] Standard Kalman recursion:

[0097] Prior estimation: ;

[0098] Prior covariance: ;

[0099] Calculate the Kalman gain: ;

[0100] in, Let k be the prior state estimate at time k. This is the posterior state estimate from the previous time step. To estimate the covariance matrix a priori, Let be the posterior error covariance matrix. The process noise covariance matrix is... To observe the noise covariance matrix, Let K be the Kalman gain at time k, and T denote the matrix transpose.

[0101] Adaptive adjustment criterion: Define residual sequence Construct a statistical criterion function based on residuals. :

[0102] ;

[0103] in, The sliding window length of the residual sequence;

[0104] like Exceeding the preset confidence interval threshold A current step is detected (circuit breaker trips). At this point, the observed noise covariance matrix is... Perform exponential decay correction to increase the observation weights:

[0105] ;

[0106] in, The corrected observation noise covariance matrix is ​​obtained when the exponential decay correction is performed for the first time. , The initial observation noise covariance matrix, This is the adjustment coefficient.

[0107] like Figure 3As shown, addressing the second problem, and considering the characteristics of signals such as small sample size, long-distance transmission, and high accuracy requirements, this embodiment employs an intelligent diagnostic method based on adaptive feature extraction. This is achieved by using the acquired simulated circuit breaker circuit current signal to classify and identify the following four states of the simulated circuit breaker and its drive circuit: State I: Normal operation; State II: Inter-turn short circuit; State III: Poor contact or intermittent connection; State IV: Mechanical jamming / operation delay. This invention employs a fault diagnosis method based on a CPSO-BP neural network. The input signal is a reconstructed current signal obtained after processing using a cascaded processing method based on improved wavelet thresholding and Kalman filtering. .

[0108] CPSO-BP Model Optimization and Diagnostic Output

[0109] After the preceding "improved wavelet threshold-adaptive Kalman" cascaded filtering, the pure current signal Feature vectors (such as peak time, coil energizing time, waveform energy, etc.) are extracted and used as input to the BP neural network.

[0110] Let the topology of the BP neural network be... (Input layer number of nodes) Number of hidden layer nodes Number of output layer nodes At this point, the total number of weights and thresholds that the network needs to optimize is... .

[0111] In this invention, the dimension of each particle in the particle swarm optimization algorithm is set to... That is, the position vector of each particle. Each set of initial weights and thresholds uniquely corresponds to a set of BP neural networks.

[0112] The Cubic mapping exhibits strong chaotic properties and complex nonlinearity, with a large dynamic range and the ability to generate more complex and random chaotic sequences. Furthermore, there is a clear logical connection between the uniformity of the Cubic mapping and the high sensitivity of detecting circuit breaker fault characteristics (such as subtle feature changes caused by minute inter-turn short circuits). This invention selects the Cubic mapping to generate chaotic sequences for particle swarm initialization and chaotic perturbation when trapped in local optima.

[0113] The iterative mathematical model for the Cubic chaotic mapping is as follows:

[0114]

[0115] in Indicates the first The chaotic variable in the next iteration has a value range of... ,and , Indicates system control parameters, when At that time, the Cubic mapping is in a completely chaotic state.

[0116] Randomly generate one Initial vector of dimension Iterative generation using the Cubic mapping formula A sequence of chaotic vectors Linearly mapping chaotic sequences to the range of parameters in a BP network. Inside, the location where the initial particle swarm is generated. :

[0117] ;

[0118] Where d represents the d-th dimension, which represents the position of each particle. The weights and thresholds are decoded into a BP neural network, and forward propagation is performed using a training set (such as samples of circuit breaker normal operation, core jamming, and inter-turn short circuit in the coil). The fitness function of the particles is defined. The mean square error (MSE) between the network output and the expected output:

[0119] ;

[0120] in The total number of training samples, This represents the number of output layer nodes (i.e., the number of fault types). For the first The nth sample, the nth in the output layer The true expected output of each node for The predicted value. The smaller the value, the higher the diagnostic accuracy of the set of parameters.

[0121] In each iteration, the particle is based on its historical best position. and the global optimal position Update speed and location :

[0122] ;

[0123] ;

[0124] in Inertial weights control the algorithm's global exploration and local development capabilities; The learning factor represents the step size by which a particle learns from its own experience and the experience of the group. To obey Uniformly distributed random numbers, This represents the historical optimal position of the i-th particle in dimension d during the t-th iteration. Let be the globally optimal position of the i-th particle in the d-th dimension at the t-th iteration. Represents the value at the t-th iteration. , Let d represent the d-th dimension of the velocity of the i-th particle at the t-th iteration.

[0125] To prevent population stagnation, a monitoring mechanism will be introduced. Specifically, the rate of change of the optimal fitness will be detected globally. If the rate of change is not consistently high after multiple iterations, the global optimal position will be identified. If no significant decrease occurs, the algorithm is considered to be trapped in a local optimum.

[0126] At this point, extract the current globally optimal particle. Chaotic mutation is performed on it using Cubic mapping:

[0127] ;

[0128] in, This represents the globally optimal position after the chaotic mutation. The radius is the perturbation radius, which decreases adaptively with the number of iterations. The fitness of a chaotic vector is calculated based on a randomized Cubic mapping formula. If the fitness is better than the current global optimum, a replacement is implemented.

[0129] The algorithm terminates when the maximum number of iterations is reached or the target error requirement is met. The final result... The values ​​are mapped onto a backpropagation (BP) neural network as optimal initial weights and thresholds. Subsequently, the network is fine-tuned using the standard backpropagation algorithm (gradient descent) to achieve accurate classification of circuit breaker states.

[0130] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the present invention will not describe the various possible combinations separately.

Claims

1. A relay protection verification method based on improved wavelet thresholding and Kalman filtering, characterized in that, Specifically: At the moment of the simulated circuit breaker operation, the discrete current signal in the drive coil is acquired in real time; An improved wavelet thresholding denoising algorithm is adopted. The db4 wavelet basis with compact support and orthogonality is selected to perform J-level decomposition on the acquired discrete current signal. The soft thresholding function is applied to denoise the high-frequency detail components, and then the inverse wavelet transform is performed to obtain the denoised current signal. A Kalman filter is used to perform optimal estimation of the denoised current signal. By adaptively adjusting the observation noise covariance matrix, the circuit breaker operating current is reconstructed.

2. The relay protection verification method based on improved wavelet thresholding and Kalman filtering according to claim 1, characterized in that, The discrete current signal was decomposed into J-levels using discrete wavelet transform: ； in, Let be the approximation coefficients for the j-th layer. Here, n represents the detail coefficients of the j-th layer, which are also known as high-frequency detail components, and n represents the n-th sampling point. Represents the discrete current after wavelet transform; Calculate the wavelet detail coefficients of the j-th layer. : ； in, These are wavelet basis functions; The following smoothing threshold function is used to adjust the wavelet detail coefficients. Processing: ； in, For the threshold, , Let N be the noise standard deviation and N be the total number of sampling points. Here are the processed wavelet detail coefficients, m is the adjustment factor, and sgn is the sign function; use and right Perform inverse wavelet transform to obtain the denoised current signal.

3. The relay protection verification method based on improved wavelet thresholding and Kalman filtering according to claim 1, characterized in that, The noise-reduced current signal As an observation, it is introduced into the Kalman filter, specifically as follows: The state-space equation of the Kalman filter is: ； in, for The estimated current at time [time]. Here, A represents the observed values ​​after wavelet denoising, and H represents the state transition equation. For process noise, To observe the noise, for The estimated current value at time; Prior estimation: ; Prior covariance: ; Kalman gain: ; in, Let be the prior state estimate at time k. This is the posterior state estimate from the previous time step. To estimate the covariance matrix a priori, Let be the posterior error covariance matrix. The process noise covariance matrix is... To observe the noise covariance matrix, Let K be the Kalman gain at time k, and T denote the matrix transpose.

4. The relay protection verification method based on improved wavelet thresholding and Kalman filtering according to claim 3, characterized in that, The observation noise covariance matrix is ​​adaptively adjusted as follows: Define residual sequence : ； Constructing a statistical criterion function based on residuals : ； in, The sliding window length of the residual sequence; when Exceeding the preset confidence interval threshold At that time, the observation noise covariance matrix is ​​corrected by exponential decay: ； in, The corrected observation noise covariance matrix is ​​obtained when the exponential decay correction is performed for the first time. , The initial observation noise covariance matrix, This is the adjustment coefficient.

5. The relay protection verification method based on improved wavelet thresholding and Kalman filtering according to claim 1, characterized in that, The method also includes using the CPSO-BP model to identify the state of the circuit breaker and drive circuit based on the reconstructed current, the state including: state I: normal operation, state II: inter-turn short circuit, state III: poor contact or loose connection, and state IV: mechanical jamming or delayed operation.

6. The relay protection verification method based on improved wavelet thresholding and Kalman filtering according to claim 5, characterized in that, The number of input layer nodes in the BP neural network of the CPSO-BP model is The number of hidden layer nodes is E, and the number of output layer nodes is... Then the total number of weights and thresholds that the BP neural network needs to optimize is: .

7. The relay protection verification method based on improved wavelet thresholding and Kalman filtering according to claim 6, characterized in that, Chaotic particle swarm optimization (CPSO) employs the Cubic chaotic map: A random number is generated at the beginning. Initial vector of dimension Iterative generation using the Cubic mapping formula A sequence of chaotic vectors : ； in, This represents the i-th chaotic vector. Indicates system control parameters, and ; Will The range of values ​​mapped to the parameters of the BP neural network Inside, the location where the initial particle swarm is generated. : ； in, This represents the position of the i-th particle. express The d-th dimension, express The d-th dimension, This represents the minimum value of the parameters in a BP neural network. This represents the maximum value of the parameters in the BP neural network; Define the fitness function of a particle. for: ； in, The total number of training samples, In order to target the The nth sample, the nth in the output layer The true expected output of each node for The predicted value; In each iteration, the particle is based on its historical best position. and the global optimal position Update speed and location : ； ； in, For inertial weights, and As a learning factor, and For random numbers that follow a uniform distribution, This represents the historical optimal position of the i-th particle in dimension d during the t-th iteration. Let be the globally optimal position of the i-th particle in the d-th dimension at the t-th iteration. Represents the value at the t-th iteration. , The d-th dimension represents the velocity of the i-th particle at the t-th iteration; If the iteration gets stuck in a local optimum, then extract the current global optimum position. Chaotic mutation is performed on it using Cubic mapping: ； in, This represents the globally optimal position after the chaotic mutation. The radius is the perturbation radius, which decreases adaptively with the number of iterations. The fitness of a chaotic vector, randomly generated according to the Cubic mapping formula, is calculated. If the fitness is better than the current global optimum, a replacement is implemented.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, When the processor executes the computer program, it implements the relay protection verification method based on improved wavelet threshold and Kalman filtering as described in claim 1.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the relay protection verification method based on improved wavelet threshold and Kalman filtering as described in claim 1.

Images