A power transmission line equipment fault classification method, system and device based on DWT-RBF

Abstract

A power transmission line equipment fault classification method, system and device based on DWT-RBF, the method first builds a simulation system of three-phase series compensation power, and defines the fault type, then acquires the sampling signal of transient current when the power transmission line is faulty, uses DWT to decompose, then denoises based on an improved wavelet threshold algorithm, and acquires the energy value of the sampling signal, then normalizes it, finally optimizes the DWT-RBF simulation system based on the IWOA algorithm, and inputs the energy value to complete fault classification; the sampling signal is decomposed by DWT, and the fault features are extracted, the response speed of fault classification is accelerated, the wavelet threshold denoising algorithm is used for filtering, the identification accuracy is improved, the RBF model is optimized by the IWOA algorithm, the convergence speed and generalization fitting ability of the model are improved, the network performance of the application is good, and different types of power transmission line equipment faults can be accurately and quickly classified.

Description

Technical Field

[0001] This invention relates to fault classification methods, belonging to the field of transmission line fault diagnosis technology, and particularly to a fault classification method, system and equipment for transmission line equipment based on DWT-RBF. Background Technology

[0002] As an indispensable infrastructure in modern society, the safety and stability of the power system are crucial for ensuring social life and economic development. Power system failures can lead to power outages, equipment damage, and even accidents. Therefore, timely and accurate diagnosis and handling of failures are necessary to ensure the safe operation of the power system.

[0003] In power system faults, when two or more conductors come into contact with each other or with the ground, the fault can be classified as a single-line-to-ground, double-line, double-line-to-ground, or three-line fault. These faults can cause serious damage, affecting power system equipment and power quality. However, the fault state is difficult to detect due to its short duration. Therefore, it is necessary to use high-sampling-rate transient detection instruments and combine them with transient signal processing and analysis methods to classify the fault. In recent years, there have been various researches on transmission line fault type judgment methods based on intelligent theory. However, in existing methods, fault type judgment based on expert systems is prone to "infinite loops" in reasoning search, resulting in poor accuracy. While fault type judgment based on conventional neural networks has a certain fault tolerance, the network utilization efficiency is low, and the network learning and training speed is slow. Summary of the Invention

[0004] The purpose of this invention is to overcome the above-mentioned defects and problems in the prior art and to provide a method, system and equipment for classifying transmission line equipment faults based on DWT-RBF with high accuracy and good performance.

[0005] To achieve the above objectives, the technical solution of the present invention is: a fault classification method for transmission line equipment based on DWT-RBF, comprising:

[0006] S1. Construct a DWT-RBF simulation system for three-phase series compensated power;

[0007] S2. Obtain the sampling signal of transient current during transmission line faults, decompose the sampling signal using DWT, then denoise it based on the improved wavelet thresholding algorithm, and obtain the energy value of the sampling signal.

[0008] S3. Normalize the energy value of the sampled signal and obtain the maximum and minimum values ​​of the energy value;

[0009] S4. Optimize the DWT-RBF simulation system based on the IWOA algorithm; the IWOA algorithm optimization includes any one or any combination of the following: optimal point set optimization for initializing the population, elite reverse learning strategy, nonlinear parameter control, and random difference mutation strategy;

[0010] S5. Based on the optimized DWT-RBF simulation system, the energy value is input into the optimized DWT-RBF neural network to complete the fault classification.

[0011] In step S2, the improved wavelet thresholding algorithm for noise reduction specifically includes:

[0012] S21. Based on the normal distribution principle of Gaussian distribution, the noise of the sampled signal is distributed in the interval [-3σ, 3σ]. A maximum-minimum threshold is used to denoise the sampled data of each layer component within the interval [-3σ, 3σ]. The denoising expression is as follows:

[0013]

[0014] Where: λ is the maximum and minimum threshold, and N is the number of sampling points;

[0015] S22. Reconstruct the denoised sampled signal to obtain the denoised signal; the expression for the reconstruction is as follows:

[0016] σ=mediab(s(z)) / 0.6745, z=0, 1,...,N-1;

[0017] Where: σ is the approximate variance of the sampled signal containing noise, median is the median, s(z) is the data point of the component, and z is the current sampling point;

[0018] S23. Calculate the energy value of the sampled signal; the calculation formula is as follows:

[0019]

[0020] Where: X 2 (z) is the sampled signal when the sampling point is z.

[0021] In step S3, the formulas for calculating the maximum and minimum energy values ​​are as follows:

[0022]

[0023] Where: X i Let X be the energy value of the i-th sampled signal. max X represents the maximum energy of the sampled signal. min This represents the minimum energy of the sampled signal.

[0024] The optimal point set optimization initialization of the population in step S4 refers to: initializing the starting position of individual whales by introducing the optimal point set method, as follows:

[0025] Let G s Let G be a unit cube in S-dimensional Euclidean space, and let the optimal point g∈G s The optimal point set is as follows:

[0026] P n (k)={({g1·k}, {g2·k},…,{g s ·k}), 1≤k≤n};

[0027] Where: P n (k) is the set of best points, {g} s ·k} is the value of g s ·k represents the decimal part of the result, g represents the optimal point, n represents the number of samples, and k represents the k-th sample out of n samples;

[0028] The deviation then satisfies the following:

[0029]

[0030] in: Let C(g, ε) be the deviation, C(g, ε) be a constant, ε be any positive number, and g be the optimal point;

[0031]

[0032] Wherein: g j Let p be the optimal point in the j-th dimension; j is the spatial dimension, and p is the smallest prime number.

[0033] For an unknown uniformly distributed object, the deviation of n points selected randomly is as follows:

[0034] O(n -1 / 2 (loglogn) 1 / 2 );

[0035] The location of individual whales in the population, based on the optimal point set method, is mapped to the search space as follows:

[0036]

[0037] in: Let ub be the position of the k-th whale in the j-th dimension. j lb j These represent the upper and lower limits of the search range in the j-th dimension, respectively.

[0038] The elite reverse learning strategy in step S4 refers to: by introducing an elite strategy, the initialized whale population is sorted according to fitness, and individuals with higher fitness are selected as elite individuals. The reverse solution of the elite solution is calculated, and the better individuals in the elite solution and the reverse solution are selected as elite individuals in the next generation population, as detailed below:

[0039] Assume that each whale after each iteration is an elite individual, and its expression is as follows:

[0040]

[0041] in: Let be the position vector of the i-th whale in the j-th dimensional space, where D is the spatial dimension and N is the population size. Let N be the position of the Nth whale in D-dimensional space;

[0042] The strategy formula for calculating the inverse solution of the elite solution is as follows:

[0043]

[0044]

[0045] in: and All are reverse solutions, where c is a random number in the interval [0,1], and ub j lb j These are the upper and lower limits of the dynamic boundary, respectively;

[0046] If the inverse solution exceeds the boundary, the inverse solution is reset using random generation; the reset formula is as follows:

[0047]

[0048] Where: rand() is a random function.

[0049] The nonlinear parameter control in step S4 refers to: by changing the amount of change of the nonlinear convergence factor, the local exploitation capability is enhanced to escape the local optimum, as detailed below:

[0050] The expression for the nonlinear convergence factor is as follows:

[0051]

[0052] Where: a is the nonlinear convergence factor, b and e are both control factors, and T max t represents the maximum number of iterations, and t represents the current number of iterations.

[0053] The random difference mutation strategy in step S4 refers to: using a random difference mutation strategy to randomly mutate and perturb the whale population during the search phase, and then obtaining the optimal position before and after the change again, as detailed below:

[0054] For the search phase, the random difference mutation strategy is as follows:

[0055] X(t+1)=δ1(X * (t)-X(t))+δ2(X rand (txt));

[0056] Where: X(t+1) is the whale's position after each iteration, X * X(t) represents the optimal position, X(t) represents the current position of the whale, δ1 and δ2 are both random numbers in the interval [0, 1], and X... rand (t) represents a randomly selected whale individual from the population.

[0057] Step S5 specifically includes:

[0058] S51. Settings include population size N and maximum number of iterations T. max The initial variables are determined by initializing the whale population using the optimal point set method, and the fitness is calculated.

[0059] S52. Select elite whale individuals using an elite reverse learning strategy, calculate the dynamic boundary of the elite whale individuals, and calculate the reverse solution to obtain the reverse population.

[0060] S53. Compare elite whale individuals and reverse populations, and select the top N whale populations with better fitness to participate in the next iteration as new whale populations.

[0061] S54. Update the coefficient vector of the whale population to control the movement pattern of the whale population and update the convergence factor; the expression of the coefficient vector is as follows:

[0062]

[0063] in: Let be the distance vector between the individual whale and its prey. This is the position vector of the currently obtained optimal solution. Let be the position vector of the individual whale in the t-th iteration. Let be the position vector of the individual whale in the (t+1)th iteration. Both are coefficient vectors;

[0064] The expression for updating the convergence factor is as follows:

[0065]

[0066] in: For the reverse solution, c is a random number in the interval [0, 1], ub j lb j These are the upper and lower limits of the dynamic boundary, respectively;

[0067] S55. Generate a random number p in the interval [0, 1] to determine the fishing method of the whale population; the determination result includes any of the following:

[0068] The first scenario: When p < 0.5, the IWOA algorithm performs the search encirclement phase;

[0069] Among them, if In this case, a whale individual is randomly selected from the whale population for optimization, and its position is updated using the following formula:

[0070]

[0071] Where: X(t+1) is the position of the individual whale, D is the spatial dimension, and X * (t) represents the optimal position, and e is the control factor;

[0072] like In this case, a whale individual is randomly selected from the whale population for optimization, and its position is updated using the following formula:

[0073]

[0074] Where: δ(||Yc i ||) is a Gaussian function, Y is the input vector, and c i Let ||Yc be the center point of the neuron in the hidden layer of the RBF neural network. i || represents the input vector Y to the center point c i Euclidean distance, σ i The width of the Gaussian kernel in the hidden layer of the RBF neural network;

[0075] The second method: When p ≥ 0.5, the IWOA algorithm updates the position in a spiral manner according to the following formula:

[0076]

[0077] in: Let a be a random vector on [0, 1], a be the convergence factor, and t be the number of iterations;

[0078] S56. Iterate the IWOA algorithm and determine whether the maximum number of iterations has been reached; the determination result includes any of the following:

[0079] First scenario: If the number of iterations t < T maxThen, according to the following formula, the random difference variation perturbation of the whale population is performed. After the number of iterations t = t + 1, the process returns to step S4 to continue the iteration.

[0080]

[0081] in: For the reverse solution, rand() is a random function;

[0082] The second scenario: If the number of iterations t < T max If the condition is not met, then output the globally optimal solution.

[0083] A fault classification system for transmission line equipment based on DWT-RBF, which applies the above-mentioned method, the system comprising:

[0084] The simulation system construction module is used to build a DWT-RBF simulation system for three-phase series compensated power.

[0085] The signal decomposition module is used to acquire the sampled signal of transient current during transmission line faults. It uses DWT to decompose the sampled signal, then uses an improved wavelet thresholding algorithm to denoise it, and finally obtains the energy value of the sampled signal.

[0086] The normalization module is used to normalize the energy value of the sampled signal and obtain the maximum and minimum values ​​of the energy value;

[0087] The algorithm optimization module is used to optimize the DWT-RBF simulation system based on the IWOA algorithm; the IWOA algorithm optimization includes any one or any combination of the following: optimal point set optimization for initializing the population, elite reverse learning strategy, nonlinear parameter control, and random difference mutation strategy;

[0088] The fault classification module is used to input energy values ​​into the optimized DWT-RBF neural network based on the optimized DWT-RBF simulation system to complete fault classification.

[0089] A fault classification device for transmission line equipment based on DWT-RBF, the device including a processor and a memory;

[0090] The memory is used to store computer program code and to transmit the computer program code to the processor;

[0091] The processor is used to execute the above-described fault classification method for transmission line equipment based on DWT-RBF according to the instructions in the computer program code.

[0092] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0093] This invention discloses a method, system, and equipment for classifying transmission line equipment faults based on DWT-RBF. The method first establishes a simulation system for three-phase series-compensated power and defines fault types. Then, it acquires sampled signals of transient currents during transmission line faults, decomposes them using DWT, denoises them using an improved wavelet thresholding algorithm, and obtains the energy value of the sampled signals. This energy value is then normalized. Finally, the DWT-RBF simulation system is optimized using the IWOA algorithm, and the energy value is input to complete fault classification. In application, this design accelerates fault classification response speed by decomposing the sampled signals and extracting fault features using DWT, and improves recognition accuracy by employing a wavelet thresholding denoising algorithm. Furthermore, the optimization of the RBF model using the IWOA algorithm enhances the model's convergence speed and generalization fitting ability, resulting in better network performance and the ability to accurately and quickly classify different types of transmission line equipment faults. Attached Figure Description

[0094] Figure 1 This is a flowchart of the method steps of the present invention.

[0095] Figure 2 This is a schematic diagram of the structure of the three-phase series compensation power system in Embodiment 1 of the present invention.

[0096] Figure 3 This is a schematic diagram of wavelet three-level decomposition constructed by MRA in Embodiment 1 of the present invention.

[0097] Figure 4 This is a schematic diagram of the fault classification and processing structure in Embodiment 1 of the present invention.

[0098] Figure 5 This is a schematic diagram of the fast Fourier transform of the A-phase fault signal in Embodiment 1 of the present invention.

[0099] Figure 6 This is a schematic diagram of the db4 wavelet three-level decomposition technique for the A-phase fault signal in Embodiment 1 of the present invention.

[0100] Figure 7 This is a schematic diagram showing the relationship between the fault initiation angle and wavelet transform energy in Embodiment 1 of the present invention.

[0101] Figure 8 This is a schematic diagram of the radial basis network structure in Embodiment 1 of the present invention.

[0102] Figure 9 This is a schematic diagram of the network performance of the training samples in Phase A of Embodiment 1 of the present invention.

[0103] Figure 10 This is a schematic diagram of the system structure of the present invention.

[0104] Figure 11This is a schematic diagram of the device structure of the present invention.

[0105] In the diagram: Simulation system construction module 1, signal decomposition module 2, normalization module 3, algorithm optimization module 4, fault classification module 5, processor 6, memory 7, computer program code 71. Detailed Implementation

[0106] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0107] Example 1:

[0108] See Figure 1 A fault classification method for transmission line equipment based on DWT-RBF, comprising:

[0109] S1. Construct a DWT-RBF simulation system for three-phase series compensated power;

[0110] See Figure 2 The DWT-RBF simulation system for the three-phase series-compensated power system was built in MATLAB or SIMULINK. It is a three-phase, 60Hz, 735kV power system that transmits power from a power plant consisting of six 350MVA generators to an equivalent system via a 600km transmission line. The transmission line consists of two 300km lines connected between busbars B1, B2, and B3. To increase transmission capacity, each line is compensated in series with capacitors representing 40% of the line reactance. These two lines are also compensated in parallel with 330Mvar parallel reactances. The parallel and series compensation devices are located at substation B2. A 300MVA-735 / 230kV transformer supplies power to a 230kV-250MW load. Each series compensation group is protected by metal oxide varistors (MOV1 and MOV2). CB1 and CB2 represent two circuit breakers on line 1. The simulation time is 50μs, only a quarter cycle, and the fault resistance is 1Ω.

[0111] In this embodiment, it is also necessary to define the fault types. Ten types of faults (ag, bg, cg, abg, ab, bcg, bc, acg, ca, abc) and four fault initiation angles (0, 45, 90, 135) occur on the transmission line between the 300-kilometer-long busbars (B1-B2). All fault types and four fault locations are processed from the distance of the transmitting end (25 km, 100 km, 175 km, 250 km) before and after series compensation. Therefore, the total number of cases in this embodiment is 160 (10*4*4=160).

[0112] In this embodiment, the sampling signal of the transient current during a transmission line fault is the current signal and fault time collected in real time by a current sensor installed on the transmission line or by an online monitoring system.

[0113] S2. Obtain the sampling signal of transient current during transmission line faults, decompose the sampling signal using DWT, then denoise it based on the improved wavelet thresholding algorithm, and obtain the energy value of the sampling signal.

[0114] In this embodiment, the principle of DWT signal decomposition is as follows:

[0115] DWT consists of a scaling function and a wavelet function, expressed as follows:

[0116] φ jk =2 j / 2 φ 00 (2 j zk);

[0117]

[0118] Where: φ jk and Let be the scaling function and wavelet function for scaling parameter j and translation parameter k, respectively, and z be the current sampling point. φ is a basis function of the scaling function. 00 These are the basis functions of the wavelet function, also known as the mother wavelet;

[0119] Multi-resolution analysis (MRA) provides different temporal and frequency resolutions at each analysis level. MRA uses a scaling function φ and wavelet functions. This allows the sampled signal to be extended to:

[0120]

[0121] Where: X(z) is the sampled signal at sampling point z, a jk d jk These are the approximation coefficients and detail coefficients when the scale parameter is j and the translation parameter is k, respectively, collectively referred to as DWT coefficients;

[0122] The basic idea of ​​DWT is to decompose the signal into different scales and perform low-pass and high-pass filtering on the sub-signals of each scale. The output of the high-pass filter contains the detail coefficients d of the input signal, while the output of the low-pass filter is the approximation coefficients a. By downsampling the signal through the filter, the frequency resolution is improved, as shown below:

[0123] The low-frequency component a at the j-th scale when the sampling point is z j (z) and high-frequency component d j (z), its expression is as follows:

[0124]

[0125] Where: a j+1 (z) represents the low-frequency component at the (j+1)th scale when the sampling point is z. G0 and H0 are coefficients, corresponding to the filter coefficients of the low-pass filter and the high-pass filter, respectively.

[0126] See Figure 3 The MRA-constructed wavelet three-level decomposition, using db4 as the mother wavelet, achieves the third level of signal decomposition. At each analysis level, the signal is decomposed by applying consecutive low-pass filters G0 and H0. After each filtering, half of the samples are eliminated, and decomposition is achieved at the end row, ultimately yielding detail coefficients d1, d2, d3 and approximation coefficients a3, thus simultaneously extracting time and frequency information. Therefore, lower frequencies can achieve higher time resolution. Zero-phase wavelets like db4 have the smallest energy envelope width, confining their energy to a minimum time frame. Thus, zero-phase wavelets offer the highest resolution; the total energy represents both the potential and kinetic energy of the wavelet signal. Assuming the wavelet signal is the output of a velocity detector, the partial energy represents only kinetic energy, while the total energy better represents the wavelet energy propagating through the Earth. Although the energy distribution is the same, the partial energy used (only the real signal) introduces a bias, making the zero-phase wavelet superior to other wavelets of the same class. In training and testing, instead of using detailed and approximate information, the total energy of each decomposition level is used as a feature for data classification, and the ED (Equation of Detailed Energy) representing the detailed energy of decomposition level q is used. q And EA represents the approximate energy of the decomposition level p. p The sum of these values ​​represents each signal. The energy of the wavelet coefficients varies at different scales depending on the input signal, and is contained in both the approximation and detail portions.

[0127] The transient current amplitude during a line fault is weak and easily affected by noise. To ensure the accuracy of the fault identification algorithm, an improved wavelet thresholding algorithm is used for noise filtering. The transient current signal acquired by the equipment consists of a superposition of the original signal and noise signal. After wavelet transform, approximate and detail components are obtained. Each component still consists of the original signal and noise signal, and the noise is mostly high-frequency, located in the detail components of the wavelet transform. The magnitude of the original signal is often greater than that of the noise signal, and the actual noise is composed of multiple sources, which can be regarded as the sum of variables with different independent probability distributions. According to the central limit theorem, the noise distribution will approach a Gaussian distribution. Therefore, the denoising process in this embodiment is as follows:

[0128] S21. Based on the normal distribution principle of Gaussian distribution, the noise of the sampled signal is distributed in the interval [-3σ, 3σ]. A maximum-minimum threshold is used to denoise the data of each layer component within the interval [-3σ, 3σ]. The denoising expression is as follows:

[0129]

[0130] Where: λ is the maximum and minimum threshold, and N is the number of sampling points;

[0131] S22. Reconstruct the denoised sampled signal to obtain the denoised signal; the expression for the reconstruction is as follows:

[0132] σ=median(s(z)) / 0.6745, z=0, 1,...,N-1;

[0133] Where: σ is the approximate variance of the sampled signal containing noise, median is the median of the data, s(z) is the data point of a certain layer component, and z is the current sampling point;

[0134] In the application of wavelet thresholding denoising algorithms, it is necessary to consider not only the distribution range of noise but also the influence of variance σ on the threshold. Here, the new threshold is obtained by multiplying σ of each layer component by the maximum and minimum thresholds of each layer. Threshold calculation methods can be divided into hard thresholding and soft thresholding. Hard thresholding calculates the threshold λ and sets points with moduli less than λ in the wavelet component to be denoised to zero, while retaining points with moduli greater than λ. Soft thresholding subtracts λ from points with moduli greater than λ and sets points with moduli less than λ to zero. Compared to hard thresholding, soft thresholding removes noise more smoothly and ensures waveform restoration. Therefore, this technical solution preferably uses soft thresholding for denoising.

[0135] S23. Calculate the energy value of the sampled signal; the calculation formula is as follows:

[0136]

[0137] Where: X 2 (z) is the sampled signal at sampling point z.

[0138] S3. Normalize the energy values ​​after decomposition and denoising to obtain the maximum and minimum energy values.

[0139] Furthermore, the formulas for calculating the maximum and minimum energy values ​​are as follows:

[0140]

[0141] Where: X i Let X be the energy value of the i-th sampled signal. max X represents the maximum energy of the sampled signal.min This represents the minimum energy of the sampled signal.

[0142] See Figure 4 After normalizing the energy value to between 0 and 1, it is input into the IWOA-optimized RBF network for fault classification. The output target is 0 when normal and 1 when faulty. For example, the output target for a BCG fault is (0 11 1).

[0143] S4. Optimize the DWT-RBF simulation system based on the IWOA algorithm; the IWOA algorithm optimization includes any one or any combination of the following: optimal point set optimization for initializing the population, elite reverse learning strategy, nonlinear parameter control, and random difference mutation strategy;

[0144] In this embodiment, the principle of the RBF neural network is as follows:

[0145] RBF neural networks are typical multi-layer feedforward neural networks. They use a linear approach to map problems that cannot be solved in low-dimensional space to high-dimensional space for solution. An RBF neural network consists of an input layer, hidden layers, and an output layer, where the hidden layers are composed of RBF network neurons. An RBF neural network is represented as follows:

[0146]

[0147] Where: Y is the input vector, Y∈R n ω i Let be the weights between the i-th hidden layer neuron and the output layer in the RBF neural network. c i represents the center point of the i-th neuron in the RBF neural network, m represents the number of neurons in the hidden layer, and ||Yc i || represents the input vector Y to the center point c i Euclidean distance; δ(||Yc i ||) is R + → A linear function of R, commonly known as the Gaussian function, is represented as follows:

[0148]

[0149] Where: σ i The width of the Gaussian kernel in the hidden layer of the RBF neural network;

[0150] The training speed and accuracy of the RBF neural network prediction model are related to the linear weights ω of the neurons in the output layer. i The center point c of the neurons in the hidden layer i and the Gaussian kernel width σ of the hidden layer i Regarding this, clustering algorithms are typically used to first determine c. i and σ iThen, the least squares method is used to obtain ω. i However, this method of obtaining the parameters of the RBF neural network will slow down the convergence speed of the neural network. IWOA has a strong optimization ability. Therefore, in this invention, IWOA is used to optimize these three parameters, which can speed up the training speed of the prediction model and improve the prediction accuracy.

[0151] In this embodiment, the principle of the WOA algorithm is as follows:

[0152] WOA is a swarm intelligence optimization algorithm that simulates the hunting behavior of humpback whales. The basic model of this algorithm is described as follows: Assuming there are only n individual whales in the d-dimensional search space, the i-th individual in iteration t is represented as... Where T max This represents the maximum number of iterations.

[0153] The algorithm consists of three phases: the humpback whale surrounding the prey phase, the local search phase, and the global search phase.

[0154] The stage in which a humpback whale surrounds its prey is described as follows:

[0155]

[0156]

[0157] in: Let be the distance vector between the individual whale and its prey; This is the position vector of the currently obtained optimal solution; Let be the position vector of the individual whale in the t-th iteration; Let be the position vector of the individual whale in the (t+1)th iteration; Both are coefficient vectors used to control the whale's swimming style; |·| is the absolute value.

[0158] In the above formula and The calculation process is as follows:

[0159]

[0160] in: Let be a random vector on the interval [0, 1]. The convergence factor decreases linearly from 2 to 0 as the number of iterations increases during the iteration process.

[0161] The local search phase is represented as follows:

[0162]

[0163] in: is the distance from the i-th whale to its prey (the currently obtained optimal solution); b is a constant defining the shape of the logarithmic spiral, and l is a random number in [-1, 1].

[0164] When a humpback whale approaches its prey, its contraction and spiraling position updates occur simultaneously. In this embodiment, it is assumed that the humpback whale has a 50% probability of both contraction and spiraling position updates. The mathematical model is described below:

[0165]

[0166] Where: X(t+1) is the position of the individual whale, D is the spatial dimension, and X * (t) represents the optimal position, and e represents the control factor.

[0167] The global search phase is represented as follows:

[0168] The global search mechanism randomly selects individual whales to learn from, updates their positions, and combines this information with... Perform a global search; when At this stage, whales will perform random and global searches to avoid getting trapped in local optima. The mathematical model for this stage is shown below:

[0169]

[0170] in: This represents the distance vector from a randomly selected individual whale to its prey. This is a vector representing the location of a randomly selected individual whale.

[0171] Although WOA has advantages such as fewer parameters, simple structure, and good convergence performance, it still has the following problems:

[0172] 1. Uneven population initialization: WOA randomly initializes the population, which may result in uneven or overly concentrated population distribution. This is not conducive to finding the global optimum and affects the efficiency and accuracy of the optimization process.

[0173] 2. Slow convergence speed: The parameter a changes linearly from 2 to 0, which slows down the convergence speed of the algorithm in the later stages;

[0174] 3. Difficulty in escaping local optima: In the later stages of the algorithm, as the parameter A continues to decrease, the update range of the current optimal solution position also continues to shrink. Once trapped in a local optimum, it will be difficult to escape.

[0175] To address the above shortcomings, this embodiment optimizes the WOA algorithm by using any one or any combination of the following: optimal point set optimization for initializing the population, elite reverse learning strategy, nonlinear parameter control, and random difference mutation strategy.

[0176] The optimal point set is used to optimize the initialization of the population, as detailed below:

[0177] In swarm intelligence optimization algorithms, good population initialization optimizes the process of finding the global optimum and improves the accuracy of the optimal solution. A uniform and diverse population distribution is beneficial to improving the algorithm's optimization performance. Therefore, a set of optimal points is introduced to initialize the starting positions of individual whales, as follows:

[0178] Let G s Let G be a unit cube in s-dimensional Euclidean space, and let the optimal point g∈G s The optimal point set is as follows:

[0179] Pn(k)={({g1·k}, {g2·k},…,{g s ·k}), 1≤k≤n};

[0180] Where: P n (k) is the set of best points, {g} s ·k} is the value of g s The fractional part of the result obtained by k, g is the optimal point, n is the number of samples, and k is the k-th sample out of n samples;

[0181] The deviation then satisfies the following:

[0182]

[0183] in: Let C(g, ε) be the deviation, and let C(g, ε) be a constant related to g and ε, where ε is any positive number and g is the optimal point.

[0184]

[0185] Wherein: g j Let be the optimal point in the j-th dimension; j is the spatial dimension, and p is the smallest prime number that satisfies (p-3) / 2≥s;

[0186] For an unknown uniformly distributed object, the deviation of n points selected randomly is as follows:

[0187] O(n -1 / 2 (loglogn) 1 / 2 );

[0188] From the formula for randomly selecting the position vector of individual whales in the global search mechanism, it can be seen that using a set of optimal points to select n points results in a deviation of O(n^2). -1+ε Therefore, the deviation of selecting points using the optimal point set method is very small, only at the square root level of random point selection. Initializing the individual whale positions through the optimal point set can increase population diversity, improve the global search capability of the whale optimization algorithm to a certain extent, and accelerate the convergence speed of the algorithm.

[0189] The location of individual whales in the population, based on the optimal point set method, is mapped to the search space as follows:

[0190]

[0191] in: Let ub be the position of the k-th whale in the j-th dimension. j lb j These represent the upper and lower limits of the search range in the j-th dimension, respectively.

[0192] The elite reverse learning strategy is as follows:

[0193] The basic idea of ​​reverse learning is to find the corresponding reverse solution based on the current solution, compare the fitness values ​​of the original solution and the reverse solution, and save the better solutions. Introducing reverse solutions can expand the search space of the algorithm, but because reverse learning has a certain degree of blindness, excessive calculation of reverse solutions will waste the algorithm's search time. Therefore, an elite strategy is introduced: the initialized whale population is sorted by fitness, individuals with higher fitness are selected as elite individuals, and their reverse solutions are calculated. The better individuals from the elite solutions and their reverse solutions are selected as elite individuals for the next generation of the population. This ensures that the population continuously approaches the optimal value while reducing unnecessary computation. Specifically:

[0194] Assume that each whale after each iteration is an elite individual, and its expression is as follows:

[0195]

[0196] in: Let be the position vector of the i-th whale in the j-th dimensional space, where D is the spatial dimension and N is the population size. Let N be the position of the Nth whale in D-dimensional space;

[0197] The strategy formula for calculating the inverse solution of the elite solution is as follows:

[0198]

[0199] in: and All are reverse solutions, where c is a random number in the interval [0,1], and ub j lb j These are the upper and lower bounds of the dynamic boundary, respectively. The advantage of dynamic boundaries is that they can compensate for the inability of fixed boundaries to preserve search experience, which helps to shorten search time.

[0200] If the inverse solution exceeds the boundary, it is reset using random generation; the reset formula is as follows:

[0201]

[0202] Where: rand() is a random function.

[0203] The nonlinear parameter control is specifically as follows:

[0204] In the basic Whale Optimization (WOA) algorithm, the global exploration and local exploitation capabilities of the whale pod primarily rely on the convergence factor 'a'. In the early stages of the WOA algorithm, the number of optimal individuals in the pod is relatively small, necessitating an enhanced global exploration capability. However, in the later stages, the pod is more likely to encounter local optima, requiring adjustments to the convergence factor to strengthen local exploitation and escape local optima. But since the convergence factor 'a' decreases linearly from 2 to 0, it cannot effectively balance the algorithm's global exploration and local exploitation capabilities. Therefore, a non-linear convergence factor is used to modify its variation, thereby enhancing local exploitation and escaping local optima, as detailed below:

[0205] The expression for the nonlinear convergence factor is as follows:

[0206]

[0207] Where: a is the nonlinear convergence factor, b and e are both control factors, and T max t represents the maximum number of iterations, and t represents the current number of iterations.

[0208] In the early stages of the algorithm, the convergence factor decreases rapidly, enhancing the global search capability of the population. In the later stages of iteration, the value converges slowly, which is more conducive to local exploration and improves the convergence accuracy of the algorithm.

[0209] The specific random difference mutation strategy is as follows:

[0210] Traditional Whale Algorithm (WOA) is prone to premature convergence. Therefore, the purpose of employing a random difference mutation strategy is to prevent the Whale Algorithm from getting stuck in local optima in its later stages. Specifically, each individual whale, after hunting prey or using a bubble net, updates its position to obtain its current optimal position. Before the maximum number of iterations is reached, a random difference mutation strategy is used to perturb the population, obtaining the optimal position again before and after the change. This overcomes the weakness of traditional WOA in getting stuck in local optima and accelerates the convergence speed of the algorithm. For the search phase, the random difference mutation strategy is shown in the following equation:

[0211] X(t+1)=δ1(X * (t)-X(t))+δ2(X rand (txt));

[0212] Where: X(t+1) is the whale's position after each iteration, X *X(t) represents the optimal position, X(t) represents the current position of the whale, δ1 and δ2 are both random numbers in the interval [0, 1], and X... rand (t) represents a randomly selected whale individual from the population.

[0213] S5. Based on the optimized DWT-RBF simulation system, the energy value is input into the optimized DWT-RBF neural network to complete the fault classification.

[0214] Furthermore, in this embodiment, the IWOA algorithm flow is as follows:

[0215] S51. Settings include population size N and maximum number of iterations T. max The initial variables are determined by initializing the whale population using the optimal point set method, and the fitness is calculated.

[0216] S52. Select elite whale individuals using an elite reverse learning strategy, calculate the dynamic boundary of the elite whale individuals, and calculate the reverse solution to obtain the reverse population.

[0217] S53. Compare elite whale individuals and reverse populations, and select the top N whale populations with better fitness to participate in the next iteration as new whale populations.

[0218] S54. Update the coefficient vector of the whale population to control the movement pattern of the whale population and update the convergence factor; the expression of the coefficient vector is as follows:

[0219]

[0220] in: Let be the distance vector between the individual whale and its prey. This is the position vector of the currently obtained optimal solution. Let be the position vector of the individual whale in the t-th iteration. Let be the position vector of the individual whale in the (t+1)th iteration. Both are coefficient vectors;

[0221] The expression for updating the convergence factor is as follows:

[0222]

[0223] in: For the reverse solution, c is a random number in the interval [0, 1], ub j lb j These are the upper and lower limits of the dynamic boundary, respectively;

[0224] S55. Generate a random number p in the interval [0,1], and determine the whale's fishing method; the determination result includes any of the following:

[0225] The first scenario: When p < 0.5, the IWOA algorithm performs the search encirclement phase;

[0226] Among them, if In this case, a whale individual is randomly selected from the whale population for optimization, and its position is updated using the following formula:

[0227]

[0228] Where: X(t+1) is the position of the individual whale, D is the spatial dimension, and X * (t) represents the optimal position, and e is the control factor;

[0229] like In this case, a whale individual is randomly selected from the whale population for optimization, and its position is updated using the following formula:

[0230]

[0231] Where: δ(||Yc i ||) is a Gaussian function, Y is the input vector, and c i Let ||Yc be the center point of the neuron in the hidden layer of the RBF neural network. i || represents the input vector Y to the center point c i Euclidean distance, σ i The width of the Gaussian kernel in the hidden layer of the RBF neural network;

[0232] The second method: When p ≥ 0.5, the IWOA algorithm updates the position in a spiral manner according to the following formula:

[0233]

[0234] in: Let a be a random vector on [0, 1], a be the convergence factor, and t be the number of iterations;

[0235] S56. Iterate the IWOA algorithm and determine whether the maximum number of iterations has been reached; the determination result includes any of the following:

[0236] First scenario: If the number of iterations t < T max Then, according to the following formula, the random difference variation perturbation of the whale population is performed. After the number of iterations t = t + 1, the process returns to step S4 to continue the iteration.

[0237]

[0238] in: For the reverse solution, rand() is a random function;

[0239] The second method: If the number of iterations is t <Tmax If the condition is not met, then output the globally optimal solution.

[0240] In this embodiment, the effectiveness of the method is verified by simulating the current signal during a single-wire grounding fault in the AG circuit.

[0241] The Fast Fourier Transform (FFT) has limited ability to analyze signals and provide time-frequency information because it uses a global function `ejwt` to transform the signal from the time domain to the frequency domain. Therefore, all information in the time domain, which can help detect the start and duration of interference, is lost. The FFT of the A-phase fault signal is shown below. Figure 5 As shown (THD represents Total Harmonic Distortion), the wavelet three-level decomposition technique db4 is used to achieve this signal. Figure 6 As shown, it can be used to locate distortion events in a timely manner, and can detect the exact end time of distortion events even when the oscillation phenomenon decays to a very small value.

[0242] The energy of the current signal under normal, fault-free conditions is shown in the table below:

[0243] normal 1.26E+05 1.99E+05 1.63E+05

[0244] As can be seen from the table above, the energy values ​​of different phases are very close or almost the same. These values ​​are used as a reference for the different fault effects of noise at the signal energy value. In the case of a single-phase ground fault, although the other two phases change slightly, the energy of the affected line increases sharply from its normal state.

[0245] The signal energy during a single-phase ground fault is shown in the table below:

[0246]

[0247] The table above shows that, for example, during an AG fault, the maximum energy occurs in phase A. The fault initiation angle affects the current energy, but for the same fault type and location, such as an AG single-line ground fault occurring at a distance of 25km, the fault initiation angle has a greater impact on the three-phase current (A) signal energy value. The relationship between the fault initiation angle and wavelet transform energy is as follows: Figure 7 As shown, when there is a grounding fault in the double circuit, although the normal phase changes slightly, the energy of the affected double circuit increases sharply compared to the normal state.

[0248] The signal energy during a double-circuit w / wo ground fault is shown in the table below:

[0249]

[0250] As can be seen from the table above, for example, when ABG or AB fails, the minimum response to the fault increases to the normal phase C.

[0251] The signal energy during a three-wire fault is shown in the table below:

[0252]

[0253] Furthermore, for the precise design method, the number of hidden neurons is equal to the number of training samples, n. h =N=144, the training sample error is zero, but for larger neural network systems, the number of hidden neurons would be too large and unacceptable. However, an RBF network was established using IWOA, and after experiments, the network's distribution degree was 0.065, and the tolerance error was 0.01. The radial basis function network structure diagram established in this embodiment is as follows. Figure 8 As shown; the proposed network structure has three input signals, 108 hidden neurons, and 4 neurons in the output layer. The network performance of the training samples in phase A is as follows. Figure 9 As shown in the figure, the training objective and the network output are very close at this stage.

[0254] To test the generalization ability of the RBF network, we selected other samples not present in the training set for testing. The test results are shown in the table below:

[0255]

[0256] The examples in the table above demonstrate that all faulty elements can be correctly identified, thus the network performance of this technical solution is good.

[0257] The transmission line equipment typically includes line towers, conductors, insulators, line hardware, guy wires and tower foundations, grounding devices, lightning protection wires, circuit breakers, transformers, voltage transformers, etc. In this embodiment, the focus is on equipment such as conductors, transformers, switches, insulators, and circuit breakers that are prone to short-circuit faults. After a short circuit, their signal characteristics will show sudden increases in current, frequency oscillations, etc. By filtering and separating the signal characteristics, the fault type of the transmission circuit equipment can be identified.

[0258] Example 2:

[0259] A fault classification system for transmission line equipment based on DWT-RBF, which is applied to the method described in Example 1, the system comprising:

[0260] Simulation system construction module 1 is used to build a DWT-RBF simulation system for three-phase series compensated power and define fault types;

[0261] Signal decomposition module 2 is used to acquire the sampled signal of transient current during transmission line faults. The sampled signal is decomposed using DWT, then denoised based on an improved wavelet thresholding algorithm, and the energy value of the sampled signal is obtained.

[0262] Furthermore, the signal decomposition module 2 performs noise reduction according to the relevant steps S21-S23 in Embodiment 1:

[0263] Normalization module 3 is used to normalize the energy value after decomposition and denoising to obtain the maximum and minimum values ​​of the energy value.

[0264] Furthermore, the normalization module 3 performs calculations according to the relevant steps of step S3 in embodiment 1;

[0265] Algorithm optimization module 4 is used to optimize the DWT-RBF simulation system based on the IWOA algorithm; the IWOA algorithm optimization includes any one or any combination of the following: optimal point set optimization initialization population, elite reverse learning strategy, nonlinear parameter control, and random difference mutation strategy;

[0266] The algorithm optimization module 4 optimizes the algorithm according to the relevant steps of step S4 in embodiment 1:

[0267] The fault classification module 5 is used to input energy values ​​into the optimized DWT-RBF neural network based on the optimized DWT-RBF simulation system to complete fault classification.

[0268] Furthermore, the fault classification module 5 performs fault classification according to steps S51-S56 in Embodiment 1:

[0269] Example 3:

[0270] A fault classification device for transmission line equipment based on DWT-RBF, the device including a processor 6 and a memory 7;

[0271] The memory 7 is used to store computer program code 71 and to transmit the computer program code 71 to the processor 6;

[0272] The processor 6 is used to execute the DWT-RBF-based transmission line equipment fault classification method described in Embodiment 1 according to the instructions in the computer program code 71.

[0273] In this embodiment, a computer-readable storage medium is also included, which stores computer-executable instructions. When the computer-executable instructions are executed on a computer, the transmission line fault classification method based on DWT-RBF described in Embodiment 1 is implemented.

[0274] Generally, the computer instructions for implementing the method of the present invention can be carried on any combination of one or more computer-readable storage media. Non-transitory computer-readable storage media can include any computer-readable medium except for the signal itself, which is temporarily propagating.

[0275] Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EKROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0276] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smarttalk, and C++—as well as conventional procedural programming languages—such as the "C" language or similar programming languages. In particular, Python, suitable for neural network computation, and platform frameworks based on TensorFlow, PyTorch, etc., can be used. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer or to an external computer (e.g., via the Internet using an Internet service provider) through any type of network, including a local area network (LAN) or a wide area network (WAN).

[0277] For details regarding the aforementioned equipment and non-transitory computer-readable storage media, please refer to the specific description of a DWT-RBF-based transmission line equipment fault classification method and its beneficial effects, which will not be repeated here.

[0278] Although embodiments of the present invention have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A fault classification method for transmission line equipment based on DWT-RBF, characterized in that, include: S1. Construct a DWT-RBF simulation system for three-phase series compensated power; S2. Obtain the sampling signal of transient current during transmission line faults, decompose the sampling signal using DWT, then denoise it based on the improved wavelet thresholding algorithm, and obtain the energy value of the sampling signal. S3. Normalize the energy value of the sampled signal and obtain the maximum and minimum values ​​of the energy value; S4. Optimize the DWT-RBF simulation system based on the IWOA algorithm; The IWOA algorithm optimization includes any one or any combination of the following: optimal point set optimization for initializing the population, elite reverse learning strategy, nonlinear parameter control, and random difference mutation strategy; S5. Based on the optimized DWT-RBF simulation system, input the energy value into the optimized DWT-RBF neural network to complete the fault classification; In step S2, the improved wavelet thresholding algorithm for noise reduction specifically includes: S21. Based on the Gaussian distribution and normal distribution principle, the noise of the sampled signal is distributed in... Within the range, the maximum and minimum thresholds are used to evaluate the components of each layer. The sampled data within the interval is denoised; the denoising expression is as follows: ； in: For the maximum and minimum threshold, This represents the number of sampling points; S22. Reconstruct the denoised sampled signal to obtain the denoised signal; the expression for the reconstruction is as follows: ； in: This is an approximate variance of the noisy sampled signal. The median value. For the data points of the components, This is the current sampling point; S23. Calculate the energy value of the sampled signal; the calculation formula is as follows: ； in: For sampling points The sampling signal at that time; Step S5 specifically includes: S51, Settings including population size With maximum number of iterations The initial variables are determined by initializing the whale population using the optimal point set method, and the fitness is calculated. S52. Select elite whale individuals using an elite reverse learning strategy, calculate the dynamic boundary of the elite whale individuals, and calculate the reverse solution to obtain the reverse population. S53. Compare elite individual whales and reverse populations, selecting those with better fitness. One whale population will participate in the next iteration as a new whale population; S54. Update the coefficient vector of the whale population to control the movement pattern of the whale population and update the convergence factor; the expression of the coefficient vector is as follows: ； ； in: Let be the distance vector between the individual whale and its prey. This is the position vector of the currently obtained optimal solution. For the first The position vector of the individual whale in the next iteration. For the first The position vector of the individual whale in the next iteration. , Both are coefficient vectors; The expression for updating the convergence factor is as follows: ； in: The solution is the reverse. for Random numbers within the interval , These are the upper and lower limits of the dynamic boundary, respectively; S55, the generation interval is random numbers To determine the fishing methods of whale populations; the determination result includes any of the following: The first type: when When this happens, the IWOA algorithm performs the search encirclement phase; Among them, if In this case, a whale individual is randomly selected from the whale population for optimization, and its position is updated using the following formula: ； in: Location of an individual whale. For spatial dimensions, For the best position, As a control factor; like In this case, a whale individual is randomly selected from the whale population for optimization, and its position is updated using the following formula: ； in: It is a Gaussian function. For the input vector, The center point of the neurons in the hidden layer of the RBF neural network. For input vectors To the center point European distance, The width of the Gaussian kernel in the hidden layer of the RBF neural network; The second type: when The IWOA algorithm then updates the position in a spiral manner according to the following formula: ； in: for random vectors on, The convergence factor is This represents the number of iterations. S56. Iterate the IWOA algorithm and determine whether the maximum number of iterations has been reached; the determination result includes any of the following: The first method: If the number of iterations... Then, according to the following formula, the random difference variation perturbation of the whale population is applied, and the number of iterations is set to... Then, return to step S4 to continue the iteration; ； in: The solution is the reverse. It is a random function; The second method: if the number of iterations... If the condition is not met, then output the globally optimal solution.

2. The method for classifying faults in transmission line equipment based on DWT-RBF according to claim 1, characterized in that: In step S3, the formulas for calculating the maximum and minimum energy values ​​are as follows: ； in: For the first The energy value of each sampled signal The maximum energy of the sampled signal. This represents the minimum energy of the sampled signal.

3. The method for classifying faults in transmission line equipment based on DWT-RBF according to claim 1, characterized in that: The optimal point set optimization initialization of the population in step S4 refers to: initializing the starting position of individual whales by introducing the optimal point set method, as follows: set up for The unit cube in Euclidean space, if the best point The optimal point set is as follows: ； in: For the best collection of points, To obtain The decimal part of the result For best results, For the sample size, for The first sample One sample; The deviation then satisfies the following: ； in: For deviation, It is a constant. For any positive number, Best point; ； in: For the first The best spot for Vie; For spatial dimensions, It is the smallest prime number; For objects with unknown uniform distribution, points are selected randomly. The deviations at each point are as follows: ； The location of individual whales in the population, based on the optimal point set method, is mapped to the search space as follows: ； in: For the first The first dimension in space The location of the whale. , The first The upper and lower limits of the search range in 3D space.

4. The method for classifying faults in transmission line equipment based on DWT-RBF according to claim 1, characterized in that: The elite reverse learning strategy in step S4 refers to: by introducing an elite strategy, the initialized whale population is sorted according to fitness, and individuals with higher fitness are selected as elite individuals. The reverse solution of the elite solution is calculated, and the better individuals in the elite solution and the reverse solution are selected as elite individuals in the next generation population, as detailed below: Assume that each whale after each iteration is an elite individual, and its expression is as follows: ； in: For the first The first dimension in space The position vector of the whale. For spatial dimensions, For population size, for The first dimension in space The location of the whale; The strategy formula for calculating the inverse solution of the elite solution is as follows: ； ； in: and All are reverse solutions. for Random numbers within the interval , These are the upper and lower limits of the dynamic boundary, respectively; If the inverse solution exceeds the boundary, the inverse solution is reset using random generation; the reset formula is as follows: ； in: It is a random function.

5. The method for classifying faults in transmission line equipment based on DWT-RBF according to claim 1, characterized in that: The nonlinear parameter control in step S4 refers to: by changing the amount of change of the nonlinear convergence factor, the local exploitation capability is enhanced to escape the local optimum, as detailed below: The expression for the nonlinear convergence factor is as follows: ； in: It is a nonlinear convergence factor. , All are control factors. The maximum number of iterations, This represents the current iteration number.

6. The method for classifying faults in transmission line equipment based on DWT-RBF according to claim 1, characterized in that: The random difference mutation strategy in step S4 refers to: using a random difference mutation strategy to randomly mutate and perturb the whale population during the search phase, and then obtaining the optimal position before and after the change again, as detailed below: For the search phase, the random difference mutation strategy is as follows: ； in: This represents the whale's position after each iteration. For the best position, This is the whale's current location. , All Random numbers within the interval A whale individual randomly selected from the population.

7. A fault classification system for transmission line equipment based on DWT-RBF, characterized in that: The system is applied to the method according to any one of claims 1-6, the system comprising: The simulation system construction module (1) is used to build a DWT-RBF simulation system for three-phase series compensation power. The signal decomposition module (2) is used to obtain the sampling signal of the transient current during the fault of the transmission line, decompose the sampling signal using DWT, then denoise it based on the improved wavelet threshold algorithm, and obtain the energy value of the sampling signal. The signal decomposition module (2) performs noise reduction using an improved wavelet thresholding algorithm through the following steps: S21. Based on the Gaussian distribution and normal distribution principle, the noise of the sampled signal is distributed in... Within the range, the maximum and minimum thresholds are used to evaluate the components of each layer. The sampled data within the interval is denoised; the denoising expression is as follows: ； in: For the maximum and minimum threshold, This represents the number of sampling points; S22. Reconstruct the denoised sampled signal to obtain the denoised signal; the expression for the reconstruction is as follows: ； in: This is an approximate variance of the noisy sampled signal. The median value. For the data points of the components, This is the current sampling point; S23. Calculate the energy value of the sampled signal; the calculation formula is as follows: ； in: For sampling points The sampling signal at that time; The normalization module (3) is used to normalize the energy value of the sampled signal and obtain the maximum and minimum values ​​of the energy value; The algorithm optimization module (4) is used to optimize the DWT-RBF simulation system based on the IWOA algorithm; the IWOA algorithm optimization includes any one or any combination of the following: optimal point set optimization initialization population, elite reverse learning strategy, nonlinear parameter control, and random difference mutation strategy; The fault classification module (5) is used to input the energy value into the optimized DWT-RBF neural network based on the optimized DWT-RBF simulation system to complete the fault classification. The fault classification module (5) completes fault classification through the following steps: S51, Settings including population size With maximum number of iterations The initial variables are determined by initializing the whale population using the optimal point set method, and the fitness is calculated. S52. Select elite whale individuals using an elite reverse learning strategy, calculate the dynamic boundary of the elite whale individuals, and calculate the reverse solution to obtain the reverse population. S53. Compare elite individual whales and reverse populations, selecting those with better fitness. One whale population will participate in the next iteration as a new whale population; S54. Update the coefficient vector of the whale population to control the movement pattern of the whale population and update the convergence factor; the expression of the coefficient vector is as follows: ； ； in: Let be the distance vector between the individual whale and its prey. This is the position vector of the currently obtained optimal solution. For the first The position vector of the individual whale in the next iteration. For the first The position vector of the individual whale in the next iteration. , Both are coefficient vectors; The expression for updating the convergence factor is as follows: ； in: The solution is the reverse. for Random numbers within the interval , These are the upper and lower limits of the dynamic boundary, respectively; S55, the generation interval is random numbers To determine the fishing methods of whale populations; the determination result includes any of the following: The first type: when When this happens, the IWOA algorithm performs the search encirclement phase; Among them, if In this case, a whale individual is randomly selected from the whale population for optimization, and its position is updated using the following formula: ； in: Location of an individual whale. For spatial dimensions, For the best position, As a control factor; like In this case, a whale individual is randomly selected from the whale population for optimization, and its position is updated using the following formula: ； in: It is a Gaussian function. For the input vector, The center point of the neurons in the hidden layer of the RBF neural network. For input vectors To the center point European distance, The width of the Gaussian kernel in the hidden layer of the RBF neural network; The second type: when The IWOA algorithm then updates the position in a spiral manner according to the following formula: ； in: for random vectors on, The convergence factor is This represents the number of iterations. S56. Iterate the IWOA algorithm and determine whether the maximum number of iterations has been reached; the determination result includes any of the following: The first method: If the number of iterations... Then, according to the following formula, the random difference variation perturbation of the whale population is applied, and the number of iterations is set to... Then, return to step S4 to continue the iteration; ； in: The solution is the reverse. It is a random function; The second method: if the number of iterations... If the condition is not met, then output the globally optimal solution.

8. A fault classification device for transmission line equipment based on DWT-RBF, characterized in that: The device includes a processor (6) and a memory (7); The memory (7) is used to store computer program code (71) and to transmit the computer program code (71) to the processor (6). The processor (6) is used to execute the fault classification method for transmission line equipment based on DWT-RBF according to any one of the instructions in the computer program code (71).

Images