A distribution network fault detection method and device based on traveling wave signal

Abstract

The present invention discloses a distribution network fault detection method and device based on traveling wave signals, the method comprising the following steps: obtaining a fault traveling wave signal of the distribution network and determining the fault location of the distribution network; performing frequency decomposition of the fault traveling wave signal based on an optimization decomposition algorithm to obtain an instantaneous frequency; performing feature extraction on the instantaneous frequency based on a pre-constructed optimization fault detection model to provide the fault characteristics of the distribution network; providing the fault category corresponding to the fault traveling wave signal according to the fault characteristics of the distribution network. The present invention can accurately determine the specific location and category of the distribution network fault, and improve the accuracy and efficiency of fault detection.

Description

Technical Field

[0001] The present invention belongs to the technical field of distribution network fault detection, and in particular relates to a distribution network fault detection method and device based on traveling wave signals. Background Art

[0002] With the large-scale commissioning of power cables in distribution networks, the resulting difficulties in cable fault location and classification have become particularly prominent.

[0003] Traditional cable fault distance measurement is mainly based on offline methods. Due to the complexity of external equipment, limited scope of application, and low distance measurement accuracy, offline methods are difficult to meet the needs of smart grid development. In addition, when classifying cable faults, Hilbert-Huang transform (HHT) is mainly used, where the traditional HHT consists of empirical mode decomposition (EMD) and Hilbert transform. For example, patent CN106501668A discloses a traditional distribution network single-phase disconnection fault line selection method, including: using the symmetrical component method to calculate the negative sequence current of each feeder; when the amplitude of the negative sequence current is greater than the negative sequence setting value, the negative sequence current of each feeder is decomposed by EMD to obtain each order of intrinsic mode components IMF; each order of intrinsic mode components IMF is Hilbert transformed to obtain its corresponding instantaneous amplitude waveform; the change in the sum of the instantaneous amplitudes of the intrinsic mode components IMF of the set order of each feeder before and after the fault for m cycles is calculated; and the disconnection fault line is selected according to the size of the change. The complex signal is adaptively decomposed into a finite number of high-frequency and low-frequency intrinsic mode functions (IMFs) with meaningful instantaneous frequencies and modulated amplitudes or frequencies through EMD. The frequency components contained in each IMF component are related to the sampling frequency and change with the signal itself. However, the HHT algorithm will experience modal aliasing when performing modal decomposition. Modal aliasing means that the same IMF component contains components of different scales, which directly leads to the lack of sufficient physical meaning of the aliased IMF, resulting in confusion in the subsequent time-frequency distribution.

[0004] Therefore, how to provide a high-precision fault detection method is an urgent problem to be solved by those skilled in the art. Summary of the invention

[0005] In view of the defects existing in the above-mentioned prior art, the present invention provides a distribution network fault detection method based on traveling wave signals, which can accurately determine the specific location and type of distribution network faults and improve the accuracy and efficiency of fault detection.

[0006] In a first aspect, the present invention provides a distribution network fault detection method based on a traveling wave signal, which specifically comprises the following steps:

[0007] Obtain the fault traveling wave signal of the distribution network and determine the fault location of the distribution network;

[0008] Based on the optimization decomposition algorithm, the fault traveling wave signal is frequency decomposed to obtain the instantaneous frequency;

[0009] Based on the pre-built optimized fault detection model, the instantaneous frequency is feature extracted to give the fault characteristics of the distribution network;

[0010] According to the fault characteristics of the distribution network, the fault category corresponding to the fault traveling wave signal is given.

[0011] Further, the fault location of the distribution network is determined, including:

[0012] Determine the corresponding power cable according to the acquired fault traveling wave signal;

[0013] According to the determined power cable, the time difference of the fault traveling wave signal transmitted to the two ends of the corresponding power cable is given;

[0014] Based on the time difference of the fault traveling wave signal being transmitted to the two ends of the corresponding power cable and combined with the length of the corresponding power cable, the fault location on the power cable is given.

[0015] Furthermore, the fault traveling wave signal is frequency decomposed based on the optimization decomposition algorithm to obtain the instantaneous frequency, including:

[0016] A noise coefficient is given based on the fault traveling wave signal and a plurality of first white noises are generated accordingly;

[0017] combining the plurality of first white noises with the fault traveling wave signal respectively;

[0018] Performing empirical mode decomposition on each combined fault traveling wave signal to obtain a plurality of first intrinsic mode function components;

[0019] All first intrinsic mode function components are averaged, and the averaged results are Hilbert transformed to obtain instantaneous frequencies.

[0020] Furthermore, a noise coefficient is given based on the fault traveling wave signal and a plurality of first white noises are generated accordingly, including:

[0021] Based on the standard deviation of the fault traveling wave signal, the value range of the noise coefficient is determined;

[0022] Taking the noise coefficient as a particle and combining the value range of the noise coefficient, the particles are initialized to give a particle swarm composed of multiple random particles;

[0023] Calculate the fitness value of the particle swarm, give the local optimal fitness value of each particle, and take the minimum value of all local optimal fitness values ​​as the global optimal fitness value;

[0024] Update the position of the particle swarm based on the evolution function and using extreme value perturbations;

[0025] Repeat the iterative fitness value calculation and position update steps until convergence, giving the final global optimal fitness value;

[0026] Based on the final global optimal fitness value, the corresponding noise coefficient value is determined, and different Gaussian white noises are combined to obtain multiple first white noises of the noise coefficient.

[0027] Furthermore, the fitness value of the particle swarm is calculated to give the local optimal fitness value of each particle, including:

[0028] The noise coefficient value of each particle in the particle group is combined with the same Gaussian white noise to obtain multiple second white noises;

[0029] Combining the plurality of second white noises with the fault traveling wave signal to obtain a plurality of signals to be decomposed;

[0030] Perform empirical mode decomposition on multiple signals to be decomposed respectively to obtain multiple groups of second intrinsic mode function components.

[0031] Based on the fitness value function, multiple groups of second intrinsic mode function components and fault traveling wave signals are analyzed and processed to give the local optimal fitness value of each particle.

[0032] Furthermore, the fitness value function satisfies the following relationship:

[0033]

[0034] Where RRMSE is the fitness value, N1 is the total number of the second intrinsic mode function components, x0(k) is the fault traveling wave signal, c max (k) is the second intrinsic modal function component with the highest correlation coefficient with the fault traveling wave signal in the kth group of second intrinsic modal function components obtained by empirical mode decomposition, k∈[1,K], K is the number of groups of second intrinsic modal function components, that is, the number of signals to be decomposed;

[0035] The evolution function satisfies the following relationship:

[0036]

[0037] In the formula, is the position of the particle, indicating the position of the i-th particle in the d-th dimension in the t+1-th iteration, i=1,2,…,m, d=1,2,…,D, m is the number of particles in the particle swarm, and D is the set search dimension; ω is the momentum inertia coefficient; c1 and c2 are non-negative learning factors; r1, r2, r3, and r4 are all random numbers that obey the U(0,1) distribution; p is the extreme point of the particle individual and the population, t0 and t gare the evolutionary stagnation steps of individual particle extrema and global extrema respectively; T0 and T g are the stagnation step thresholds that need to be disturbed for individual particle extrema and global extrema respectively.

[0038] Furthermore, the plurality of first white noises are respectively combined with the fault traveling wave signal to satisfy the following relationship:

[0039]

[0040] Where: x m (t) is the fault traveling wave signal after combining with the first white noise, x(t) is the fault traveling wave signal, n m (t) is the first white noise, A noise is the noise factor, δ (t) is Gaussian white noise.

[0041] Furthermore, the pre-construction of the optimized fault detection model includes:

[0042] Obtain the historical fault traveling wave signal of the distribution network and give the corresponding historical instantaneous frequency based on the optimization decomposition algorithm;

[0043] Normalize the historical instantaneous frequency to obtain normalized data and divide it into training set, validation set and test set;

[0044] Train the neural network model through the training set and give the trained neural network model;

[0045] The trained neural network model is verified by the validation set until the training end conditions are met, and the performance of the neural network model is tested using the test set to obtain the optimized fault detection model.

[0046] Furthermore, the neural network model is trained through the training set, and the trained neural network model is given, including:

[0047] S31, initializing the neural network model and loss function, giving initial model parameters and loss function coefficients; wherein the model parameters include a weight matrix and a bias vector;

[0048] S32, the neural network model performs forward propagation on the training data of the training set and calculates the value of the loss function;

[0049] S34, calculating the gradient of the loss function relative to the model parameters of the neural network model based on the back propagation algorithm, and using the gradient descent algorithm to update the model parameters and loss function coefficients of the neural network model according to the gradient to minimize the value of the loss function;

[0050] S35, repeating S32 and S33 until convergence, and providing a trained neural network model;

[0051] Among them, the loss function satisfies the following relationship:

[0052]

[0053] Where E is the loss function, α and β are the loss function coefficients, and n is the total number of network weights of the neural network model. l is the average network weight of the neural network model, w i is the network weight of the ith neural network model, W k represents the network weight vector of the kth iteration, J is the Jacobian matrix, ε(W k ) is based on W k The error vector, μ is the step size control parameter, which is 0.001, and γ is the number of effective parameters of the neural network model, γ=N-2β'tr(H) -1 , β' is the β value of the previous iteration, N is the total number of model parameters of the neural network model; H is the Hessian matrix of the loss function, and tr(H) represents the trace of the matrix H.

[0054] Furthermore, the fault detection model is optimized to satisfy the following relationship:

[0055]

[0056] In the formula, H s,t is the input normalized data of the optimized fault detection model at time t, is the current input cell state, W f is the weight matrix of the forget gate, b f is the bias of the forget gate, W i is the weight matrix of the input gate, b i is the bias of the input gate, W c is the weight matrix of the cell state, b c is the bias of the cell state, W o is the weight matrix of the output gate, b o is the bias of the output gate, tanh is the activation function, h t-1 is the fault characteristic output at time t-1, f t is the proportional coefficient of the forgotten information controlled in the forget gate, i tis the proportional coefficient of the input gate that controls the input information, c t is the status information after the cell status is updated, o t is the proportional coefficient of the output gate controlling the output information, σ is the Sigmoid activation function, C t is the cell state at time t, h t is the fault characteristic of the output.

[0057] Furthermore, according to the fault characteristics of the distribution network, the fault category corresponding to the fault traveling wave signal is given, including:

[0058] The fault features of the distribution network are input into the pre-built classifier to give the probability of mapping the fault features to each fault category;

[0059] The fault category with the highest probability is taken as the fault category corresponding to the fault traveling wave signal.

[0060] In a second aspect, the present invention further provides a distribution network fault detection device based on traveling wave signals, using the above-mentioned distribution network fault detection method based on traveling wave signals, the distribution network fault detection device comprises:

[0061] A signal acquisition and positioning module is used to acquire the fault traveling wave signal of the distribution network and determine the fault location of the distribution network;

[0062] A frequency decomposition module, which is used to perform frequency decomposition on the fault traveling wave signal based on an optimized decomposition algorithm to obtain an instantaneous frequency;

[0063] A feature extraction module is used to extract features of the instantaneous frequency based on a pre-built optimized fault detection model and provide fault features of the distribution network;

[0064] The fault identification module is used to give the fault category corresponding to the fault traveling wave signal according to the fault characteristics of the distribution network.

[0065] Furthermore, the signal acquisition and positioning module includes:

[0066] Determine the corresponding power cable according to the acquired fault traveling wave signal;

[0067] According to the determined power cable, the time difference of the fault traveling wave signal transmitted to the two ends of the corresponding power cable is given;

[0068] Based on the time difference of the fault traveling wave signal being transmitted to the two ends of the corresponding power cable and combined with the length of the corresponding power cable, the fault location on the power cable is given.

[0069] Furthermore, the frequency decomposition module includes:

[0070] A noise coefficient is given based on the fault traveling wave signal and a plurality of first white noises are generated accordingly;

[0071] combining the plurality of first white noises with the fault traveling wave signal respectively;

[0072] Performing empirical mode decomposition on each combined fault traveling wave signal to obtain a plurality of first intrinsic mode function components;

[0073] All first intrinsic mode function components are averaged, and the averaged results are Hilbert transformed to obtain instantaneous frequencies.

[0074] Furthermore, the frequency decomposition module includes:

[0075] Based on the standard deviation of the fault traveling wave signal, the value range of the noise coefficient is determined;

[0076] Taking the noise coefficient as a particle and combining the value range of the noise coefficient, the particles are initialized to give a particle swarm composed of multiple random particles;

[0077] Calculate the fitness value of the particle swarm, give the local optimal fitness value of each particle, and take the minimum value of all local optimal fitness values ​​as the global optimal fitness value;

[0078] Update the position of the particle swarm based on the evolution function and using extreme value perturbations;

[0079] Repeat the iterative fitness value calculation and position update steps until convergence, giving the final global optimal fitness value;

[0080] Based on the final global optimal fitness value, the corresponding noise coefficient value is determined, and different Gaussian white noises are combined to obtain a plurality of first white noises with noise coefficients.

[0081] Furthermore, the frequency decomposition module also includes:

[0082] The noise coefficient value of each particle in the particle group is combined with the same Gaussian white noise to obtain multiple second white noises;

[0083] Combining the plurality of second white noises with the fault traveling wave signal to obtain a plurality of signals to be decomposed;

[0084] Perform empirical mode decomposition on multiple signals to be decomposed respectively to obtain multiple groups of second intrinsic mode function components.

[0085] Based on the fitness value function, multiple groups of second intrinsic mode function components and fault traveling wave signals are analyzed and processed to give the local optimal fitness value of each particle.

[0086] Further, the feature extraction module includes a model building module, which includes:

[0087] Obtain the historical fault traveling wave signal of the distribution network and give the corresponding historical instantaneous frequency based on the optimization decomposition algorithm;

[0088] Normalize the historical instantaneous frequency to obtain normalized data and divide it into training set, validation set and test set;

[0089] Train the neural network model through the training set and give the trained neural network model;

[0090] The trained neural network model is verified by the validation set until the training end conditions are met, and the performance of the neural network model is tested using the test set to obtain the optimized fault detection model.

[0091] Furthermore, the model building module also includes:

[0092] S31, initializing the neural network model and loss function, giving initial model parameters and loss function coefficients; wherein the model parameters include a weight matrix and a bias vector;

[0093] S32, the neural network model performs forward propagation on the training data of the training set and calculates the value of the loss function;

[0094] S33, calculating the gradient of the loss function relative to the model parameters of the neural network model based on the back propagation algorithm, and using the gradient descent algorithm to update the model parameters and the loss function coefficients according to the gradient to minimize the value of the loss function;

[0095] S34, repeat iterating S32 and S33 until convergence, and give a trained neural network model.

[0096] Furthermore, the fault identification module includes:

[0097] The fault features of the distribution network are input into the pre-built classifier to give the probability of mapping the fault features to each fault category;

[0098] The fault category with the highest probability is taken as the fault category corresponding to the fault traveling wave signal.

[0099] The present invention provides a distribution network fault detection method and device based on traveling wave signals, which at least have the following beneficial effects:

[0100] (1) After extracting the features of the fault traveling wave signal to obtain the instantaneous frequency, the pre-built optimized fault detection model is used to extract the fault features of the instantaneous frequency, and finally complete the fault classification of the fault traveling wave signal. This can improve the accuracy of fault feature extraction and ultimately improve the accuracy of fault classification, that is, improve the accuracy of distribution network fault detection.

[0101] (2) By optimizing the existing empirical mode decomposition, we can adaptively optimize the characteristics of the signal itself to obtain more accurate numerical parameters of the empirical mode decomposition, thereby improving the accuracy of the empirical mode decomposition. We can adaptively obtain more repeatable results based on the characteristics of the signal with higher computational efficiency.

[0102] (3) By optimizing the neural network model, overfitting can be effectively prevented, and the accuracy and robustness of the model can be taken into account, thereby improving the generalization ability and making it more suitable for fault feature extraction in the current application scenario.

[0103] (4) By combining the optimized empirical mode decomposition with the optimized neural network model, it can adapt to the complex and changeable environment of the distribution network. By continuously learning and optimizing the model parameters, the recognition ability and diagnostic accuracy of different types of faults can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0104] Figure 1 A flow chart of a distribution network fault detection method based on traveling wave signals provided by the present invention;

[0105] Figure 2 A flow chart for obtaining instantaneous frequency provided by an embodiment of the present invention;

[0106] Figure 3 A flowchart of providing a first white noise provided by an embodiment of the present invention;

[0107] Figure 4 A flowchart of providing a local optimal fitness value provided by an embodiment of the present invention;

[0108] Figure 5 A flowchart of a pre-built optimized fault detection model provided by an embodiment of the present invention;

[0109] Figure 6 A flowchart of a trained neural network model provided in an embodiment of the present invention;

[0110] Figure 7 A schematic diagram of a distribution network fault detection device based on traveling wave signals provided by the present invention. DETAILED DESCRIPTION

[0111] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0112] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings, and "multiple" generally includes at least two.

[0113] It should also be noted that the term "includes", "comprising" or any other variation thereof is intended to cover non-exclusive inclusion, so that a commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprising a ..." do not exclude the existence of other identical elements in the commodity or device including the elements.

[0114] In order to suppress modal aliasing, Ensemble Empirical Mode Decomposition (EEMD) is used. EEMD is a noise-assisted data analysis method that can well restore the essence of the signal. In addition, when processing long sequences, traditional RNNs often have difficulty capturing long-distance information dependencies due to the problem of gradient vanishing or gradient exploding. In order to solve the problem of RNN processing, the LSTM neural network algorithm is used. The LSTM neural network algorithm is a special neural network structure specially designed to process sequence data with long-term dependencies. LSTM effectively solves this problem by introducing a gating mechanism. However, when the LSTM neural network is directly applied to distribution network fault diagnosis, it is difficult to directly capture the physical characteristics of the fault signal.

[0115] Based on this, Figure 1 As shown, the present invention provides a distribution network fault detection method based on traveling wave signals, which specifically includes the following steps:

[0116] Obtain the fault traveling wave signal of the distribution network and determine the fault location of the distribution network;

[0117] Based on the optimization decomposition algorithm, the fault traveling wave signal is frequency decomposed to obtain the instantaneous frequency;

[0118] Based on the pre-built optimized fault detection model, the instantaneous frequency is feature extracted to give the fault characteristics of the distribution network;

[0119] According to the fault characteristics of the distribution network, the fault category corresponding to the fault traveling wave signal is given.

[0120] After determining the fault location and fault type of the distribution network, the present invention can complete the fault detection of the distribution network. Wherein, determining the fault location of the distribution network includes:

[0121] Determine the corresponding power cable according to the acquired fault traveling wave signal;

[0122] According to the determined power cable, the time difference of the fault traveling wave signal transmitted to the two ends of the corresponding power cable is given;

[0123] Based on the time difference of the fault traveling wave signal being transmitted to the two ends of the corresponding power cable and combined with the length of the corresponding power cable, the fault position on the power cable is given, satisfying the following relationship:

[0124]

[0125] Where t1 and t2 are the time taken for the signal to reach the two measuring ends of the line, v is the line mode wave velocity, l is the total length of the line, and x is the calculated fault distance.

[0126] like Figure 2 As shown, this embodiment performs frequency decomposition on the fault traveling wave signal based on the optimization decomposition algorithm to obtain the instantaneous frequency, including:

[0127] A noise coefficient is given based on the fault traveling wave signal and a plurality of first white noises are generated accordingly;

[0128] Combine multiple first white noises with the fault traveling wave signal to satisfy the following relationship:

[0129]

[0130] Where: x m (t) is the fault traveling wave signal after combining with the first white noise, x(t) is the fault traveling wave signal, n m (t) is the first white noise, A noise is the noise factor, δ (t) is Gaussian white noise;

[0131] Perform empirical mode decomposition on each combined fault traveling wave signal to obtain multiple first intrinsic mode function components that satisfy the following relationship:

[0132]

[0133] In the formula, r n (t) is the residual component, c i (t) is the first intrinsic modal function component, and n is the number of the first intrinsic modal function components;

[0134] Performing mean processing on all first intrinsic mode function components, and performing Hilbert transform on the mean processing result to obtain the instantaneous frequency;

[0135] Among them, all first intrinsic mode function components are averaged to satisfy the following relationship:

[0136]

[0137] In the formula, is the mean result of all first intrinsic mode function components, M is the number of decompositions of empirical mode decomposition, c ij (t) is the sum of all first intrinsic mode function components corresponding to the j-th decomposition.

[0138] like Figure 3 As shown, this embodiment provides a plurality of first white noises with different noise coefficients based on the fault traveling wave signal, which may include:

[0139] Based on the standard deviation of the fault traveling wave signal, the value range of the noise coefficient is determined;

[0140] Taking the noise coefficient as a particle and combining the value range of the noise coefficient, the particles are initialized to give a particle swarm composed of multiple random particles;

[0141] Calculate the fitness value of the particle swarm, give the local optimal fitness value of each particle, and take the minimum value of all local optimal fitness values ​​as the global optimal fitness value;

[0142] Based on the evolution function of the time series particle swarm optimization algorithm, the position of the particle swarm is updated using extreme value perturbations;

[0143] Repeat the iterative fitness value calculation and position update steps until convergence, and give the final global optimal fitness value; where the position of the particle group represents the position of each particle, and the position of each particle is the value of the noise coefficient; the convergence condition is to reach the maximum number of iterations;

[0144] Based on the final global optimal fitness value, the corresponding noise coefficient value is determined, and different Gaussian white noises are combined to obtain a plurality of first white noises with noise coefficients.

[0145] Among them, the standard deviation of the fault traveling wave signal, σ0, satisfies the following relationship:

[0146]

[0147] Where p is the mean value of the fault traveling wave signal, N2 is the number of time windows, and x i is the fault traveling wave signal data in the i-th time window.

[0148] After obtaining the standard deviation of the fault traveling wave signal, the noise coefficient range is [0.1σ0, 0.5σ0].

[0149] like Figure 4 As shown, the fitness value of the particle swarm is calculated, and the local optimal fitness value of each particle is given, including:

[0150] The noise coefficient value of each particle in the particle group is combined with the same Gaussian white noise to obtain multiple second white noises;

[0151] Combining the plurality of second white noises with the fault traveling wave signal to obtain a plurality of signals to be decomposed;

[0152] Perform empirical mode decomposition on multiple signals to be decomposed respectively to obtain multiple groups of second intrinsic mode function components.

[0153] Based on the fitness value function, multiple groups of second intrinsic mode function components and fault traveling wave signals are analyzed and processed to give the local optimal fitness value of each particle.

[0154] The fitness value function satisfies the following relationship:

[0155]

[0156] Where RRMSE is the fitness value, N1 is the total number of the second intrinsic mode function components, x0(k) is the fault traveling wave signal, c max (k) is the second intrinsic modal function component with the highest correlation coefficient with the fault traveling wave signal among the second intrinsic modal function components of the kth group obtained by empirical mode decomposition;

[0157] In this embodiment, when updating the position of the particle swarm, the evolution function of the time series particle swarm optimization algorithm used needs to be initialized. Specifically, according to the characteristics of the historical fault traveling wave signal, the number of individuals m in the population, the dimension D of the target search space, the total number of iterations N, the momentum inertia coefficient ω, the learning factors c1, c2, and the initial stagnation step thresholds T0, T2 that the individual and global extreme values ​​need to be disturbed are set. g The particle swarm position is updated and iterated based on the initialization parameters of the evolution function, where the evolution function satisfies the following relationship:

[0158]

[0159] In the formula, is the position of the particle, indicating the position of the i-th particle in the d-th dimension in the t+1-th iteration, i=1,2,…,m, d=1,2,…,D, m is the number of particles in the particle swarm, and D is the set search dimension; ω is the momentum inertia coefficient; c1 and c2 are non-negative learning factors; r1, r2, r3, and r4 are all random numbers that obey the U(0,1) distribution; p is the extreme point of the particle individual and the population, t0 and t g are the evolutionary stagnation steps of individual particle extrema and global extrema respectively; T0 and T g are the stagnation step thresholds that need to be disturbed for individual particle extrema and global extrema respectively.

[0160] like Figure 5 As shown, the pre-construction of the optimized fault detection model may include:

[0161] Obtain the historical fault traveling wave signal of the distribution network and give the corresponding historical instantaneous frequency based on the optimization decomposition algorithm;

[0162] Normalize the historical instantaneous frequency to obtain normalized data and divide it into training set, validation set and test set;

[0163] Train the neural network model (LSTM model) through the training set and give the trained neural network model;

[0164] The trained neural network model is verified by the validation set until the training end condition is met, and the performance of the neural network model is tested with the test set to obtain the optimized fault detection model; specifically, the trained neural network model is verified by the validation set to determine whether the training end condition is met. If so, the training is stopped, and the performance of the neural network model is tested with the test set to obtain the optimized fault detection model; otherwise, the internal parameters of the neural network model are adjusted and retrained. The training end condition is that the network generalization ability of the neural network model reaches the maximum and the network performance meets the requirements.

[0165] The historical instantaneous frequency is normalized, including: vectorizing the historical instantaneous frequency into a fault feature vector H=[H1,H2,...,H n ], and then the fault feature vector is linearly normalized so that its amplitude is within [0,1] and satisfies the following relationship:

[0166]

[0167] In the formula, H s is the normalized data, H is the fault feature vector, H min is the minimum fault feature vector data, H max is the largest fault feature vector data.

[0168] At the same time, when this embodiment extracts features of the instantaneous frequency based on the pre-built optimized fault detection model, the instantaneous frequency is first processed using the same normalization method as mentioned above, and then the normalized instantaneous frequency is input into the pre-built optimized fault detection model for detection to give the fault characteristics of the distribution network.

[0169] like Figure 6 As shown, the neural network model is trained through the training set, and the trained neural network model is given, including:

[0170] S31, initializing the neural network model and loss function, giving initial model parameters and loss function coefficients; wherein the model parameters include a weight matrix and a bias vector;

[0171] S32, the neural network model performs forward propagation on the training data of the training set and calculates the value of the loss function;

[0172] S33, calculating the gradient of the loss function relative to the model parameters of the neural network model based on the back propagation algorithm, and using the gradient descent algorithm to update the model parameters and the loss function coefficients according to the gradient to minimize the value of the loss function;

[0173] S34, repeat iterating S32 and S33 until convergence, and give a trained neural network model.

[0174] Among them, the fault detection model is optimized to satisfy the following relationship:

[0175]

[0176] In the formula, H s,t is the input normalized data of the optimized fault detection model at time t, is the current input cell state, W f is the weight matrix of the forget gate, b f is the bias of the forget gate, W i is the weight matrix of the input gate, b i is the bias of the input gate, W c is the weight matrix of the cell state, b c is the bias of the cell state, W o is the weight matrix of the output gate, b o is the bias of the output gate, tanh is the activation function, h t-1 is the fault characteristic output at time t-1, ft is the proportional coefficient of the forgotten information controlled in the forget gate, i t is the proportional coefficient of the input gate that controls the input information, c t is the status information after the cell status is updated, o t is the proportional coefficient of the output gate controlling the output information, σ is the Sigmoid activation function, C t is the cell state at time t, h t is the fault characteristic of the output.

[0177] The optimization algorithm is used to optimize the neural network model from two aspects: weight optimization and structure adjustment, including adjusting the loss function of the neural network model to meet the following relationship:

[0178]

[0179] Where E is the loss function, α and β are the loss function coefficients, and n is the total number of network weights of the neural network model. l is the average network weight of the neural network model, w i is the network weight of the ith neural network model, W k represents the network weight vector of the kth iteration, J is the Jacobian matrix, μ is the step size control parameter, which is 0.001, and γ is the number of effective parameters of the neural network model, γ=N-2β'tr(H) -1 , β' is the β value of the previous iteration, N is the total number of model parameters of the neural network model; H is the Hessian matrix of the loss function, tr(H) represents the trace of the matrix H, W k Because of w i The matrix composed of ε(W k ) represents the value based on the current weight W k The error vector.

[0180] After pre-building the optimized fault detection model, this embodiment inputs the instantaneous frequency into the optimized fault detection model, performs corresponding processing and analysis on the instantaneous frequency, and finally outputs the fault characteristics of the distribution network.

[0181] According to the fault characteristics of the distribution network, the fault categories corresponding to the fault traveling wave signals are given, including:

[0182] The fault characteristics of the distribution network are input into the classifier, and the probability of mapping the fault characteristics to each fault category is given;

[0183] The fault category with the highest probability is taken as the fault category corresponding to the fault traveling wave signal.

[0184] The fault feature h extracted by the neural network model t , a softmax classifier is used to map each fault feature ht into the probability distribution y=[y1,y2,y3,...,y n ], select the category corresponding to the maximum value of the calculated probability as the classification result, so as to realize the classification of the fault cause

[0185]

[0186] Where W is the weight of the output layer, b is the bias, and n is the total number of fault cause categories.

[0187] like Figure 7 As shown, the present invention also provides a distribution network fault detection device based on traveling wave signals, which adopts the above-mentioned distribution network fault detection method based on traveling wave signals, and the distribution network fault detection device includes:

[0188] A signal acquisition and positioning module is used to acquire the fault traveling wave signal of the distribution network and determine the fault location of the distribution network;

[0189] A frequency decomposition module, which is used to perform frequency decomposition on the fault traveling wave signal based on an optimized decomposition algorithm to obtain an instantaneous frequency;

[0190] A feature extraction module is used to extract features of the instantaneous frequency based on a pre-built optimized fault detection model and provide fault features of the distribution network;

[0191] The fault identification module is used to give the fault category corresponding to the fault traveling wave signal according to the fault characteristics of the distribution network.

[0192] Furthermore, the signal acquisition and positioning module includes:

[0193] Determine the corresponding power cable according to the acquired fault traveling wave signal;

[0194] According to the determined power cable, the time difference of the fault traveling wave signal transmitted to the two ends of the corresponding power cable is given;

[0195] Based on the time difference of the fault traveling wave signal being transmitted to the two ends of the corresponding power cable and combined with the length of the corresponding power cable, the fault location on the power cable is given.

[0196] Furthermore, the frequency decomposition module includes:

[0197] A noise coefficient is given based on the fault traveling wave signal and a plurality of first white noises are generated accordingly;

[0198] combining the plurality of first white noises with the fault traveling wave signal respectively;

[0199] Performing empirical mode decomposition on each combined fault traveling wave signal to obtain a plurality of first intrinsic mode function components;

[0200] All first intrinsic mode function components are averaged, and the averaged results are Hilbert transformed to obtain instantaneous frequencies.

[0201] Furthermore, the frequency decomposition module includes:

[0202] Based on the standard deviation of the fault traveling wave signal, the value range of the noise coefficient is determined;

[0203] Taking the noise coefficient as a particle and combining the value range of the noise coefficient, the particles are initialized to give a particle swarm composed of multiple random particles;

[0204] Calculate the fitness value of the particle swarm, give the local optimal fitness value of each particle, and take the minimum value of all local optimal fitness values ​​as the global optimal fitness value;

[0205] Update the position of the particle swarm based on the evolution function and using extreme value perturbations;

[0206] Repeat the iterative fitness value calculation and position update steps until convergence, giving the final global optimal fitness value;

[0207] Based on the final global optimal fitness value, the corresponding noise coefficient value is determined, and different Gaussian white noises are combined to obtain a plurality of first white noises with noise coefficients.

[0208] Furthermore, the frequency decomposition module also includes:

[0209] The noise coefficient value of each particle in the particle group is combined with the same Gaussian white noise to obtain multiple second white noises;

[0210] Combining the plurality of second white noises with the fault traveling wave signal to obtain a plurality of signals to be decomposed;

[0211] Perform empirical mode decomposition on multiple signals to be decomposed respectively to obtain multiple groups of second intrinsic mode function components.

[0212] Based on the fitness value function, multiple groups of second intrinsic mode function components and fault traveling wave signals are analyzed and processed to give the local optimal fitness value of each particle.

[0213] Further, the feature extraction module includes a model building module, which includes:

[0214] Obtain the historical fault traveling wave signal of the distribution network and give the corresponding historical instantaneous frequency;

[0215] Normalize the historical instantaneous frequency to obtain normalized data and divide it into training set, validation set and test set;

[0216] Train the neural network model through the training set and give the trained neural network model;

[0217] The trained neural network model is verified by the validation set until the training end conditions are met, and the performance of the neural network model is tested using the test set to obtain the optimized fault detection model.

[0218] Furthermore, the model building module also includes:

[0219] S31, initializing the neural network model and loss function, giving initial model parameters and loss function coefficients; wherein the model parameters include a weight matrix and a bias vector;

[0220] S32, the neural network model performs forward propagation on the training data of the training set and calculates the value of the loss function;

[0221] S33, calculating the gradient of the loss function relative to the model parameters of the neural network model based on the back propagation algorithm, and using the gradient descent algorithm to update the model parameters and the loss function coefficients according to the gradient to minimize the value of the loss function;

[0222] S34, repeat iterating S32 and S33 until convergence, and give a trained neural network model.

[0223] Furthermore, the fault identification module includes:

[0224] The fault features of the distribution network are input into the pre-built classifier to give the probability of mapping the fault features to each fault category;

[0225] The fault category with the highest probability is taken as the fault category corresponding to the fault traveling wave signal.

[0226] Although preferred embodiments of the present invention have been described, additional changes and modifications may be made to these embodiments by those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A distribution network fault detection method based on traveling wave signals, characterized in that: The specific steps include: Obtain the fault traveling wave signal of the distribution network and determine the fault location of the distribution network; Performing frequency decomposition on the fault traveling wave signal based on an optimized decomposition algorithm to obtain an instantaneous frequency, including: providing a noise coefficient based on the fault traveling wave signal and correspondingly generating a plurality of first white noises; combining the plurality of first white noises with the fault traveling wave signal respectively; performing empirical mode decomposition on each combined fault traveling wave signal to obtain a plurality of first intrinsic mode function components; performing mean processing on all first intrinsic mode function components, and performing Hilbert transform on the mean processing result to obtain an instantaneous frequency; Based on the pre-built optimized fault detection model, the instantaneous frequency is feature extracted to give the fault characteristics of the distribution network; According to the fault characteristics of the distribution network, the fault category corresponding to the fault traveling wave signal is given; Among them, the pre-construction of the optimized fault detection model includes: The historical fault traveling wave signals of the distribution network are obtained, and the corresponding historical instantaneous frequencies are given based on the optimization decomposition algorithm; the historical instantaneous frequencies are normalized to obtain normalized data and divided into training set, verification set and test set; the neural network model is trained with the training set to give the trained neural network model; the trained neural network model is verified with the verification set until the training end conditions are met, and the performance of the neural network model is verified with the test set to obtain the optimized fault detection model.

2. The method for detecting a distribution network fault according to claim 1, wherein: Determine the fault location in the distribution network, including: Determine the corresponding power cable according to the acquired fault traveling wave signal; According to the determined power cable, the time difference of the fault traveling wave signal transmitted to the two ends of the corresponding power cable is given; Based on the time difference of the fault traveling wave signal being transmitted to the two ends of the corresponding power cable and combined with the length of the corresponding power cable, the fault location on the power cable is given.

3. The method for detecting a distribution network fault according to claim 1, wherein: A noise coefficient is given based on the fault traveling wave signal and a plurality of first white noises are generated accordingly, including: Based on the standard deviation of the fault traveling wave signal, the value range of the noise coefficient is determined; Taking the noise coefficient as a particle and combining the value range of the noise coefficient, the particles are initialized to give a particle swarm composed of multiple random particles; Calculate the fitness value of the particle swarm, give the local optimal fitness value of each particle, and take the minimum value of all local optimal fitness values ​​as the global optimal fitness value; Update the position of the particle swarm based on the evolution function and using extreme value perturbations; Repeat the iterative fitness value calculation and position update steps until convergence, giving the final global optimal fitness value; Based on the final global optimal fitness value, the corresponding noise coefficient value is determined, and different Gaussian white noises are combined to obtain a plurality of first white noises with noise coefficients.

4. The method for detecting a distribution network fault according to claim 3, wherein: Calculate the fitness value of the particle swarm and give the local optimal fitness value of each particle, including: The noise coefficient value of each particle in the particle group is combined with the same Gaussian white noise to obtain multiple second white noises; Combining the plurality of second white noises with the fault traveling wave signal to obtain a plurality of signals to be decomposed; Perform empirical mode decomposition on multiple signals to be decomposed respectively to obtain multiple groups of second intrinsic mode function components. Based on the fitness value function, multiple groups of second intrinsic mode function components and fault traveling wave signals are analyzed and processed to give the local optimal fitness value of each particle.

5. The method for detecting a distribution network fault according to claim 1, wherein: The neural network model is trained through the training set, and the trained neural network model is given, including: S31, initializing the neural network model and loss function, giving initial model parameters and loss function coefficients; wherein the model parameters include a weight matrix and a bias vector; S32, the neural network model performs forward propagation on the training data of the training set and calculates the value of the loss function; S33, calculating the gradient of the loss function relative to the model parameters of the neural network model based on the back propagation algorithm, and using the gradient descent algorithm to update the model parameters and the loss function coefficients according to the gradient to minimize the value of the loss function; S34, repeat iterating S32 and S33 until convergence, and give a trained neural network model.

6. The method for detecting a distribution network fault according to claim 5, characterized in that: The loss function satisfies the following relationship: ； Where E is the loss function, α and β are the loss function coefficients, and n is the total number of network weights of the neural network model. l is the average network weight of the neural network model, w i is the network weight of the ith neural network model, W k represents the network weight vector of the kth iteration, J is the Jacobian matrix, ε(W k ) is based on W k The error vector, μ is the step size control parameter, γ is the number of effective parameters of the neural network model, γ=N-2β'tr(H) -1 , β' is the β value of the previous iteration, N is the total number of model parameters of the neural network model; H is the Hessian matrix of the loss function, and tr(H) represents the trace of the matrix H.

7. The distribution network fault detection method according to claim 5 or 6, characterized in that: According to the fault characteristics of the distribution network, the fault categories corresponding to the fault traveling wave signals are given, including: The fault features of the distribution network are input into the pre-built classifier to give the probability of mapping the fault features to each fault category; The fault category with the highest probability is taken as the fault category corresponding to the fault traveling wave signal.

8. A distribution network fault detection device based on traveling wave signals, characterized in that: Using the distribution network fault detection method based on traveling wave signals according to any one of claims 1 to 7, the distribution network fault detection device comprises: A signal acquisition and positioning module is used to acquire the fault traveling wave signal of the distribution network and determine the fault location of the distribution network; A frequency decomposition module, which is used to perform frequency decomposition on the fault traveling wave signal based on an optimized decomposition algorithm to obtain an instantaneous frequency; A feature extraction module is used to extract features of the instantaneous frequency based on a pre-built optimized fault detection model and provide fault features of the distribution network; The fault identification module is used to give the fault category corresponding to the fault traveling wave signal according to the fault characteristics of the distribution network.

Images