CN118244057A - Power system fault detection method, device, equipment and storage medium - Google Patents

Abstract

The invention discloses a power system fault detection method, a device, equipment and a storage medium, belonging to the technical field of power grid fault detection, wherein the method comprises the steps of collecting a zero sequence current mathematical expectation T ₁, a zero sequence current variance T ₂, a frequency variance T ₃ of sampling points of each phase, a correlation coefficient T ₄ between each two phases of current, and a characteristic quantity T ₅,T₆,T₇ extracted by PCA dimension reduction processing of a wavelet coefficient obtained after each phase of current scale decomposition of a power transmission line under a fault state; and taking the seven characteristic value sequences of T ₁-T₇ and the corresponding power transmission line fault states as training data sets, training the preset parameters of the radial basis neural network by using an Adam algorithm based on the training data sets, and obtaining a trained power system fault detection model for detecting power system faults. The invention can realize the effect of efficiently and accurately identifying the type of the short-circuit fault, and solves the problem that the current technical scheme is difficult to efficiently and accurately detect the fault of the power system.

Description

Power system fault detection method, device, equipment and storage medium

Technical Field

The present invention belongs to the technical field of power grid fault detection, and in particular relates to a power system fault detection method, device, equipment and storage medium.

Background technique

Transmission lines are one of the foundations for providing stable and reliable power supply. However, due to various reasons, short-circuit faults may occur in transmission lines during operation, which will not only affect the quality of power supply, but also cause damage to cables and even cause serious consequences such as fire. Therefore, short-circuit fault detection and diagnosis in transmission lines is particularly important.

Common transmission line fault detection methods mainly rely on manual inspections and the use of protection devices for fault detection. Manual inspections require a lot of manpower and material resources, and the detection efficiency is low. Fault detection of protection devices is usually carried out through current differential protection, tripping protection and other technologies, but these methods have low detection accuracy for certain fault types, and the diagnosis of transmission line faults is unclear. It is difficult to determine the fault location and cause in the first time when a fault occurs, which is not conducive to the inspection and maintenance work of power workers.

Summary of the invention

The purpose of the present invention is to provide a power system fault detection method, device, equipment and storage medium, which can achieve the effect of accurately identifying the type of short-circuit fault, and solve the problem that the current technical solution is difficult to detect power system faults efficiently and accurately.

To achieve the above purpose, the technical solution adopted by the present invention is as follows:

A method for detecting faults in an electric power system, the method comprising:

Collect the mathematical expectation T ₁ of zero-sequence current of the transmission line under fault condition, the variance T ₂ of zero-sequence current, the frequency variance T ₃ of each phase sampling point, the correlation coefficient T ₄ between each two-phase current, and the wavelet coefficients obtained after the scale decomposition of each phase current, and extract the feature quantities T ₅ , T ₆ , T ₇ after PCA dimensionality reduction processing;

The seven eigenvalue sequences T ₁ -T ₇ and the corresponding transmission line fault states are used as training data sets. The parameters of the preset radial basis function neural network are trained using the Adam algorithm based on the training data sets to obtain a trained power system fault detection model for detecting power system faults.

The mathematical expectation T ₁ of the zero-sequence current of the transmission line under fault condition, the variance T ₂ of the zero-sequence current, the frequency variance T ₃ of each phase sampling point, the correlation coefficient T ₄ between each two-phase current, and the wavelet coefficients obtained after the scale decomposition of each phase current are processed by PCA dimensionality reduction, and the extracted feature quantities T ₅ , T ₆ , T ₇ are specifically:

Extract the three-phase current after the fault occurs (i.e., phase A current, phase B current, phase C current respectively), and calculate the zero-sequence current component/> , zero-sequence current mathematical expectation/> (E() represents mathematical expectation), zero-sequence current variance/> ,/> (The rest of the calculations are similar, for example ), the frequency variance of each phase sampling point/> ,/> 、/> 、/> are the frequencies of the sampling points of phase A, phase B, and phase C respectively. The correlation coefficient between the currents of each two phases is T ₄ , which is the Peelman coefficient between each two phases. （/> That is, the Pielman coefficient between phase A and phase B, and the rest are similar). The characteristic quantities T ₅ , T ₆ , and T ₇ are the characteristic vectors extracted after the discrete wavelet transform of each phase current signal decomposes the original signal into low-frequency wavelet coefficients and high-frequency wavelet coefficients of different scales and performs PCA dimensionality reduction processing. Among them, /> ,/> , ,/> is the average value of the fault current of phase A,/> is the average value of the fault current of phase B,/> is the average value of the fault current of phase C.

The discrete wavelet transform comprises:

Using a binary sampling network, take the scale factor , displacement/> , j represents the number of decomposition levels, Z is an integer, /> represents a positive integer, k is the parameter of the displacement factor (responsible for the displacement of the wavelet function in the time domain);

The corresponding discrete wavelet function is:

;

Discrete wavelet coefficients:

;

In the formula, t is time, represents the function of the A phase fault current changing with time, and * represents taking the conjugate; the Mallat algorithm is used to perform multi-scale analysis on the signal, and the signal to be analyzed is decomposed into various frequency scales, which is equivalent to performing multiple low-pass and high-pass filtering on the signal to obtain low-frequency wavelet coefficients and high-frequency wavelet coefficients at different layers;

The original signal is decomposed into the first layer of wavelet using wavelet basis functions to obtain a set of high-frequency wavelet coefficients and a set of low-frequency wavelet coefficients; the low-frequency wavelet coefficients are decomposed into the next layer of wavelet, and the process is repeated until the desired decomposition level is reached or the decomposition termination condition is met.

The PCA dimensionality reduction process includes:

Assume that the matrix X _n×m is composed of [CA ₄ , CD ₁ , CD ₂ , CD ₃ , CD ₄ ], where CA ₄ is the low-frequency wavelet coefficient of the fourth decomposition, CD ₁ is the high-frequency wavelet coefficient of the first decomposition, CD ₂ is the high-frequency wavelet coefficient of the second decomposition, CD ₃ is the high-frequency wavelet coefficient of the third decomposition, and CD ₄ is the high-frequency wavelet coefficient of the fourth decomposition. n is the number of samples of the current signal, and m is the number of variables, m=5;

For each column in Xn _×5 Elements/> ,/> , through the mean centering formula/> , and obtain the decentralized matrix Y _n×m ;

Calculate the covariance matrix between the feature quantities, the formula is , where Y is the feature quantity and Y ^T is the transposed matrix of Y;

Perform eigenvalue decomposition on the covariance matrix C and calculate the eigenroot and the eigenvector/> ;

The characteristic equation for Y is , sort the desired eigenvalues:/> , the corresponding eigenvector is/> ;

The cumulative contribution rate is greater than or equal to the preset threshold yz, and it is determined that the accuracy of the principal component model meets the condition, that is,

;

Take the eigenvectors corresponding to the largest K eigenvalues ;

The characteristic matrix X _AK of the wavelet coefficients of phase A current after dimension reduction is composed of K eigenvectors. , repeat the calculation and get/> ,/> .

The parameters of the preset radial basis neural network are trained using the Adam algorithm based on the training data set, including:

Calculate the gradients of the weights and bias parameters of a preset radial basis neural network , where N is the size of the training data set; _Li (P) is the loss function of the neural network; P represents the weight and bias parameters of the neural network;

The Adam algorithm calculates the parameter gradient The first moment estimate of and second moment estimates/> , and constantly correct the first-order moment estimate/> and second moment estimates/> The deviation is used to obtain the correction amount of the parameter, and the parameters of the preset radial basis neural network are determined based on the correction amount of the parameter. The specific calculation formula is as follows:

;

;

in, Represents the decay rate of the moment estimate, usually/> ,/> .

The method further comprises:

Compute the corrected first moment estimate bias and the second moment estimate bias/> , N is the current iteration number, /> , ;

based on and/> , determine the updated values of the neural network weights and bias parameters as/> , where/> Indicates the step length; /> is a small constant.

Said Take 0.001, the /> Take 10 ^-8 .

The present invention also provides a power system fault detection device, the device comprising:

The sampling module is used to collect the mathematical expectation T ₁ of the zero-sequence current of the transmission line under fault conditions, the variance T ₂ of the zero-sequence current, the frequency variance T ₃ of each phase sampling point, the correlation coefficient T ₄ between each two-phase current, and the wavelet coefficients obtained after the scale decomposition of each phase current, and the feature quantities T ₅ , T ₆ , T ₇ extracted after the PCA dimensionality reduction processing;

The processing module is used to use the seven eigenvalue sequences T ₁ -T ₇ and the corresponding transmission line fault states as training data sets, and train the parameters of the preset radial basis neural network using the Adam algorithm based on the training data sets to obtain a trained power system fault detection model for detecting power system faults.

The present invention also provides an electronic device, the electronic device comprising:

at least one processor;

and a memory communicatively coupled to the at least one processor;

The memory stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor to implement the method of the first aspect of the present invention.

The present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the method according to the first aspect of the present invention is implemented.

The beneficial effects of the present invention are as follows:

By collecting the current signal of each phase of the transmission line, the fault characteristic quantity _T1-7 , the mathematical expectation of zero-sequence current _T1 , the zero-sequence current variance _T2 , the frequency variance _T3 of each phase sampling point, the correlation coefficient _T4 between each two-phase current, the wavelet coefficients obtained after the scale decomposition of each phase current are subjected to principal component analysis (PCA) dimensionality reduction processing, and the characteristic quantities _T5 , _T6 , _T7 are extracted; subsequently, the above seven eigenvalue sequences are used as training data sets and the radial basis function neural network (RBFNN) input layer is used to train the parameters of the RBFNN based on the training data set using the adaptive moment estimation (Adam) algorithm to achieve the effect of accurately identifying the type of short-circuit fault, solving the problem that the current technical solution is difficult to efficiently and accurately detect faults in the power system.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic diagram showing a flow chart of a method for detecting a fault in a power system according to an embodiment of the present invention;

FIG2 shows a schematic diagram of a short circuit fault provided by an embodiment of the present invention;

FIG3 shows a wavelet decomposition tree diagram provided by an embodiment of the present invention;

FIG4 shows a block diagram of a power system fault detection device provided by an embodiment of the present invention;

FIG5 shows a block diagram of an exemplary electronic device capable of implementing embodiments of the present invention.

Detailed ways

The embodiments of the present invention are further described below in conjunction with the accompanying drawings.

As shown in FIG1 , the power system fault detection method includes:

S101, collecting the mathematical expectation T ₁ of the zero-sequence current of the transmission line in a fault state, the variance T ₂ of the zero-sequence current, the frequency variance T ₃ of each phase sampling point, the correlation coefficient T ₄ between each two-phase current, and the wavelet coefficients obtained after the scale decomposition of each phase current, and extracting the feature quantities T ₅ , T ₆ , T ₇ after PCA dimensionality reduction processing;

S102, taking seven characteristic value sequences T ₁ -T ₇ and corresponding transmission line fault states as training data sets, and training parameters of a preset radial basis neural network using an Adam algorithm based on the training data sets to obtain a trained power system fault detection model for detecting power system faults.

The power system fault detection method provided by the present invention collects the current signal of each phase of the transmission line to calculate the fault characteristic quantity _T1-7 , the zero-sequence current mathematical expectation _T1 ; the zero-sequence current variance _T2 ; the frequency variance _T3 of each phase sampling point; the correlation coefficient _T4 between each two-phase current; the wavelet coefficient obtained after the scale decomposition of each phase current is subjected to PCA dimensionality reduction processing, and the characteristic quantities _T5 , _T6 , _T7 are extracted; subsequently, the above seven characteristic value sequences are used as training data sets and radial basis neural network input layer, and the parameters of RBFNN are trained by using the Adam algorithm based on the training data sets to achieve the effect of accurately identifying the short-circuit fault type, so as to solve the problem that the current technical solution is difficult to efficiently and accurately detect the faults of the power system.

It should be noted that after a short circuit occurs in a transmission line, the measured values of current, voltage, frequency, etc. will change dramatically, which is a non-stationary random signal. The short circuit fault of the transmission line, that is, the fault state corresponding to the above-mentioned acquisition transmission line mainly includes single-phase short circuit to ground, two-phase short circuit, two-phase short circuit to ground and three-phase short circuit fault. The schematic diagram of the above four short circuit faults is shown in Figure 2.

The mathematical expectation T ₁ of the zero-sequence current of the transmission line under fault condition, the variance T ₂ of the zero-sequence current, the frequency variance T ₃ of each phase sampling point, the correlation coefficient T ₄ between each two-phase current, and the wavelet coefficients obtained after the scale decomposition of each phase current are processed by PCA dimensionality reduction, and the extracted feature quantities T ₅ , T ₆ , T ₇ are specifically as follows:

Extract the three-phase current after the fault occurs and calculate the zero-sequence current component , zero-sequence current mathematical expectation/> , zero sequence current variance/> ,/> , frequency variance of each phase sampling point/> ,/> 、/> 、/> are the frequencies of the sampling points of phase A, phase B, and phase C respectively, and the correlation coefficient T ₄ between each two-phase current is the Peelman coefficient between each two phases, that is, /> , the characteristic quantities T ₅ , T ₆ , and T ₇ are the characteristic vectors extracted after the discrete wavelet transform of each phase current signal decomposes the original signal into low-frequency wavelet coefficients and high-frequency wavelet coefficients of different scales and performs PCA dimensionality reduction processing, where, /> ,/> , ,/> is the average value of the fault current of phase A,/> is the average value of the fault current of phase B,/> is the average value of the fault current of phase C.

In the process of realizing the discrete wavelet transform in the present invention, the present invention discretizes the continuous wavelet. The discrete wavelet transform greatly reduces the computational complexity by discretizing the scale and displacement factors, making the wavelet transform more feasible in practical applications such as signal processing and data compression. The core of the discrete wavelet transform is to use a finite wavelet basis function to approximate the continuous wavelet transform, thereby achieving high computational efficiency while ensuring a certain degree of accuracy. In practice, a binary sampling network is often used to take the scale factor , displacement/> , j represents the number of decomposition levels, Z is an integer, /> represents a positive integer, k is the parameter of the displacement factor;

The corresponding discrete wavelet function is:

;

Discrete wavelet coefficients:

;

The present invention adopts Mallat algorithm to perform multi-scale analysis on the signal and adopts Daubechies4 (db4) as wavelet basis function. DaubechiesN wavelet has orthogonal properties, which can reduce the error and information loss of wavelet transform. At the same time, it has the characteristics of fast calculation and can efficiently decompose the signal through fast multi-scale transformation. The most important thing is that DaubechiesN wavelet has good stability to the noise and interference of the signal, and can effectively suppress the noise and extract the main features of the signal.

The Mallat algorithm is used to perform multi-scale analysis on the signal, and the signal to be analyzed is decomposed into various frequency scales, which is equivalent to performing multiple low-pass and high-pass filtering on the signal to obtain low-frequency wavelet coefficients and high-frequency wavelet coefficients at different levels.

The original signal is decomposed into the first layer of wavelet using wavelet basis functions to obtain a set of high-frequency wavelet coefficients and a set of low-frequency wavelet coefficients; the low-frequency wavelet coefficients are decomposed into the next layer of wavelet, and the process is repeated until the desired decomposition level is reached or the decomposition termination condition is met.

In a specific example, the wavelet decomposition tree f(t) is shown in FIG3 . The present invention uses four-layer wavelet decomposition to obtain the signal characteristics. In FIG3 , CA _1A —CA _4A are low-frequency wavelet coefficients of phase A; CD _1A —CD _4A are high-frequency wavelet coefficients of phase A.

PCA dimensionality reduction processing, including:

Assume that the matrix X _n×m is composed of [CA ₄ , CD ₁ , CD ₂ , CD ₃ , CD ₄ ], where CA ₄ is the low-frequency wavelet coefficient of the fourth decomposition, CD ₁ is the high-frequency wavelet coefficient of the first decomposition, CD ₂ is the high-frequency wavelet coefficient of the second decomposition, CD ₃ is the high-frequency wavelet coefficient of the third decomposition, and CD ₄ is the high-frequency wavelet coefficient of the fourth decomposition. n is the number of samples of the current signal, and m is the number of variables, m=5;

For each column in Xn _×5 Elements/> ,/> , through the mean centering formula/> , and obtain the decentralized matrix Y _n×m ;

Calculate the covariance matrix between the feature quantities, the formula is , where Y is the feature quantity and Y ^T is the transposed matrix of Y;

Perform eigenvalue decomposition on the covariance matrix C and calculate the eigenroot and the eigenvector/> ;

The characteristic equation for Y is , sort the desired eigenvalues:/> , the corresponding eigenvector is/> ;

The cumulative contribution rate is greater than or equal to the preset threshold yz, and it is determined that the accuracy of the principal component model meets the condition, that is,

;

Take the eigenvectors corresponding to the largest K eigenvalues ;

The characteristic matrix X _AK of the wavelet coefficients of phase A current after dimension reduction is composed of K eigenvectors. , repeat the calculation and get/> ,/> .

Due to the complexity of the factors that cause transmission line failures, using PCA to reduce the dimension of the wavelet coefficient data matrix can more accurately find and extract the most important components in the data, that is, the most representative fault features, which can be better used as the input layer of the neural network for training.

Based on the training data set, the parameters of the preset radial basis neural network are trained using the Adam algorithm, including:

Calculate the gradients of the weights and bias parameters of a preset radial basis neural network , where N is the size of the training data set; _Li (P) is the loss function of the neural network; P represents the weight and bias parameters of the neural network;

The Adam algorithm calculates the parameter gradient The first moment estimate of and second moment estimates/> , and constantly correct the first-order moment estimate/> and second moment estimates/> The deviation is used to obtain the correction amount of the parameter, and the parameters of the preset radial basis neural network are determined based on the correction amount of the parameter. The specific calculation formula is as follows:

;

;

in, represents the decay rate of moment estimation,/> ,/> .

The method also includes:

Compute the corrected first moment estimate bias and the second moment estimate bias/> , N is the current iteration number, /> , ;

based on and/> , determine the updated values of the neural network weights and bias parameters as/> , where/> Indicates the step length; /> is a small constant.

Take 0.001, the /> Take 10 ^-8 .

The present invention uses a radial basis function neural network to identify short-circuit faults in power transmission lines. RBF neural networks perform well in nonlinear approximation and can usually converge to the global optimal solution faster because they have stronger approximation ability and nonlinear feature capture ability. RBF neural networks have better robustness and generalization capabilities to a certain extent. Because its hidden layer uses radial basis functions, it can better handle noisy data and nonlinear relationships, and can not only perform feature mapping on data and more efficiently extract core features from data, but also have a faster convergence speed, adapt to the distribution characteristics of data more quickly, and do not need to perform iterative optimization of back propagation like BP neural networks.

Establish an RBF neural network model with an input layer, a hidden layer, and an output layer in advance. The number of neurons in the input layer is the same as the dimension of the fault feature set, and the number of neurons in the output layer is related to the short-circuit fault type (short circuit, single-phase short circuit to ground, two-phase short circuit, two-phase short circuit to ground). In order to enable the neural network to memorize the connection between data and fault diagnosis results through the adjustment of internal parameters. It is necessary to use the collected fault feature set and the corresponding fault type to train the neural network system; that is, the extracted feature quantity T _1-7 is used as the input of the RBF neural network, and the short-circuit fault type is used as the output.

The present invention uses Adam to optimize the internal parameters of the neural network, making the final fault diagnosis more accurate. The advantage of the Adam algorithm in optimizing the neural network is that it can automatically adjust the learning rate of each parameter and update the parameter according to the first-order moment estimation and second-order moment estimation of the gradient of the parameter. This can avoid the trouble of manually adjusting the learning rate and also help to converge to the optimal solution faster.

Based on the power system fault detection method in Figure 1, an alarm can be issued based on the fault information obtained after detection, and the detection platform transmits the fault information and diagnostic data to the mobile terminal so as to timely determine the type of short-circuit fault, which is beneficial to the power staff for inspection and maintenance.

The above is an introduction to a method embodiment. The following is a further explanation of the scheme of the present invention through an apparatus embodiment.

FIG4 shows a block diagram of a power system fault detection device according to an embodiment of the present invention.

As shown in FIG4 , the device comprises:

The sampling module 401 is used to collect the mathematical expectation T ₁ of the zero-sequence current of the transmission line under fault conditions, the variance T ₂ of the zero-sequence current, the frequency variance T ₃ of each phase sampling point, the correlation coefficient T ₄ between each two-phase current, and the wavelet coefficients obtained after the scale decomposition of each phase current, and the feature quantities T ₅ , T ₆ , T ₇ extracted after the PCA dimensionality reduction processing;

The processing module 402 is used to use the seven feature value sequences T ₁ -T ₇ and the corresponding transmission line fault states as training data sets, and train the parameters of the preset radial basis neural network using the Adam algorithm based on the training data sets to obtain a trained power system fault detection model for detecting power system faults.

This embodiment also provides an electronic device, a readable storage medium, and a computer program product.

Fig. 5 shows a block diagram of an exemplary electronic device capable of implementing an embodiment of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and/or required herein.

The device 500 includes a computing unit 501, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the device 500 can also be stored. The computing unit 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input/output (I/O) interface 505 is also connected to the bus 504.

A number of components in the device 500 are connected to the I/O interface 505, including: an input unit 506, such as a keyboard, a mouse, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a disk, an optical disk, etc.; and a communication unit 509, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 509 allows the device 500 to exchange information/data with other devices through a computer network such as the Internet and/or various telecommunication networks.

The computing unit 501 may be a variety of general and/or special processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 501 performs the various methods and processes described above, such as the method in FIG. 1. For example, the method in FIG. 1 may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 508. Part or all of the computer program may be loaded and/or installed on the device 500 via the ROM 502 and/or the communication unit 509. When the computer program is loaded into the RAM 503 and executed by the computing unit 501, one or more steps of the method in FIG. 1 described above may be performed.

Claims (10)

1. A method of power system fault detection, the method comprising:

Acquiring a zero sequence current mathematical expectation T ₁, a zero sequence current variance T ₂, a frequency variance T ₃ of sampling points of each phase, a correlation coefficient T ₄ between each two phases of currents and a wavelet coefficient obtained after each phase of current scale decomposition of a power transmission line in a fault state, and extracting a characteristic quantity T ₅,T₆,T₇ after PCA dimension reduction treatment;

And taking the seven characteristic value sequences of T ₁-T₇ and the corresponding power transmission line fault states as training data sets, training the preset parameters of the radial basis neural network by using an Adam algorithm based on the training data sets, and obtaining a trained power system fault detection model for detecting power system faults.

2. The power system fault detection method according to claim 1, wherein the extracted feature quantity T ₅,T₆,T₇ is specifically that after the acquisition of the zero sequence current mathematical expectation T ₁, the zero sequence current variance T ₂, the frequency variance T ₃ of each phase sampling point, the correlation coefficient T ₄ between each two phases of current, and the wavelet coefficient obtained after each phase of current scale decomposition in the fault state is subjected to PCA dimension reduction processing:

Extracting three-phase current after fault occurrence, and calculating zero sequence current component Zero sequence current math expectation/>Zero sequence current variance/>，/>Frequency variance of sampling points of each phase，/>、/>、/>The frequencies of sampling points of phase A, phase B and phase C are respectively that the correlation coefficient T ₄ between each two phases of current is the Pelman coefficient between each two phases, namely/>The characteristic quantity T ₅,T₆,T₇ is the characteristic vector extracted by decomposing the original signal into low-frequency wavelet coefficients and high-frequency wavelet coefficients with different scales through discrete wavelet transformation and PCA dimension reduction treatment on the original signal, wherein/>，/>，，/>Is the average value of A phase fault current,/>Is the average value of B phase fault current,/>Is the average value of the C-phase fault current.

3. The power system fault detection method of claim 2, wherein the discrete wavelet transform comprises:

Taking scale factors by adopting binary sampling network Displacement/>J represents the number of decomposition layers, Z is an integer,/>Representing a positive integer, k being a parameter of the displacement factor;

the corresponding discrete wavelet function is:

；

discrete wavelet coefficients:

；

Wherein t is the time, Representing a function of the A phase fault current with time, wherein the function represents conjugation; carrying out multi-scale analysis on the signals by adopting a Mallat algorithm, decomposing the signals to be analyzed into frequency scales, and equivalently, carrying out low-pass and high-pass filtering on the signals for a plurality of times to obtain low-frequency wavelet coefficients and high-frequency wavelet coefficients under different layers;

Performing first-layer wavelet decomposition on an original signal by using a wavelet basis function to obtain a group of high-frequency wavelet coefficients and a group of low-frequency wavelet coefficients; the next layer of wavelet decomposition is carried out on the low-frequency wavelet coefficient, and the process is repeated until the required decomposition level is reached or the decomposition termination condition is reached.

4. A power system fault detection method according to claim 3, wherein the PCA dimension reduction process comprises:

Let matrix X _n×m consist of [ CA ₄,CD₁,CD₂,CD₃,CD₄ ], wherein CA ₄ is the fourth decomposed low frequency wavelet coefficient, CD ₁ is the first decomposed high frequency wavelet coefficient, CD ₂ is the second decomposed high frequency wavelet coefficient, CD ₃ is the third decomposed high frequency wavelet coefficient, CD ₄ is the fourth decomposed high frequency wavelet coefficient, n is the number of samples of the current signal, m is the variable number, m=5;

for each column X _n×5 Element/>，/>Through a mean value centering formulaObtaining a decentered matrix Y _n×m;

calculating covariance matrix among characteristic quantities by the formula Wherein Y is a characteristic quantity, and Y ^T is a transposed matrix of Y;

performing eigenvalue decomposition on the covariance matrix C, and calculating eigenvalue roots And feature vector/>；

The characteristic equation about Y isOrdering the determined eigenvalues: /(I)The corresponding feature vector is/>；

The accumulated contribution rate is larger than or equal to a preset threshold yz, and the accuracy of the principal component model is determined to meet the condition, namely

；

Taking feature vectors corresponding to the maximum K feature values；

A characteristic matrix X _AK of the A-phase current wavelet coefficient after dimension reduction, the matrix is composed of K characteristic vectors,Repeating the calculation to obtain/>，/>。

5. The power system fault detection method according to claim 1, wherein the training parameters of a preset radial basis neural network using Adam algorithm based on the training data set comprises:

Calculating the gradient of the weight and bias parameters of a preset radial basis function neural network Wherein N is the size of the training data set; l _i (P) is a loss function of the neural network; p represents the weight and bias parameters of the neural network;

adam algorithm calculates parameter gradients First moment estimation/>And second moment estimation/>Continuously correcting the first moment estimation/>And second moment estimation/>Obtaining a correction amount of the parameter, and determining a preset parameter of the radial basis neural network based on the correction amount of the parameter, wherein the specific calculation formula is as follows:

；

；

Wherein, Attenuation Rate representing moment estimation,/>，/>。

6. The power system fault detection method of claim 5, further comprising:

Calculating and correcting first moment estimated deviation And second moment estimation bias/>N is the current iteration number,/>，；

Based onAnd/>Determining the neural network weight and the bias parameter update value as/>Wherein/>Representing a step size; is a small constant.

7. The power system fault detection method of claim 6, wherein theTake 0.001, the/>10 ^-8 Parts were taken.

8. An electrical power system fault detection device, the device comprising:

The sampling module is used for collecting a zero sequence current mathematical expectation T ₁, a zero sequence current variance T ₂, a frequency variance T ₃ of sampling points of each phase, a correlation coefficient T ₄ between each two phases of currents and a wavelet coefficient obtained after each phase of current scale decomposition of the power transmission line under a fault state, and extracting a characteristic quantity T ₅,T₆,T₇ after PCA dimension reduction treatment;

The processing module is used for training parameters of a preset radial basis neural network by utilizing an Adam algorithm based on a training data set by taking the T ₁-T₇ seven eigenvalue sequences and the corresponding power transmission line fault states as the training data set to obtain a trained power system fault detection model for detecting power system faults.

9. An electronic device, comprising:

At least one processor;

and a memory communicatively coupled to the at least one processor;

characterized in that the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A non-transitory computer readable storage medium having stored thereon computer instructions that,

The computer instructions for causing the computer to perform the method according to any one of claims 1-7.