Fault positioning method and system for 35kV wind power plant current collection line

Abstract

The invention discloses a fault positioning method for a 35kV wind power plant current collection line, and relates to the field of fault positioning, and the method comprises the steps: collecting a line state signal when a fault occurs; carrying out decoupling processing on the line state signal; extracting a line time-frequency characteristic of the target line state signal; extracting a line random feature of the target line state signal; constructing and training a fault positioning model; and outputting a fault section position through the target fault positioning model. According to the invention, accurate and low-cost fault positioning can be effectively carried out in the wind power plant current collection line.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of fault location, and in particular to a fault location method and system for a 35kV wind farm collector line. Background Art

[0002] As wind energy gradually becomes one of the main sources of clean energy, the scale and number of wind farms are constantly expanding. However, in the operation of wind farms, the problem of collector line faults has always been one of the bottlenecks restricting the stable development of wind farms. The problem of locating collector line faults has always been the focus and difficulty of wind farm management.

[0003] Traditional collector line fault location methods include fault location based on pulse signal injection method. The pulse signal injection method achieves fault location by injecting high-voltage pulse signals on the faulty line and then detecting the fault point along the line. However, the use of the pulse signal injection method requires power outage and disconnection of the faulty line for fault location, which may affect the normal operation of the power system. Therefore, it is difficult to be widely used. Traditional collector line fault location methods also include installing fault indicators on the collector line to detect line current in real time. However, this method not only requires a large amount of capital investment, but also requires a lot of manpower for equipment maintenance, which seriously wastes manpower and material resources. Summary of the invention

[0004] The embodiments of the present application provide a fault location method and system for a 35kV wind farm collector line, which are used to solve the problem that it is difficult to accurately and cost-effectively locate faults in a wind farm collector line in the prior art.

[0005] To achieve the above objectives, the embodiments of the present application adopt the following technical solutions: In a first aspect, a fault location method for a 35 kV wind farm collector line is provided, the method comprising: The line status signal when a fault occurs is collected by using high-frequency sampling equipment pre-deployed in the target wind farm collection line; Decoupling the line state signal using a decoupling algorithm to obtain a target line state signal; Extracting line time-frequency characteristics from the target line state signal based on Fourier transform; Extracting feature vectors of the target line state signal based on random matrix theory to obtain line random features; Building a fault location model based on the neural network model, and performing model training on the fault location model based on a pre-built fault training set to obtain a target fault location model; The line time-frequency characteristics and the line random characteristics are input into the target fault location model, and the fault section position of the target wind farm collector line is output through the target fault location model.

[0006] Optionally, the decoupling process of the line state signal by using a decoupling algorithm to obtain a target line state signal comprises the following steps: Performing noise reduction processing on the line state signal by wavelet transform method to obtain a low-noise line state signal; Using a filtering algorithm to filter the low-noise line status signal; The low-noise line state signal that has completed filtering is decoupled using the Karenberg transformation formula to obtain a target line state signal.

[0007] Optionally, the performing noise reduction processing on the line state signal by using a wavelet transform method comprises the following steps: Performing multi-scale decomposition on the line state signal by using wavelet transform to obtain a plurality of wavelet line coefficients; All the wavelet line coefficients are threshold processed based on a wavelet threshold function, and the wavelet threshold function is as follows:

[0008] in, For the The scale of wavelet line coefficients, After threshold processing, The scale of wavelet line coefficients, is the preset adjustment factor, is the preset wavelet threshold, is the symbolic function, is an exponential function; All the wavelet line coefficients after threshold processing are subjected to signal reconstruction to obtain a low-noise line state signal.

[0009] Optionally, extracting the line time-frequency features from the target line state signal based on Fourier transform comprises the following steps: Decomposing the target line state signal into a plurality of local line state signals by using a sliding window method; Performing time domain analysis on all the local line state signals to obtain local time domain features of all the local line state signals, wherein the local time domain features include a local signal mean, a local signal variance, and a local signal peak value; Performing fast Fourier transform on all the local line status signals to obtain multiple local line frequency domain signals; Performing frequency domain analysis on all the local line frequency domain signals to obtain local frequency domain features of all the local line status signals, wherein the local frequency domain features include local spectrum amplitude, local spectrum phase and local spectrum frequency; The local time domain feature and the local frequency domain feature are integrated to obtain the line time-frequency feature of the target line state signal.

[0010] Optionally, extracting a feature vector from the target line state signal based on random matrix theory to obtain a line random feature comprises the following steps: Constructing a signal covariance matrix according to the target line state signal and based on random matrix theory; Performing eigenvalue decomposition on the signal covariance matrix to obtain a plurality of signal eigenvalues ​​and a plurality of signal eigenvectors; The signal feature vector is subjected to feature screening based on the signal feature value, and the line random feature of the target line state signal is determined according to the screening result.

[0011] Optionally, constructing a fault location model based on a neural network model, and performing model training on the fault location model based on a pre-constructed fault training set to obtain a target fault location model comprises the following steps: A fault location model is constructed based on a neural network model, wherein the fault location model includes a fault feature extraction module, a fault feature fusion module and a fault location classification module, wherein the fault feature extraction module is constructed based on a twin neural network model, the fault feature fusion module is constructed based on a self-attention mechanism, and the fault location classification module is constructed based on a probabilistic neural network model; The pre-constructed fault training set is input into the fault location model for model training to obtain a target fault location model.

[0012] Optionally, the step of inputting the pre-built fault training set into the fault localization model for model training to obtain a target fault localization model comprises the following steps: Inputting the fault training set into the fault feature extraction module, and extracting a first fault feature and a second fault feature from the fault training set through a first feature extraction layer and a second feature extraction layer in the fault feature extraction module; Inputting the first fault feature and the second fault feature into a similarity measurement layer of the fault feature extraction module, and outputting the fault feature similarity between the first fault feature and the second fault feature through the similarity measurement layer; Constructing a feature extraction loss function of the fault feature extraction module based on the fault feature similarity; Inputting the first fault feature and the second fault feature into the fault feature fusion module, using the fault feature fusion module to output a fused fault feature, and constructing a feature fusion loss function of the fault feature fusion module based on the fused fault feature; Inputting the fused fault feature into the fault location classification module, outputting a fault classification result through the fault location classification module, and constructing a fault classification loss function based on the fault classification result; Combining the feature extraction loss function, the feature fusion loss function and the fault classification loss function to obtain a comprehensive loss function value of the fault location model; The module parameters of the fault feature extraction module, the fault feature fusion module and the fault location classification module are updated and iterated based on the comprehensive loss function value until the comprehensive loss function value reaches a minimum value, thereby obtaining a target fault location model.

[0013] Optionally, the step of inputting the line time-frequency characteristics and the line random characteristics into the target fault location model and outputting the fault section position of the target wind farm collector line through the target fault location model comprises the following steps: Inputting the line time-frequency feature and the line random feature into the first feature extraction layer and the second feature extraction layer in the fault feature extraction module respectively to extract fault features, so as to obtain fault time-frequency feature and fault random feature; Inputting the fault time-frequency feature and the fault random feature into the fault feature fusion module for fault weighted fusion to obtain a fault fusion feature; The fault fusion feature is input into the fault location classification module for feature mapping, and the fault section location of the target wind farm collector line is output through the fault location classification module.

[0014] In a second aspect, the present application provides a fault location system for a 35kV wind farm collector line, characterized in that it includes: a memory configured to store instructions; and A processor is configured to call the instruction from the memory and implement the fault location method for a 35kV wind farm collector line according to any one of the first aspects when executing the instruction.

[0015] In a third aspect, the present application provides a machine-readable storage medium, characterized in that the machine-readable storage medium stores instructions for enabling a machine to execute a fault location method for a 35kV wind farm collector line according to any one of the first aspects.

[0016] Through the above technical scheme, the noise in the line status signal is removed by wavelet transform, and the de-noised line status signal is decoupled by using the Karen Bell transform formula, which can effectively separate the signals of different components, so as to more accurately identify the signal characteristics related to the fault, improve the accuracy of subsequent fault location, and use Fourier transform and random matrix theory to extract the characteristics of the target line status signal, which can effectively extract the key signal characteristics and improve the accuracy and efficiency of fault location. A target fault location model is constructed based on a neural network model, and the target fault location model is used to locate the fault. While improving the efficiency of fault location, it can also quickly adapt to other different types of collector lines and can be widely used. In summary, the present invention effectively solves the problem that the prior art is difficult to accurately and low-cost fault location in the wind farm collector line, and provides a strong guarantee for wind farm power transmission.

[0017] Other features and advantages of the embodiments of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic flow chart of a fault location method for a 35kV wind farm collector line provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the specific implementation methods described herein are only used to illustrate and explain the embodiments of the present application, and are not used to limit the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0020] It should be noted that if the embodiments of the present application involve directional indications (such as up, down, left, right, front, back...), such directional indications are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0021] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present application, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0022] Figure 1 The following schematically shows a flow chart of a method for locating a fault in a 35kV wind farm collector line according to an embodiment of the present application. Figure 1 As shown, the embodiment of the present application provides a method for locating a fault of a 35kV wind farm collector line, and the method may include the following steps: S101. Collect line status signals when a fault occurs through high-frequency sampling equipment pre-deployed in the target wind farm collection line.

[0023] In this embodiment, the high-frequency sampling equipment includes high-frequency current sensors, high-frequency voltage sensors and other equipment, and the line status signal includes the current, voltage and other signal data in the target wind farm collector line that can be used to judge the target wind farm collector line. This is because if a line fault occurs in the wind farm collector line, it often causes abnormal signals such as current and voltage. For example, when a line is short-circuited or open-circuited, the current may suddenly increase or decrease to zero, while the voltage may fluctuate or decrease abnormally. In practical applications, multiple high-frequency sampling devices may be deployed on the line to improve positioning accuracy through multi-point measurement. When a fault occurs, these devices will trigger recording at the same time, capture the line status signal when the fault occurs, and store the collected line status signal. In this way, high-quality, high-time resolution line status signals can be obtained, providing a reliable data basis for subsequent fault location analysis.

[0024] S102: Decoupling the line state signal using a decoupling algorithm to obtain a target line state signal.

[0025] In this embodiment, the line status signal is first denoised and filtered to remove noise and interference in the signal, improve the signal-to-noise ratio, and highlight the fault characteristics. Specifically, the signal is first denoised using the wavelet transform method. Wavelet transform is a time-frequency analysis method for continuous time signals. It can decompose the signal into frequency bands of different frequencies and amplitudes, provide more comprehensive and detailed frequency domain information, and is particularly suitable for processing non-stationary signals. Then, the low-noise line status signal after wavelet denoising is filtered using a filtering algorithm. Commonly used filtering algorithms include median filtering, mean filtering, Gaussian filtering, and Kalman filtering. Finally, the low-noise line status signal that has completed the above denoising and filtering steps is decoupled using the Karen Bell transform formula to obtain the target line status signal. This is because in a complex wind farm collector line, the target line status signal often contains multiple components, and there may be mutual interference or coupling between these components. If decoupling is not performed, these interference components may affect the accuracy of fault location. By decoupling the target line status signal, the signals of different components can be separated, which can simplify the target line status signal, thereby more accurately identifying the signal characteristics related to the fault, thereby improving the accuracy of subsequent fault location results.

[0026] S103. Extracting line time-frequency features from the target line status signal based on Fourier transform.

[0027] In this embodiment, Fourier transform can convert the time domain signal into the frequency domain. The target line state signal is analyzed in the frequency domain by Fourier transform, and the frequency domain characteristics of the target line state signal are extracted, which is a key step in fault location. The target line state signal is first decomposed into multiple local line state signals by using the sliding window method. This step can decompose a complex, long-term continuous signal into multiple shorter, easy-to-process local signals, thereby facilitating subsequent signal analysis, feature extraction or fault location. Among them, the sliding window method is a commonly used signal processing technology. It divides the target line state signal into multiple local line state signals by setting a fixed-length window on the time axis and sliding the window on the signal. Then, each local line status signal is analyzed in the time domain and frequency domain separately to obtain the local frequency domain features and line time-frequency features. Among them, the local frequency domain features include the local spectrum amplitude, the local spectrum phase and the local spectrum frequency. The local time domain features reflect the overall morphology and change characteristics of the local line status signal, which helps to distinguish different types of faults. The local frequency domain features include the local spectrum amplitude, the local spectrum phase and the local spectrum frequency. The frequency domain analysis can reveal the periodic components and specific frequency features in the fault signal that may be ignored by the time domain analysis, which is of great significance for identifying different types of faults. In addition, the frequency domain features are also very effective for detecting high-frequency transients and harmonic interference. Finally, the local time domain features and local frequency domain features are integrated into the line time-frequency features. The line time-frequency features are a multidimensional feature vector that provides comprehensive data support for subsequent fault location.

[0028] S104. Extract feature vectors of the target line state signal based on random matrix theory to obtain line random features.

[0029] In this embodiment, the random matrix theory is a mathematical theory that studies the intrinsic properties and statistical laws of matrices with random elements. First, a target line state matrix is ​​constructed according to the target line state signal, and then the target line state matrix is ​​converted into a signal covariance matrix using the covariance matrix formula. Then, the signal covariance matrix is ​​subjected to eigenvalue decomposition to obtain multiple eigenvalues, namely signal eigenvalues. Each signal eigenvalue corresponds to an eigenvector, namely a signal eigenvector. The signal eigenvalue represents the variance of the data in the direction of the signal eigenvector, namely the degree of dispersion of the data in this direction. The signal eigenvalues ​​are sorted in order from large to small, and the eigenvectors corresponding to the first few eigenvalues ​​are selected as key eigenvectors. The selected key eigenvectors are formed into a feature space, and the original signal, namely the target line state signal, is projected into the feature space to obtain the eigenvector after dimensionality reduction, namely the line random feature. The line random feature can accurately describe the key features in the target line state signal and improve the accuracy of the subsequent fault location results.

[0030] S105: construct a fault location model based on the neural network model, and perform model training on the fault location model based on a pre-constructed fault training set to obtain a target fault location model.

[0031] In this embodiment, the fault location model includes a fault feature extraction module, a fault feature fusion module and a fault location classification module. The fault feature extraction module is constructed based on a twin neural network model. The twin neural network model is a coupling architecture based on two artificial neural networks. The core of the twin neural network lies in its weight sharing characteristics. Specifically, the twin neural network model includes two neural networks with the same structure and weight sharing. The two networks receive different inputs, but share the same weights and parameters. By comparing the outputs of the two networks, the similarity of the two input samples can be calculated. The fault feature fusion module is constructed based on the self-attention mechanism. The self-attention mechanism pays higher attention to more discriminative features by learning attention masks, and weights are assigned to features of different dimensions respectively, thereby improving the performance of the fault location model. The fault location classification module is constructed based on a probabilistic neural network model. The probabilistic neural network model is a neural network model based on statistical principles, which aims to minimize the expected risk of misclassification through Bayesian decision rules. The probabilistic neural network model mainly includes an input layer, a hidden layer and an output layer. The fault location classification module can efficiently map multi-dimensional and highly complex fault feature data to a specific fault location, thereby realizing accurate fault location of the target wind farm collector line.

[0032] The pre-constructed fault training set includes wind farm collector line fault samples with pre-annotated faults. The wind farm collector line fault samples are input into the fault location model for model training. The model performance is evaluated by constructing a suitable loss function. Optimizers such as Adam or SGD are selected for model training. The model parameters are updated through the back propagation algorithm. The loss and accuracy during the training process are monitored. During the model training process, the model parameters of the fault location model are continuously optimized until the comprehensive loss function value reaches the minimum value, thereby obtaining the target fault location model.

[0033] S106: Input the line time-frequency characteristics and the line random characteristics into the target fault location model, and output the fault section position of the target wind farm collector line through the target fault location model.

[0034] In this embodiment, the line time-frequency features and line random features are respectively input into the first feature extraction layer and the second feature extraction layer in the fault feature extraction module for convolution operation, and the fault time-frequency features and the fault random features are extracted. Then the fault time-frequency features and the fault random features are input into the fault feature fusion module for fault weighted fusion. The fault feature fusion module is constructed based on the self-attention mechanism. The self-attention mechanism can give different weights to features, so that the model pays more attention to key features, thereby improving the accuracy of fault identification. The attention weights of the fault time-frequency features and the fault random features are allocated based on the self-attention mechanism, and then the fault time-frequency features and the fault random features after the weight allocation are weighted summed to obtain the fault fusion features. The fault fusion features are input into the fault location classification module for feature classification. The fault location classification module can efficiently map the multi-dimensional and highly complex fault fusion features to the specific fault location, thereby realizing the accurate fault location of the target wind farm collector line and obtaining the fault section location.

[0035] In one embodiment, decoupling the line state signal using a decoupling algorithm to obtain a target line state signal includes the following steps: The line state signal is subjected to noise reduction processing by wavelet transform method to obtain a low-noise line state signal; Filtering the low-noise line status signal using a filtering algorithm; The Karenberg transform formula is used to decouple the low-noise line state signal that has completed filtering processing to obtain the target line state signal.

[0036] In this embodiment, the wavelet transform method is first used to remove noise and interference in the line state signal. Specifically, the line state signal is first subjected to wavelet transform to obtain its wavelet coefficients, namely, wavelet line coefficients. These wavelet line coefficients contain key information of the line state signal, wherein the wavelet line coefficients generated by normal signals are larger, while the wavelet line coefficients with noise are smaller. Therefore, threshold processing is performed on the wavelet line coefficients, that is, coefficients below a certain threshold are considered to be noise, which are set to zero or subjected to other processing (such as soft threshold processing, i.e., shrinkage processing). Finally, the wavelet line coefficients that have completed the threshold processing are inversely transformed to obtain a signal after noise elimination, i.e., a low-noise line state signal.

[0037] Then, a filtering algorithm is used to filter the low-noise line status signal after wavelet denoising. Commonly used filtering algorithms include median filtering, mean filtering, Gaussian filtering, and Kalman filtering. Taking Gaussian filtering as an example, a Gaussian function is used as a weight coefficient to perform weighted averaging on the data points in the low-noise line status signal. Since its weight decreases exponentially with increasing distance, it can well preserve the local characteristics of the signal and complete the filtering of the signal.

[0038] Finally, the Karl-Peter transform formula is used to decouple the low-noise line status signal that has completed the above denoising and filtering steps to obtain the target line status signal. This is because in complex wind farm collection lines, the target line status signal often contains multiple components, and there may be mutual interference or coupling between these components. If decoupling is not performed, these interference components may affect the accuracy of fault location. By decoupling the target line status signal, the signals of different components can be separated, and the target line status signal can be simplified, so that the signal characteristics related to the fault can be more accurately identified, thereby improving the accuracy of subsequent fault location results. The Karl-Peter transform (Clarke Transform) is a mathematical tool for converting three-phase AC signals into two-phase DC signals. It is often used to decouple three-phase signals in power systems. The Karl-Peter transform formula is as follows:

[0039] in, , , It is a three-phase signal, that is, a low-noise line status signal that has been filtered. , It is a two-phase signal after Karen Bell transformation, that is, the target line state signal.

[0040] In one embodiment, performing noise reduction processing on the line state signal by using a wavelet transform method comprises the following steps: The line state signal is decomposed into multiple scales using wavelet transform to obtain multiple wavelet line coefficients; All wavelet line coefficients are threshold processed based on the wavelet threshold function. The wavelet threshold function is as follows:

[0041] in, For the The scale of wavelet line coefficients, After threshold processing, The scale of wavelet line coefficients, is the preset adjustment factor, is the preset wavelet threshold, is the symbolic function, is an exponential function; All wavelet line coefficients after threshold processing are reconstructed to obtain low-noise line status signals.

[0042] In this embodiment, the wavelet transform can decompose the line state signal into components on multiple scales, and each component on the scale corresponds to a wavelet coefficient, namely, the wavelet line coefficient. These wavelet line coefficients not only contain the time information of the line state signal, but also contain the frequency information of the signal, so they can fully reflect the characteristics of the signal at different scales. Then, all the wavelet line coefficients are threshold processed, and the purpose is to remove noise. This is because the wavelet line coefficients generated by the normal signal are large, while the wavelet line coefficients with noise are small. Therefore, the purpose of removing noise can be achieved by threshold processing the wavelet line coefficients. In this embodiment, all the wavelet line coefficients are sequentially input into the wavelet threshold function for threshold processing. Compared with the compromise threshold function, the wavelet threshold function can better retain the detailed information of the line state signal, and the wavelet line coefficient decays according to the exponential law, and the denoising effect is further improved. Finally, all the wavelet line coefficients after threshold processing are reconstructed. Signal reconstruction is one of the key steps of the wavelet denoising method. The wavelet inverse transform algorithm is mainly used to recombine all the wavelet line coefficients that have completed the threshold processing. The combined output result is the signal after noise elimination, that is, the low-noise line state signal.

[0043] In one embodiment, extracting the line time-frequency characteristics from the target line state signal based on Fourier transform includes the following steps: Decomposing the target line state signal into multiple local line state signals by using a sliding window method; Performing time domain analysis on all local line state signals to obtain local time domain features of all local line state signals, the local time domain features including local signal mean, local signal variance, and local signal peak value; Performing fast Fourier transform on all local line status signals to obtain multiple local line frequency domain signals; Performing frequency domain analysis on all local line frequency domain signals to obtain local frequency domain features of all local line status signals, the local frequency domain features including local spectrum amplitude, local spectrum phase and local spectrum frequency; The local time domain features and the local frequency domain features are integrated to obtain the line time-frequency features of the target line state signal.

[0044] In this embodiment, a sliding window method is used to segment the target line state signal into multiple local line state signals. Specifically, a fixed-size window is slid on the time axis to divide the continuous signal into a series of overlapping short-time segments. The selection of the window size is usually based on the characteristic frequency and expected duration of the target line state signal, and the typical value may be between a few milliseconds and tens of milliseconds. The sliding step size determines the degree of overlap between adjacent windows. A smaller step size can provide more detailed analysis, but it will increase the amount of calculation. In actual operation, window functions such as rectangular windows, Hanning windows, or Hamming windows can be used to smooth signal edges and reduce spectral leakage. Multiple local line state signals are obtained, and Fourier transforms are performed on the multiple local line state signals separately. The time domain features of all local line state signals are first extracted, that is, the local time domain features. The local time domain features intuitively reflect the changing characteristics of the signal in the time dimension. Among them, the local signal mean μ reflects the overall level of the signal, and the calculation formula is , where N is the number of sampling points, is the value of the i-th sampling point. Local signal variance Describes the degree of fluctuation of the signal around the mean, and the calculation formula is , the local signal peak value K describes the sharpness of the signal distribution, and the calculation formula is , where E represents the mathematical expectation. The above local time domain features reflect the overall shape and change characteristics of the local line status signal, which helps to distinguish different types of faults.

[0045] Through Fourier transform, the frequency domain analysis of all local line frequency domain signals is performed to extract the frequency domain characteristics of the local line status signal, that is, the local frequency domain characteristics. Fourier transform can convert the time domain signal to the frequency domain and reveal the frequency composition of the signal. In actual calculations, in order to improve efficiency, the fast Fourier transform (FFT) algorithm is usually used for frequency domain analysis to extract the local frequency domain characteristics of all local line status signals. The formula of the fast Fourier transform (FFT) algorithm is as follows:

[0046] in, is the local route frequency domain signal frequency components, It is the local line status signal (time domain signal) A point in time, is a negative exponential function, representing sine and cosine waves of different frequencies. is the number of signal samples, that is, the number of local route frequency domain signals.

[0047] The local frequency domain features include local spectrum amplitude, local spectrum phase and local spectrum frequency. The local spectrum amplitude indicates the amplitude of the signal at each frequency component, usually taking the absolute value. The local spectrum phase indicates the phase information of the signal at each frequency component, describing the phase offset of the signal waveform at that frequency. The local spectrum frequency refers to each frequency component contained in the signal, which is related to the number of sampling points and sampling frequency of the fast Fourier transform. Frequency domain analysis can reveal periodic components and specific frequency features in the fault signal that may be ignored by time domain analysis, which is of great significance for identifying different types of faults (such as single-phase grounding, phase-to-phase short circuit, etc.). In addition, frequency domain features are also very effective in detecting high-frequency transients and harmonic interference.

[0048] The local time domain features and local frequency domain features are integrated to obtain the line time-frequency features of the target line status signal. This step fuses the information in the time domain and frequency domain to form a multidimensional feature vector, which provides comprehensive data support for subsequent fault location. During the integration process, the scale and importance of different features need to be considered. Feature standardization methods such as z-score standardization can be used. This can eliminate the dimensional differences between different features. This comprehensive feature not only contains the statistical characteristics and frequency information of the signal, but also retains the potential correlation between the time domain and frequency domain features. Through this integration, the characteristics of the local line status signal can be more comprehensively described, the accuracy and reliability of fault location can be improved, and strong data support can be provided for subsequent fault location.

[0049] In one embodiment, extracting a feature vector of a target line state signal based on random matrix theory to obtain a line random feature includes the following steps: Constructing a signal covariance matrix according to the target line state signal and based on random matrix theory; Performing eigenvalue decomposition on the signal covariance matrix to obtain multiple signal eigenvalues ​​and multiple signal eigenvectors; The signal feature vector is feature screened based on the signal feature value, and the line random feature of the target line state signal is determined according to the screening result.

[0050] In this embodiment, random matrix theory is combined with statistical theory to analyze the fault characteristics in the target wind farm collector line from the perspective of the stability of the matrix composed of the target line state signal, which to a certain extent avoids the problem of excessive reliance on the collector line model when processing data by traditional analysis methods. Specifically, the target line state matrix is ​​first constructed according to the target line state signal, and then the target line state matrix X is converted into the signal covariance matrix C using the covariance matrix formula. The covariance matrix formula is: , where n is the number of samples and T is the transpose of the matrix. Then, the signal covariance matrix is ​​decomposed into eigenvalues ​​to obtain multiple eigenvalues, namely signal eigenvalues. Each signal eigenvalue corresponds to an eigenvector, namely the signal eigenvector. The signal eigenvalue represents the variance of the data in the direction of the signal eigenvector, namely the degree of dispersion of the data in this direction. The signal eigenvalues ​​are sorted from large to small, and the eigenvectors corresponding to the first few eigenvalues ​​are selected as key eigenvectors. The selected key eigenvectors are combined into a feature space. By projecting the original signal, namely the target line state signal, into the feature space, the reduced eigenvector, namely the line random feature, is obtained. The line random feature can accurately describe the key features in the target line state signal and improve the accuracy of subsequent fault location results.

[0051] In one embodiment, a fault location model is constructed based on a neural network model, and the fault location model is trained based on a pre-constructed fault training set to obtain a target fault location model, including the following steps: A fault location model is constructed based on a neural network model. The fault location model includes a fault feature extraction module, a fault feature fusion module and a fault location classification module. The fault feature extraction module is constructed based on a twin neural network model, the fault feature fusion module is constructed based on a self-attention mechanism, and the fault location classification module is constructed based on a probabilistic neural network model. The pre-built fault training set is input into the fault location model for model training to obtain the target fault location model.

[0052] In this embodiment, the fault location model includes a fault feature extraction module, a fault feature fusion module and a fault location classification module. The fault feature extraction module is constructed based on a twin neural network model. The twin neural network model is a coupling architecture based on two artificial neural networks. The core of the twin neural network lies in its weight sharing characteristics. Specifically, the twin neural network model includes two neural networks with the same structure and shared weights, respectively referred to as Network1 and Network2. The two networks receive different inputs, but share the same weights and parameters. By comparing the outputs of the two networks, the similarity of the two input samples can be calculated. In the process of building and training the fault location model, in order to make the output results of the fault location model more accurate, a large number of wind farm collector line fault samples are usually required for model training, but the actual situation is that wind farm collector line fault samples are relatively scarce, and even using power system simulation software to form samples is time-consuming and laborious. This greatly limits the application of deep learning algorithms in the field of fault location. Therefore, the present invention uses twin neural networks to enhance data samples and assist in training recognition networks to solve the problem of scarce actual fault samples, effectively improving the utilization efficiency of existing samples. The fault feature fusion module is built based on the self-attention mechanism. The self-attention mechanism pays more attention to more discriminative features by learning attention masks, and assigns weights to features of different dimensions, thereby improving the performance of the fault location model. The fault location classification module is built based on the probabilistic neural network model. The probabilistic neural network model is a neural network model based on statistical principles, which aims to minimize the expected risk of misclassification through the Bayesian decision rule. The probabilistic neural network model mainly includes an input layer, a hidden layer, and an output layer. The fault location classification module can efficiently map multi-dimensional and highly complex fault feature data to the specific fault location, thereby achieving accurate fault location of the target wind farm collector line.

[0053] The pre-constructed fault training set includes wind farm collector line fault samples with pre-annotated faults. The wind farm collector line fault samples are input into the fault location model for model training. The model performance is evaluated by constructing a suitable loss function. Optimizers such as Adam or SGD are selected for model training. The model parameters are updated through the back propagation algorithm. The loss and accuracy during the training process are monitored. During the model training process, the model parameters of the fault location model are continuously optimized until the comprehensive loss function value reaches the minimum value, thereby obtaining the target fault location model.

[0054] In one embodiment, inputting a pre-built fault training set into a fault localization model for model training to obtain a target fault localization model comprises the following steps: Inputting the fault training set into the fault feature extraction module, and extracting the first fault feature and the second fault feature from the fault training set through the first feature extraction layer and the second feature extraction layer in the fault feature extraction module; Inputting the first fault feature and the second fault feature into a similarity measurement layer of a fault feature extraction module, and outputting the fault feature similarity between the first fault feature and the second fault feature through the similarity measurement layer; Constructing the feature extraction loss function of the fault feature extraction module based on the fault feature similarity; Inputting the first fault feature and the second fault feature into a fault feature fusion module, using the fault feature fusion module to output a fused fault feature, and constructing a feature fusion loss function of the fault feature fusion module based on the fused fault feature; The fused fault features are input into the fault location classification module, the fault classification result is output through the fault location classification module, and a fault classification loss function is constructed based on the fault classification result; The comprehensive loss function value of the fault location model is obtained by combining the feature extraction loss function, feature fusion loss function and fault classification loss function; The module parameters of the fault feature extraction module, the fault feature fusion module and the fault location classification module are updated and iterated based on the comprehensive loss function value until the comprehensive loss function value reaches the minimum value, and the target fault location model is obtained.

[0055] In this embodiment, first, the fault training set is input into the fault feature extraction module. The fault feature extraction module includes a first feature extraction layer, a second feature extraction layer and a similarity measurement layer. The first feature extraction layer and the second feature extraction layer are used to extract the first fault feature and the second fault feature. The similarity measurement layer is used to calculate the fault feature similarity between the first fault feature and the second fault feature. The first feature extraction layer and the second feature extraction layer can use convolution kernels for feature extraction. The similarity measurement layer can use, for example, Euclidean distance, cosine similarity, etc. to calculate the fault feature similarity between the first fault feature and the second fault feature. The loss function of the fault feature extraction module aims to reduce the distance between the fault features whose feature fault feature similarity is greater than a preset threshold, that is, to increase their similarity, and to increase the distance between the fault features whose feature fault feature similarity is greater than a preset threshold, that is, to reduce their similarity. A feature extraction loss function is constructed based on the fault feature similarity between the first fault feature and the second fault feature. A contrast loss function or a triplet loss function can be used. The feature extraction loss function is used to evaluate the similarity between the first fault feature and the second fault feature, and guide the model to optimize the feature extraction capability.

[0056] Then, the first fault feature and the second fault feature are input into the fault feature fusion module, and the first fault feature and the second fault feature are weightedly fused using the self-attention mechanism. Specifically, the first fault feature and the second fault feature are first vectorized and converted into a feature matrix. The feature matrix is ​​mapped to three vectors of query, key, and value using linear changes. The dot product between the query vector and the key vector is calculated to obtain an attention score matrix. The dot product result is divided by a scaling factor (which can be the square root of the key vector) to stabilize the gradient. The scaled attention score matrix is ​​converted into a probability value between [0,1] using a normalization function to obtain an attention weight. The normalization function can be a softmax function. The obtained attention weight is multiplied by the value vector to complete the attention weight allocation of the first fault feature and the second fault feature. The first fault feature and the second fault feature after the weight allocation are weighted summed to obtain the fused fault feature. The feature fusion loss function of the fault feature fusion module is constructed based on the fused fault feature. The loss function can use cross entropy or KL divergence. The feature fusion loss function is used to measure the matching degree between the fused fault feature and the real feature.

[0057] Then, the fused fault features are input into the fault location classification module. The fault location classification module is constructed based on the probabilistic neural network model. The probabilistic neural network model is a neural network model based on statistical principles, which aims to minimize the expected risk of misclassification through the Bayesian decision rule. The probabilistic neural network model mainly includes an input layer, a hidden layer and an output layer. The fault location classification module can efficiently map multi-dimensional and highly complex fault feature data to the specific fault location, thereby realizing the accurate fault location of the target wind farm collector line. The fault location classification module can learn and simulate the complex mapping relationship between fault features and fault locations through the nonlinear activation function between multiple layers of neurons. The fault location classification module is trained using fault samples, and its parameters such as smoothing factor and Gaussian kernel bandwidth parameter are continuously adjusted during the training process. The fault classification loss function of the fault location classification module is constructed based on the fault classification result. The loss function can sample the cross entropy loss function or the negative log-likelihood loss function, etc. The gap between the fault classification result and the true result is evaluated according to the fault classification loss function.

[0058] Finally, the comprehensive loss function of the fault location model is obtained by combining the feature extraction loss function, feature fusion loss function and fault classification loss function. The comprehensive loss function can be expressed as the weighted sum of the above three loss functions. The weights of the three loss functions can be determined by methods such as cross-validation or grid search to determine the optimal value. The comprehensive loss function value of the comprehensive loss function is calculated. The calculation process involves forward propagating the fault training set through the model, then comparing it with the true value to obtain each loss term, and finally weighted summing it. Based on the comprehensive loss function value, the module parameters of the fault feature extraction module, the fault feature fusion module and the fault location classification module are iteratively updated until the comprehensive loss function value reaches the minimum value or the preset maximum number of iterations to obtain the target fault location model. The update process can use optimization algorithms such as stochastic gradient descent (SGD) or Adam.

[0059] In one embodiment, inputting the line time-frequency characteristics and the line random characteristics into the target fault location model, and outputting the fault section location of the target wind farm collector line through the target fault location model includes the following steps: Inputting the line time-frequency feature and the line random feature into the first feature extraction layer and the second feature extraction layer in the fault feature extraction module respectively to extract the fault feature, so as to obtain the fault time-frequency feature and the fault random feature; Input the fault time-frequency features and fault random features into the fault feature fusion module for fault weighted fusion to obtain the fault fusion features; The fault fusion features are input into the fault location classification module for feature mapping, and the fault section location of the target wind farm collector line is output through the fault location classification module.

[0060] In this embodiment, the line time-frequency features and the line random features are respectively input into the first feature extraction layer and the second feature extraction layer in the fault feature extraction module for convolution operation, and the fault time-frequency features and the fault random features are extracted. Specifically, the first feature extraction layer and the second feature extraction layer respectively perform multiple convolutions (three times), introduce nonlinear factors through the activation function, perform feature dimension reduction, and finally output the features after passing through the fully connected layer to obtain the fault time-frequency features and the fault random features. Then the fault time-frequency features and the fault random features are input into the fault feature fusion module for fault weighted fusion. The fault feature fusion module is constructed based on the self-attention mechanism. The self-attention mechanism first vectorizes the fault time-frequency features and the fault random features and converts them into a feature matrix. The feature matrix is ​​mapped to three vectors of query, key and value using linear changes. The dot product between the query vector and the key vector is calculated to obtain the attention score matrix. The dot product result is divided by a scaling factor (which can be the square root of the key vector) to stabilize the gradient. The attention score matrix after scaling is converted into a probability value between [0,1] using a normalization function (such as a softmax function). The obtained attention weight is multiplied by the value vector to complete the attention weight allocation of the fault time-frequency features and the fault random features. The fault time-frequency features and the fault random features after the weight allocation are weighted and summed to obtain the fault fusion features. The fault fusion features are input into the fault location classification module for feature classification. The fault location classification module can efficiently map the multi-dimensional and highly complex fault fusion features to the specific fault location, thereby realizing the accurate fault location of the target wind farm collector line and obtaining the fault section location.

[0061] The present application also discloses a fault location system for a 35kV wind farm collector line, which is characterized by comprising: a memory configured to store instructions; and A processor is configured to call instructions from a memory and implement the fault location method for a 35kV wind farm collector line according to any one of the above items when executing the instructions.

[0062] The present application also discloses a machine-readable storage medium, characterized in that the machine-readable storage medium stores instructions, and the instructions are used to enable a machine to execute a fault location method for a 35kV wind farm collector line according to any one of the above items.

[0063] Among them, the processor can adopt a central processing unit (CPU). Of course, according to actual usage, other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be adopted. The general-purpose processor can adopt a microprocessor or any conventional processor, etc., and this application does not impose any restrictions on this.

[0064] Among them, the memory can be an internal storage unit of a computer device, such as a hard disk or memory of a computer device, or an external storage device of a computer device, such as a plug-in hard disk, a smart memory card (SMC), a secure digital card (SD) or a flash memory card (FC) equipped on the computer device, and the memory can also be a combination of an internal storage unit and an external storage device of a computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or is to be output, and this application does not impose any restrictions on this.

[0065] An embodiment of the present application further provides a machine-readable storage medium having instructions stored thereon, the instructions being used to enable a machine to execute the above-mentioned method for locating a fault of a 35kV wind farm collector line.

[0066] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0067] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0068] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0069] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0070] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0071] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0072] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0073] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, 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 process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0074] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A fault location method for a 35kV wind farm collector line, characterized in that: The method comprises the following steps: The line status signal when a fault occurs is collected by using high-frequency sampling equipment pre-deployed in the target wind farm collection line; Decoupling the line state signal using a decoupling algorithm to obtain a target line state signal; Extracting line time-frequency characteristics from the target line state signal based on Fourier transform; Extracting feature vectors of the target line state signal based on random matrix theory to obtain line random features; Building a fault location model based on the neural network model, and performing model training on the fault location model based on a pre-built fault training set to obtain a target fault location model; The line time-frequency characteristics and the line random characteristics are input into the target fault location model, and the fault section position of the target wind farm collector line is output through the target fault location model.

2. The method according to claim 1, characterized in that The decoupling process of the line state signal by using a decoupling algorithm to obtain a target line state signal comprises the following steps: Performing noise reduction processing on the line state signal by wavelet transform method to obtain a low-noise line state signal; Using a filtering algorithm to filter the low-noise line status signal; The low-noise line state signal that has completed filtering is decoupled using the Karenberg transformation formula to obtain a target line state signal.

3. The method according to claim 2, characterized in that The noise reduction process of the line status signal by wavelet transform method comprises the following steps: Performing multi-scale decomposition on the line state signal by using wavelet transform to obtain a plurality of wavelet line coefficients; All the wavelet line coefficients are threshold processed based on a wavelet threshold function, and the wavelet threshold function is as follows:

4. Among them, For the The scale of wavelet line coefficients, After threshold processing, The scale of wavelet line coefficients, is the preset adjustment factor, is the preset wavelet threshold, is the symbolic function, is an exponential function; All the wavelet line coefficients after threshold processing are subjected to signal reconstruction to obtain a low-noise line state signal.

5. The method according to claim 1, characterized in that The extracting of the line time-frequency characteristics from the target line state signal based on Fourier transform comprises the following steps: Decomposing the target line state signal into a plurality of local line state signals by using a sliding window method; Performing time domain analysis on all the local line state signals to obtain local time domain features of all the local line state signals, wherein the local time domain features include a local signal mean, a local signal variance, and a local signal peak value; Performing fast Fourier transform on all the local line status signals to obtain multiple local line frequency domain signals; Performing frequency domain analysis on all the local line frequency domain signals to obtain local frequency domain features of all the local line status signals, wherein the local frequency domain features include local spectrum amplitude, local spectrum phase and local spectrum frequency; The local time domain feature and the local frequency domain feature are integrated to obtain the line time-frequency feature of the target line state signal.

6. The method according to claim 1, characterized in that The extracting of the feature vector of the target line state signal based on random matrix theory to obtain the line random feature comprises the following steps: Constructing a signal covariance matrix according to the target line state signal and based on random matrix theory; Performing eigenvalue decomposition on the signal covariance matrix to obtain a plurality of signal eigenvalues ​​and a plurality of signal eigenvectors; The signal feature vector is subjected to feature screening based on the signal feature value, and the line random feature of the target line state signal is determined according to the screening result.

7. The method according to claim 1, characterized in that The method of constructing a fault location model based on a neural network model and training the fault location model based on a pre-constructed fault training set to obtain a target fault location model comprises the following steps: A fault location model is constructed based on a neural network model, wherein the fault location model includes a fault feature extraction module, a fault feature fusion module and a fault location classification module, wherein the fault feature extraction module is constructed based on a twin neural network model, the fault feature fusion module is constructed based on a self-attention mechanism, and the fault location classification module is constructed based on a probabilistic neural network model; The pre-constructed fault training set is input into the fault location model for model training to obtain a target fault location model.

8. The method according to claim 6, characterized in that The step of inputting the pre-built fault training set into the fault location model for model training to obtain the target fault location model comprises the following steps: Inputting the fault training set into the fault feature extraction module, and extracting a first fault feature and a second fault feature from the fault training set through a first feature extraction layer and a second feature extraction layer in the fault feature extraction module; Inputting the first fault feature and the second fault feature into a similarity measurement layer of the fault feature extraction module, and outputting the fault feature similarity between the first fault feature and the second fault feature through the similarity measurement layer; Constructing a feature extraction loss function of the fault feature extraction module based on the fault feature similarity; Inputting the first fault feature and the second fault feature into the fault feature fusion module, using the fault feature fusion module to output a fused fault feature, and constructing a feature fusion loss function of the fault feature fusion module based on the fused fault feature; Inputting the fused fault feature into the fault location classification module, outputting a fault classification result through the fault location classification module, and constructing a fault classification loss function based on the fault classification result; Combining the feature extraction loss function, the feature fusion loss function and the fault classification loss function to obtain a comprehensive loss function value of the fault location model; The module parameters of the fault feature extraction module, the fault feature fusion module and the fault location classification module are updated and iterated based on the comprehensive loss function value until the comprehensive loss function value reaches a minimum value, thereby obtaining a target fault location model.

9. The method according to claim 7, characterized in that: The step of inputting the line time-frequency characteristics and the line random characteristics into the target fault location model and outputting the fault section position of the target wind farm collector line through the target fault location model comprises the following steps: Inputting the line time-frequency feature and the line random feature into the first feature extraction layer and the second feature extraction layer in the fault feature extraction module respectively to extract fault features, so as to obtain fault time-frequency feature and fault random feature; Inputting the fault time-frequency feature and the fault random feature into the fault feature fusion module for fault weighted fusion to obtain a fault fusion feature; The fault fusion feature is input into the fault location classification module for feature mapping, and the fault section location of the target wind farm collector line is output through the fault location classification module.

10. A fault location system for a 35kV wind farm collector line, characterized in that: include: a memory configured to store instructions; as well as A processor is configured to call the instructions from the memory and implement the fault location method for a 35kV wind farm collector line according to any one of claims 1 to 8 when executing the instructions.

11. A machine-readable storage medium, characterized in that: The machine-readable storage medium stores instructions, which are used to enable a machine to execute the fault location method for a 35kV wind farm collector line according to any one of claims 1 to 8.

Images