Double-end fault positioning method based on signal enhancement

Abstract

The invention relates to the technical field of data processing, and discloses a double-end fault positioning method based on signal enhancement, which is used for improving the efficiency of double-end fault positioning based on signal enhancement. The method comprises the following steps: acquiring and preprocessing electrical signals at two ends of a power transmission line to obtain preprocessed synchronous electrical signals; performing adaptive enhancement processing on the preprocessed synchronous electrical signal to obtain an enhanced fault feature signal; performing time-frequency analysis and feature extraction on the enhanced fault feature signal to obtain a fault feature vector and a key parameter; performing machine learning analysis on the fault feature vectors and the key parameters to obtain a fault type identification result and initial fault position estimation; performing multi-algorithm fusion analysis on the preliminary fault position estimation to obtain an accurate fault position coordinate; and performing geographic information mapping and field verification on the accurate fault position coordinates to obtain a fault positioning result and system optimization data.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a double-end fault location method based on signal enhancement. Background Art

[0002] In the operation and management of modern distribution networks, inspection is increasingly valued as an important part of ensuring the normal operation of power equipment. Existing technologies usually use regular manual inspections and basic automatic monitoring methods to evaluate the status and detect faults of equipment in distribution networks. These methods collect the operating data and image information of the equipment, rely on traditional image processing technology and simple feature extraction algorithms to detect abnormalities in the equipment. However, with the increase in the types and number of distribution network equipment, traditional manual inspection methods are inefficient and prone to missing problems, and cannot reflect the status of equipment in a timely and comprehensive manner.

[0003] Although the existing technology has improved the inspection efficiency of the distribution network to a certain extent, there are still many shortcomings. On the one hand, the traditional image processing algorithm is sensitive to image noise in complex environments, resulting in poor image data processing effects and difficulty in accurately extracting equipment features. On the other hand, the existing anomaly detection methods lack intelligence and automation, cannot respond to the dynamic changes of equipment failures in real time, and cannot provide effective decision support. This makes the location and evaluation of equipment failures have a certain lag, affecting the safe and stable operation of the distribution network. Therefore, there is an urgent need for a more intelligent and efficient inspection management method to solve the many problems existing in the existing technology. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides a dual-terminal fault location method based on signal enhancement, which is used to improve the efficiency of dual-terminal fault location based on signal enhancement.

[0005] The present invention provides a double-end fault location method based on signal enhancement, comprising: collecting and preprocessing electrical signals at both ends of a transmission line to obtain preprocessed synchronous electrical signals; performing adaptive enhancement processing on the preprocessed synchronous electrical signals to obtain enhanced fault characteristic signals; performing time-frequency analysis and feature extraction on the enhanced fault characteristic signals to obtain fault characteristic vectors and key parameters; performing machine learning analysis on the fault characteristic vectors and key parameters to obtain fault type identification results and preliminary fault location estimates; performing multi-algorithm fusion analysis on the preliminary fault location estimates to obtain precise fault location coordinates; performing geographic information mapping and field verification on the precise fault location coordinates to obtain fault location results and system optimization data.

[0006] In the technical solution provided by the present invention, the electrical signals at both ends of the transmission line are collected and preprocessed to obtain preprocessed synchronous electrical signals, which effectively eliminates the noise and interference in the original signal and improves the data quality of subsequent analysis. The preprocessed synchronous electrical signal is adaptively enhanced to obtain an enhanced fault feature signal, which highlights the fault characteristics, suppresses background noise, and greatly improves the recognition rate of fault characteristics. Subsequently, the enhanced fault feature signal is subjected to time-frequency analysis and feature extraction to obtain a fault feature vector and key parameters, which comprehensively capture the time domain and frequency domain characteristics of the fault signal, laying the foundation for accurate positioning. The fault feature vector and key parameters are subjected to machine learning analysis to obtain the fault type identification result and the preliminary fault location estimation. This innovative application of machine learning technology realizes the automatic identification and preliminary positioning of the fault type, greatly improving the efficiency and accuracy of positioning. The preliminary fault location estimation is subjected to multi-algorithm fusion analysis to obtain the precise fault location coordinates, which combines the advantages of multiple positioning algorithms and significantly improves the positioning accuracy. Finally, the precise fault location coordinates were mapped and verified on the spot to obtain the fault location results and system optimization data, which not only verified the accuracy of the location results, but also continuously optimized the system performance through the feedback mechanism. Overall, this method greatly improved the accuracy, efficiency and reliability of fault location through the organic combination of signal enhancement, feature extraction, machine learning and multi-algorithm fusion. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0008] Figure 1 The present invention is a flowchart of a double-terminal fault location method based on signal enhancement in an embodiment of the present invention. DETAILED DESCRIPTION

[0009] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0010] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0011] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0012] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , Figure 1 is a flow chart of a double-terminal fault location method based on signal enhancement according to an embodiment of the present invention. Figure 1 As shown, the following steps are included: S101, collecting and preprocessing electrical signals at both ends of the transmission line to obtain preprocessed synchronous electrical signals; S102, performing adaptive enhancement processing on the pre-processed synchronous electrical signal to obtain an enhanced fault characteristic signal; S103, performing time-frequency analysis and feature extraction on the enhanced fault characteristic signal to obtain a fault characteristic vector and key parameters; S104, performing machine learning analysis on the fault feature vector and key parameters to obtain a fault type identification result and a preliminary fault location estimation; S105, performing multi-algorithm fusion analysis on the preliminary fault location estimation to obtain accurate fault location coordinates; S106. Perform geographic information mapping and field verification on the precise fault location coordinates to obtain fault location results and system optimization data.

[0013] It should be noted that the voltage and current signals are collected simultaneously at both ends of the transmission line through high-frequency sampling equipment. The sampling frequency is usually set between several thousand hertz and tens of thousands of hertz to capture transient fault characteristics. After the collected original signal is timestamped, it is processed by wavelet denoising to eliminate background noise, and then the pulse interference is removed by adaptive median filtering. The filtered signal is subjected to outlier detection and interpolation processing to ensure the continuity and reliability of the data. Then, the processed signal is Fourier transformed to extract the frequency domain features, and the target frequency band signal is retained through a bandpass filter. Finally, the Hilbert transform is used for phase correction to ensure the time synchronization of the signals at both ends and obtain the preprocessed synchronized electrical signal.

[0014] First, the signal is decomposed into signal components of multiple scales using wavelet packet decomposition. Then, the energy distribution of these components is analyzed to select the key components containing fault characteristics. The selected components are sparsely represented and singular value decomposed to extract the main characteristic components. The fault characteristics are highlighted and the background noise is suppressed through adaptive gain adjustment. Subsequently, the signal is further decomposed using empirical mode decomposition to extract the intrinsic mode function, and the mode function most related to the fault is selected through correlation analysis. Finally, an adaptive filter is designed to optimize these key mode functions and reconstruct the enhanced fault characteristic signal. The time-frequency spectrum of the signal is obtained using short-time Fourier transform, and the characteristic frequency points are identified by peak detection. Then, the time-frequency spectrum is segmented according to these characteristic points, and the energy density distribution of each region is calculated. The energy distribution characteristics are analyzed by principal component analysis to extract the main characteristic components and perform standardization. At the same time, the signal is decomposed at multiple scales using wavelet packet transform to extract statistical features. These features are fused to form the final fault feature vector, and key parameters such as the time and duration of the fault are calculated. The feature vector is reduced in dimension to reduce data redundancy. Then, cluster analysis methods such as K-means clustering are used to group the feature data after dimensionality reduction. A decision tree model is constructed based on the clustering results, and simplified classification rules are obtained through pruning optimization. The optimized decision rules are trained using the support vector machine algorithm to obtain a fault type classifier. For location estimation, a neural network model is constructed, and the fault type identification results and key parameters are used as inputs to select the best model structure through cross-validation. Finally, the particle swarm algorithm is used to optimize the parameters of the neural network model to improve the accuracy of location estimation. The advantages of two traditional location methods, the traveling wave method and the impedance method, are combined. First, the two methods are used to locate the fault respectively, and then a Bayesian network model is constructed to preliminarily integrate the results of the two methods. Using the fuzzy reasoning method, various uncertain factors are considered to obtain a fuzzy location interval. Subsequently, a genetic algorithm is used to search for the optimal solution in this interval to obtain a series of candidate location points. These candidate points are evaluated by experts through the Delphi method and weight coefficients are given. The candidate locations are weighted averaged according to the weight coefficients, and the Kalman filter is used to further optimize the location estimation. Finally, the positioning error is analyzed by the Monte Carlo simulation method, the confidence interval is calculated, and the final accurate fault location coordinates are obtained. Convert the coordinates of the fault location to geographic coordinates and locate them in the geographic information system. Analyze the surrounding terrain features based on the positioning results and plan the optimal inspection route. Conduct field surveys along the planned route and collect field data and images. Use computer vision algorithms to analyze the fault feature images and verify the accuracy of the fault location. If there is a deviation between the actual fault location and the calculated result, update the fault location information and analyze the cause of the deviation. According to the analysis results, adjust the parameters of the positioning algorithm to optimize system performance.Finally, the corrected fault location and optimized parameters are integrated to form the final fault location results and system optimization data.

[0015] By executing the above steps, the electrical signals at both ends of the transmission line are collected and preprocessed to obtain the preprocessed synchronous electrical signals, which effectively eliminates the noise and interference in the original signal and improves the data quality of subsequent analysis. The preprocessed synchronous electrical signals are adaptively enhanced to obtain the enhanced fault feature signals, which highlight the fault characteristics, suppress the background noise, and greatly improve the recognition rate of the fault characteristics. Subsequently, the enhanced fault feature signals are subjected to time-frequency analysis and feature extraction to obtain the fault feature vectors and key parameters, which fully capture the time domain and frequency domain characteristics of the fault signal and lay the foundation for accurate positioning. The fault feature vectors and key parameters are subjected to machine learning analysis to obtain the fault type identification results and preliminary fault location estimation. This innovative application of machine learning technology realizes the automatic identification and preliminary positioning of the fault type, greatly improving the efficiency and accuracy of positioning. The preliminary fault location estimation is subjected to multi-algorithm fusion analysis to obtain the precise fault location coordinates, which combines the advantages of multiple positioning algorithms and significantly improves the positioning accuracy. Finally, the precise fault location coordinates are mapped with geographic information and verified on the spot to obtain the fault location results and system optimization data, which not only verifies the accuracy of the positioning results, but also continuously optimizes the system performance through the feedback mechanism. Overall, this method greatly improves the accuracy, efficiency and reliability of fault location through the organic combination of technologies such as signal enhancement, feature extraction, machine learning and multi-algorithm fusion.

[0016] In a specific embodiment, the process of executing step S101 may specifically include the following steps: (1) Perform high-frequency sampling on the voltage and current signals at both ends of the transmission line to obtain raw electrical signal data, and perform time stamping on the raw electrical signal data to obtain raw signal data with time stamp; (2) Performing wavelet denoising on the original signal data with timestamp to obtain a preliminary denoised signal, and performing adaptive median filtering on the preliminary denoised signal to obtain a filtered signal; (3) Perform outlier detection on the filtered signal to obtain signal data without outliers, and perform interpolation processing on the signal data without outliers to obtain continuous signal data; (4) Perform Fourier transform on the continuous signal data to obtain frequency domain feature data, and perform bandpass filtering on the frequency domain feature data to obtain the target frequency band signal; (5) Performing phase correction on the target frequency band signal through Hilbert transform to obtain a phase-corrected signal, and performing time synchronization processing on the phase-corrected signal to obtain an electrical signal synchronized at both ends; (6) Data compression is performed on the electrical signals synchronized at both ends to obtain a compressed synchronization signal, and the compressed synchronization signal is quantized and encoded to obtain a preprocessed synchronization electrical signal.

[0017] Specifically, a high-precision digital sampling device is used, and the sampling frequency is usually set between 10kHz and 100kHz to ensure that transient fault characteristics can be captured. The sampled raw electrical signal data is then timestamped, which is a critical step because it lays the foundation for subsequent signal synchronization processing. The accuracy of the timestamp is usually at the microsecond level to ensure accurate alignment of the signals at both ends. Next, the raw signal data with timestamps is subjected to wavelet denoising. Wavelet denoising is an effective signal processing technique that can remove background noise while retaining important signal features. The specific operation includes selecting a suitable wavelet basis function, performing wavelet decomposition on the signal, setting a threshold to remove the noise coefficient, and then reconstructing the signal. After obtaining the preliminary denoised signal, adaptive median filtering is performed. Adaptive median filtering can effectively remove pulse interference while protecting the edge information of the signal. The size of the filter window is dynamically adjusted according to the local signal characteristics to achieve the best filtering effect.

[0018] The filtered signal also needs to be detected for outliers. Outlier detection aims to identify and remove abnormal data points, which may be caused by equipment failures or transient interference. Common detection methods include statistical-based methods (such as the 3σ criterion) or density-based methods (such as the DBSCAN algorithm). After detecting and removing outliers, the data is interpolated to ensure the continuity of the signal. The interpolation method can be linear interpolation, spline interpolation, or more complex algorithms, depending on the characteristics of the signal and the accuracy requirements. The Fourier transform of continuous signal data is to obtain the frequency domain characteristics of the signal. Fast Fourier transform (FFT) is often used for this purpose, which can efficiently convert time domain signals to frequency domain. In the frequency domain, different types of faults will show different characteristic frequencies. After obtaining the frequency domain characteristic data, a bandpass filter is applied to retain the target frequency band signal. The parameter setting of the bandpass filter is based on the prior knowledge of the fault characteristic frequency. Usually, the frequency band containing fault information (such as power frequency and its harmonics, fault transient frequency, etc.) is retained, while filtering out high-frequency noise and low-frequency interference.

[0019] Hilbert transform is used to perform phase correction on the target frequency band signal. Hilbert transform can obtain the analytical representation of the signal, thereby separating the amplitude and phase information of the signal. The purpose of phase correction is to eliminate the phase deviation caused by factors such as transmission line characteristics and measurement equipment, and ensure the phase consistency of the signals at both ends. After phase correction, the signal is time synchronized. Time synchronization is the key to the two-end positioning method. It aligns the signal by comparing the timestamps of the signals at both ends and combining line parameters (such as line length, propagation speed, etc.). Finally, the electrical signals synchronized at both ends are compressed and quantized. Data compression aims to reduce the amount of data and improve transmission and storage efficiency. Common compression methods include wavelet compression and compressed sensing. The compressed signal is quantized and encoded to map the continuous signal value to a discrete digital code. The quantization process requires a trade-off between data accuracy and storage space. Non-uniform quantization methods are usually used to reduce the amount of data while ensuring the accuracy of key information.

[0020] For example, a single-phase grounding fault occurred in a 330kV transmission line, and the fault location devices installed at both ends of the line were started simultaneously. The sampling frequency was set to 20kHz, and the voltage and current data were collected for 1 second to obtain 20,000 sampling points. Each sampling point was attached with a timestamp accurate to microseconds. The original data was decomposed and denoised using the db4 wavelet with a 5-layer decomposition. The noise threshold was set to 0.05, which removed about 90% of the background noise. The adaptive median filter window size was dynamically adjusted between 3 and 7, and 10 pulse interferences caused by switch operations were successfully filtered out. The 3σ criterion was used for outlier detection, and 5 abnormal data points were identified and removed. Then, the data gaps were filled using the cubic spline interpolation method. A 2048-point FFT was performed on the 19,995 processed data points to obtain a 0-10kHz spectrum. A bandpass filter was designed with a passband of 45-55Hz and 400-600Hz, which retained the power frequency signal and the transient characteristics of the fault. After the Hilbert transform, it was found that there was a 2-degree phase difference between the signals at both ends, which was corrected. Finally, the wavelet compression algorithm is used to compress the data to 30% of the original amount, and 8-bit non-uniform quantization is used for encoding. The entire preprocessing process takes about 50 milliseconds.

[0021] In a specific embodiment, the process of executing step S102 may specifically include the following steps: (1) Perform wavelet packet decomposition on the preprocessed synchronous electrical signal to obtain multi-scale signal components, and perform energy distribution analysis on the multi-scale signal components to obtain energy eigenvectors; (2) Adaptively select the threshold value of the multi-scale signal components according to the energy eigenvector to obtain candidate feature components, and sparsely represent the candidate feature components to obtain a sparse coefficient matrix; (3) Perform singular value decomposition on the sparse coefficient matrix to obtain the main characteristic components, and perform adaptive gain adjustment on the main characteristic components to obtain a gain-adjusted signal; (4) Decomposing the gain-adjusted signal by empirical mode decomposition to obtain an intrinsic mode function set, and performing correlation analysis on the intrinsic mode function set to obtain the key mode function; (5) Designing an adaptive filter for the key modal function to obtain optimized filter parameters, and filtering the key modal function by optimizing the filter parameters to obtain a filtered modal function; (6) Reconstruct the filtered modal function to obtain the enhanced fault characteristic signal.

[0022] Specifically, wavelet packet decomposition is an extension of wavelet transform, which also subdivides the high-frequency part of the signal and provides a finer frequency division. In specific implementation, the appropriate wavelet basis function (such as Daubechies wavelet) and the number of decomposition layers are selected to decompose the signal into sub-signals of different frequency bands. The energy distribution of these multi-scale signal components is analyzed, and the energy value of each component is calculated to form an energy feature vector. The energy feature vector reflects the energy distribution of the signal in different frequency bands and helps to identify fault characteristics. According to the energy feature vector, the adaptive threshold selection method is used to screen out candidate feature components containing fault information. The setting of the adaptive threshold takes into account the overall energy distribution and local characteristics of the signal, and is usually determined by methods based on statistical characteristics or information entropy. The selected candidate feature components are then sparsely represented, which is a method of representing the signal as a linear combination of a small number of basic elements. Sparse representation solves the optimization problem and finds a small number of atoms that best represent the signal characteristics in a predefined dictionary to obtain a sparse coefficient matrix. Sparse representation helps to highlight the essential characteristics of the signal and suppress noise and interference.

[0023] Singular value decomposition of sparse coefficient matrix is ​​an effective method to extract main characteristic components. Singular value decomposition decomposes the matrix into the product of three matrices, where the singular values ​​reflect the importance of the features. The main characteristic components are reconstructed by retaining the larger singular values ​​and their corresponding singular vectors. Adaptive gain adjustment is performed on these main characteristic components to highlight the fault characteristics and suppress background noise. The amplitude of gain adjustment is dynamically adjusted according to the local characteristics of the signal to avoid over-amplification of noise or suppression of useful information. Empirical mode decomposition is an adaptive signal processing method, which is particularly suitable for the analysis of nonlinear and non-stationary signals. It decomposes the signal into a series of intrinsic mode functions (IMFs), each of which represents an inherent oscillation mode of the signal. Empirical mode decomposition is performed on the gain-adjusted signal to obtain a set of IMFs. Subsequently, these IMFs are subjected to correlation analysis, the correlation coefficient of each IMF with the original signal is calculated, and the IMFs with higher correlation are selected as key mode functions. These key mode functions usually contain the most important fault characteristic information.

[0024] Designing an adaptive filter for the key modal function is an important step to further optimize the signal. The adaptive filter can automatically adjust its parameters according to the characteristics of the input signal to achieve the best filtering effect. Commonly used adaptive filtering algorithms include the least mean square error (LMS) algorithm and the recursive least squares (RLS) algorithm. Through these algorithms, the optimal filter parameters are determined, and then these parameters are used to filter the key modal function to further improve the signal-to-noise ratio and obtain the filtered modal function. Finally, the filtered modal function is reconstructed to obtain the enhanced fault feature signal. The reconstruction process usually directly adds the selected filtered modal functions or uses a weighted summation method, and the weights can be determined based on the energy or correlation of each modal function. The reconstructed signal retains the key fault features in the original signal while greatly reducing the impact of noise and interference.

[0025] For example, a single-phase grounding fault occurs in a 500kV transmission line, and the length of the pre-processed synchronous electrical signal collected is 10,000 points. First, the signal is decomposed into 5 layers of wavelet packets using the db4 wavelet basis to obtain signal components in 32 frequency bands. The energy of each component is calculated to form a 32-dimensional energy feature vector. The energy threshold is set to 1% of the total energy, and 10 candidate feature components with energy exceeding the threshold are selected. These 10 components are sparsely represented, and a sparse coefficient matrix is ​​obtained by using an overcomplete dictionary of 100 atoms. The matrix is ​​subjected to singular value decomposition, and the feature components corresponding to the first 5 largest singular values ​​are retained. Adaptive gain adjustment adjusts the peak values ​​of these 5 components to 80% of the original signal peak value. Eight IMFs are obtained by empirical mode decomposition, and three key IMFs with correlation coefficients greater than 0.6 are selected through correlation analysis. A 50-order adaptive FIR filter is designed using the LMS algorithm to filter these three IMFs. Finally, the three filtered IMFs are weighted and reconstructed, and the weights are determined according to the energy ratio of each IMF. The entire signal enhancement process takes about 100 milliseconds. The fault characteristics in the enhanced signal are more obvious, and the signal-to-noise ratio is improved by about 15dB, laying a solid foundation for subsequent fault location analysis.

[0026] In a specific embodiment, the process of executing step S103 may specifically include the following steps: (1) Perform short-time Fourier transform on the enhanced fault characteristic signal to obtain a time-frequency spectrum, and perform peak detection on the time-frequency spectrum to obtain a characteristic frequency point set; (2) Segment the time-frequency spectrum according to the characteristic frequency point set to obtain the time-frequency characteristic region, and calculate the energy density of the time-frequency characteristic region to obtain the energy distribution characteristics; (3) Perform principal component analysis on the energy distribution characteristics to obtain the main characteristic components, and standardize the main characteristic components to obtain the normalized characteristic vector; (4) Perform multi-scale decomposition on the enhanced fault feature signal through wavelet packet transform to obtain multi-scale coefficients, and extract statistical features from the multi-scale coefficients to obtain a statistical feature set; (5) Perform correlation analysis on the statistical feature set to obtain key statistical features, and combine the key statistical features to obtain a statistical feature vector; (6) The normalized feature vector and the statistical feature vector are fused to obtain the fault feature vector. The fault occurrence time and duration are calculated based on the fault feature vector to obtain the fault feature vector and key parameters.

[0027] Specifically, the short-time Fourier transform divides the signal into small time windows, and performs a Fourier transform on each window to obtain the frequency information of the signal changing over time. Generate a time-frequency spectrum to intuitively display the frequency components of the signal at different time points. Perform peak detection on the time-frequency spectrum to identify the characteristic frequency points with concentrated energy, which usually correspond to fault features. Peak detection uses a local maximum search algorithm, sets an appropriate threshold to filter out weak peaks, and obtains a characteristic frequency point set. The time-frequency spectrum is segmented according to the characteristic frequency point set, and the time-frequency plane is divided into multiple characteristic regions. The segmentation method can use a threshold-based segmentation or clustering algorithm, such as K-means clustering. The segmented time-frequency feature region reflects the distribution characteristics of the fault signal in the time-frequency domain. Calculate the energy density for each time-frequency feature region to obtain the energy distribution characteristics. The energy density calculation takes into account the ratio of the sum of the squares of the signal amplitudes in the region to the area of ​​the region, reflecting the concentration of the fault energy on the time-frequency plane.

[0028] Principal component analysis of energy distribution characteristics is an effective method for dimensionality reduction and extraction of main features. Principal component analysis converts possibly related variables into linearly independent variables, namely principal components, through orthogonal transformation. The first few principal components with larger contribution rates are selected as the main feature components, which not only retains the main information of the original data but also reduces the data dimension. The main feature components are standardized to convert features of different dimensions into a unified scale to obtain normalized feature vectors. The standardization process usually uses the z-score method to ensure that different features have the same weight in subsequent analysis. The enhanced fault feature signal is decomposed into multiple scales by wavelet packet transform, providing a detailed representation of the signal in the time and frequency domains. Wavelet packet transform is an extension of wavelet transform, which also subdivides the high-frequency part of the signal to provide a finer frequency division. The coefficients obtained by multi-scale decomposition reflect the characteristics of the signal at different scales and frequency bands. Statistical feature extraction is performed on these multi-scale coefficients, including calculating statistics such as mean, variance, skewness, and kurtosis to obtain a statistical feature set. These statistical features capture the distribution characteristics of the signal at different scales.

[0029] Perform correlation analysis on the statistical feature set, filter out features that are highly correlated with the fault type and location, and obtain key statistical features. Correlation analysis can use methods such as Pearson correlation coefficient or mutual information. Select features with high correlation and low correlation with each other to avoid information redundancy. Combine the selected key statistical features to form a statistical feature vector. The combination method can be simple splicing or weighted summation, and the weight can be determined according to the importance of each feature. Fuse the normalized feature vector and the statistical feature vector to obtain the final fault feature vector. The fusion method can use feature-level fusion or decision-level fusion. Feature-level fusion directly splices two vectors, while decision-level fusion classifies the two vectors separately and then integrates the results. The fused feature vector contains comprehensive information in the time-frequency domain and the statistical domain. According to the fused fault feature vector, by setting a threshold or using a machine learning model, the time and duration of the fault can be accurately determined. These are key parameters that are crucial for fault location.

[0030] For example, a two-phase short circuit fault occurred in a 220kV transmission line. The length of the enhanced fault feature signal collected was 20,000 points and the sampling frequency was 10kHz. First, a short-time Fourier transform was performed using a Hanning window with a window length of 512 points and an overlap rate of 50% to obtain a time-frequency spectrum. Peak detection was performed on the time-frequency spectrum, and the energy threshold was set to 3 times the average energy. 15 characteristic frequency points were detected. Based on these points, the K-means algorithm (K=5) was used to segment the time-frequency spectrum. The energy density of each region was calculated to obtain a 5-dimensional energy distribution feature. The energy distribution feature was subjected to principal component analysis, and the first three principal components with a cumulative contribution rate of 95% were selected as the main feature components. The z-score was normalized to obtain a normalized feature vector. At the same time, the signal was decomposed into a 4-layer wavelet packet using the db4 wavelet basis to obtain the coefficients of 16 frequency bands. The mean, variance, skewness, and kurtosis were calculated for each frequency band to form a 64-dimensional statistical feature set. By calculating the Pearson correlation coefficient, 10 features with an absolute value of correlation coefficient greater than 0.7 were selected as key statistical features. The normalized feature vector (3D) and the key statistical features (10D) were directly concatenated to obtain a 13-dimensional final fault feature vector. According to the changes in energy mutation and frequency characteristics in the feature vector, the fault was judged to occur at 8732 sampling points (corresponding to the actual time of 0.8732 seconds) and lasted for 0.15 seconds. The entire feature extraction and analysis process took about 200 milliseconds, laying the foundation for subsequent accurate fault location.

[0031] In a specific embodiment, the process of executing step S104 may specifically include the following steps: (1) Perform dimensionality reduction processing on the fault feature vector to obtain the reduced-dimensional feature data, and perform cluster analysis on the reduced-dimensional feature data to obtain feature clusters; (2) Construct a decision tree model based on the feature cluster to obtain preliminary classification rules, and then prune and optimize the preliminary classification rules to obtain optimized decision rules; (3) The optimized decision rules are trained by the support vector machine algorithm to obtain a fault type classifier, and the fault type classifier is used to classify the new fault feature vector to obtain the fault type identification result; (4) Construct a neural network model based on the fault type identification results and key parameters to obtain a position estimation model, and cross-validate the position estimation model to obtain a candidate position estimation model; (5) The particle swarm algorithm is used to optimize the parameters of the candidate location estimation model to obtain the optimized location estimation model. The fault feature vector is processed through the optimized location estimation model to obtain a preliminary fault location estimate.

[0032] Specifically, commonly used dimensionality reduction methods include principal component analysis (PCA) and linear discriminant analysis (LDA). PCA transforms possibly related variables into linearly independent variables through orthogonal transformation, retaining the principal components with larger variance contribution rate. LDA considers the category information and seeks the projection direction that maximizes the variance between classes and minimizes the variance within classes. The feature data after dimensionality reduction retains the main information of the original data while greatly reducing the data dimension. Cluster analysis is performed on the feature data after dimensionality reduction, usually using K-means or hierarchical clustering algorithm to classify similar fault features into one category to form feature clusters. Cluster analysis helps to discover the inherent structure in the data and provides a basis for subsequent fault type identification. Constructing a decision tree model based on feature clusters is an intuitive and efficient classification method. The decision tree classifies data through a series of if-then rules, each internal node represents a feature test, and each leaf node represents a category. Common decision tree algorithms include ID3, C4.5, and CART. During the construction process, the algorithm selects the best split feature based on information gain or Gini coefficient and recursively divides the data set. The initial classification rules obtained may be too complex and there is a risk of overfitting. Therefore, pruning optimization is required, including pre-pruning and post-pruning. Pre-pruning restricts the tree during growth, while post-pruning simplifies the fully grown tree. The goal of pruning is to reduce the complexity of the tree and improve the generalization ability of the model while maintaining the classification accuracy.

[0033] A more robust fault type classifier can be obtained by training the optimized decision rules through the support vector machine (SVM) algorithm. The core idea of ​​SVM is to construct an optimal separation hyperplane in high-dimensional space so that the sample interval of different categories is maximized. For nonlinear separable cases, SVM maps data to high-dimensional space through kernel techniques. Common kernel functions include linear kernel, polynomial kernel and radial basis function (RBF) kernel. The training process of SVM involves solving quadratic programming problems to obtain support vectors and decision boundaries. After training, the new fault feature vectors are classified using the fault type classifier to obtain the fault type identification results. A neural network model is constructed based on the fault type identification results and key parameters for fault location estimation. The neural network model has powerful nonlinear mapping capabilities and is suitable for handling complex fault location estimation problems. Common neural network structures include multi-layer perceptron (MLP) and long short-term memory network (LSTM). During the construction process, it is necessary to determine the network structure (such as the number of layers, the number of neurons in each layer), activation function (such as ReLU, sigmoid) and loss function (such as mean square error). The network is trained using the back propagation algorithm and the network parameters are optimized using the gradient descent method. In order to evaluate the generalization ability of the model, the location estimation model is cross-validated. A common method is K-fold cross-validation, which divides the data set into K parts, and uses K-1 parts as training sets and 1 part as validation sets in turn. Through cross-validation, multiple candidate location estimation models can be obtained.

[0034] The particle swarm algorithm (PSO) is used to optimize the parameters of the candidate position estimation model to further improve the model performance. PSO is a swarm intelligence optimization algorithm that searches for the optimal solution by simulating the foraging behavior of bird flocks. In this scheme, PSO is used to optimize the hyperparameters of the neural network, such as the learning rate, the number of hidden layers, the number of neurons in each layer, etc. Each particle represents a set of hyperparameter combinations. By iteratively updating the position and velocity of the particles, the parameter combination that optimizes the model performance is found. During the optimization process, an appropriate fitness function (such as the mean square error on the validation set) is defined to evaluate the quality of the parameter combination. Finally, the optimized position estimation model is obtained, which has better generalization ability and prediction accuracy.

[0035] For example, a single-phase grounding fault occurs in a 330kV transmission line, and the fault feature vector collected is 50-dimensional. First, PCA is used for dimensionality reduction, retaining the principal components with a cumulative variance contribution rate of 95%, and reducing the feature vector to 15 dimensions. The K-means algorithm is used to cluster the reduced-dimensional data, setting K=5 to obtain 5 feature clusters. Based on these feature clusters, the CART algorithm is used to build a decision tree with an initial tree depth of 10. The optimal pruning parameters are determined by cross-validation, and the tree depth is reduced to 6 to obtain the optimized decision rule. The decision rule is trained using the RBF kernel SVM, with kernel parameter γ=0.1 and penalty parameter C=10, achieving a classification accuracy of 98% on the validation set. According to the SVM classification results and key parameters such as the time of fault occurrence and duration, a 3-layer MLP neural network is constructed, with 20 and 10 hidden layer neurons respectively, and the ReLU activation function is used. Five-fold cross-validation is used to obtain 5 candidate location estimation models. PSO was used to optimize the neural network hyperparameters, with the number of particles set to 30, the number of iterations to 100, and the search range including the learning rate (0.001-0.1) and the number of hidden layer neurons (10-50). The average absolute error of the optimized best model on the test set was 0.5% of the total line length. The new fault feature vector was input into the optimized model, and the fault location was estimated to be 68.3 km from the starting point of the line. The entire fault type identification and location estimation process took about 300 milliseconds, providing strong support for fast and accurate fault location.

[0036] In a specific embodiment, the process of executing step S105 may specifically include the following steps: (1) Performing a traveling wave method analysis on the preliminary fault location estimate to obtain a traveling wave method location estimate, and performing an impedance method analysis on the preliminary fault location estimate to obtain an impedance method location estimate; (2) A Bayesian network model is constructed based on the position estimation using the traveling wave method and the position estimation using the impedance method to obtain a preliminary fusion result, and fuzzy reasoning is performed on the preliminary fusion result to obtain a fuzzy position interval; (3) Optimize the fuzzy position interval using a genetic algorithm to obtain a candidate position point set, and then evaluate the candidate position point set using the Delphi method to obtain a weight coefficient; (4) Perform weighted averaging on the candidate position point set according to the weight coefficient to obtain a weighted position estimate, and perform Kalman filtering on the weighted position estimate to obtain a filtered position estimate; (5) The error analysis of the filtered position estimation is performed through Monte Carlo simulation to obtain the error distribution. The confidence interval is calculated based on the error distribution to obtain the precise fault location coordinates.

[0037] Specifically, in the two-terminal fault location method based on signal enhancement, the analysis of the preliminary fault location estimation by the traveling wave method and the impedance method is a key step to improve the positioning accuracy. The traveling wave method uses the propagation characteristics of the high-frequency electromagnetic waves generated by the fault on the line for positioning. In the specific implementation, the high-frequency signals collected at both ends are firstly subjected to wavelet transformation to identify the arrival time of the initial traveling wave of the fault. Then, the distance from the fault point to both ends of the line is calculated based on the traveling wave propagation speed and the arrival time difference of the traveling waves at both ends. The impedance law locates based on the relationship between the impedance of the fault point and the impedance per unit length of the line. During the implementation process, the equivalent impedance of the fault point is calculated by measuring the voltage and current at the time of the fault, and then the fault distance is calculated in combination with the line parameters. A Bayesian network model is constructed based on the location estimation results of the traveling wave method and the impedance method, realizing the probabilistic fusion of the two methods. The Bayesian network is an inference method based on a probabilistic graph model that can handle uncertainty and complex dependencies. In this scheme, the nodes of the Bayesian network include the results of the traveling wave method, the results of the impedance method and the actual fault location, and the relationship between the nodes is described by the conditional probability table. The network parameters are trained using historical data to obtain preliminary fusion results. Subsequently, fuzzy reasoning is performed on the preliminary fusion results to introduce expert knowledge and empirical rules. The fuzzy reasoning process includes three steps: fuzzification, fuzzy rule reasoning, and defuzzification, and finally a fuzzy position interval is obtained. The genetic algorithm optimization of the fuzzy position interval is to find the optimal set of candidate position points within the fuzzy interval. The genetic algorithm simulates the biological evolution process and continuously optimizes the solution population through selection, crossover, and mutation operations. In this scheme, each chromosome represents a possible fault location, and the fitness function is based on the error design of the traveling wave method and the impedance method. Through multiple generations of evolution, a set of high-quality candidate position points is obtained. These candidate position points are evaluated by the Delphi method to integrate the opinions of multiple experts. The Delphi method is a structured group communication technology that gradually forms an expert consensus through multiple rounds of anonymous questionnaires. Each expert scores the candidate position based on his or her own experience and the data provided. After multiple rounds of iterations, the weight coefficient of each candidate position is finally obtained.

[0038] The candidate location point set is weighted averaged according to the weight coefficient to obtain a comprehensive location estimate. The credibility of each candidate location is fully considered to make the final result more reliable. The weighted location estimate is Kalman filtered to further improve the positioning accuracy. Kalman filtering is a recursive state estimation algorithm that can optimally estimate the system state in the presence of measurement noise. In this scheme, Kalman filtering uses historical data and current measurements to continuously update the estimate of the fault location to obtain the filtered location estimate. The error analysis of the filtered location estimate through Monte Carlo simulation can comprehensively evaluate the uncertainty of the positioning result. The Monte Carlo method simulates the behavior of complex systems through a large number of random sampling. In this scheme, a large number of samples are generated according to the statistical characteristics of the measurement error and model parameters, and the fault location of each sample is calculated. By analyzing these simulation results, the error distribution of the fault location estimate is obtained. Based on the error distribution, the confidence interval is calculated to give a statistically significant range of the fault location. This confidence interval not only provides accurate fault location coordinates, but also quantifies the reliability of the positioning result.

[0039] For example: a 500kV transmission line is 300km long and has a single-phase grounding fault. The initial fault location is estimated to be 180km from the starting point. The traveling wave method is used to analyze the location. By performing wavelet transform on the high-frequency signals at both ends, the arrival times of the initial traveling waves of the fault are identified as t1=0.6ms and t2=0.8ms. Considering the traveling wave propagation speed v=290m / μs, the traveling wave method location is estimated to be 175.8km. At the same time, using the impedance method, the equivalent impedance of the fault point is measured to be 45+j60Ω. Combined with the line parameters (impedance per kilometer is 0.25+j0.32Ω / km), the impedance method location is estimated to be 183.2km.

[0040] A Bayesian network model was constructed to integrate the results of the traveling wave method and the impedance method. After training, the network gave a preliminary fusion result of 179.5 km. Fuzzy reasoning was performed on this result, considering the line parameter error and measurement error, and the fuzzy position interval was obtained as [177 km, 182 km]. The genetic algorithm was used to optimize within this interval, with the population size set to 100 and the evolutionary generation number set to 50, and 10 candidate position points were obtained. Through the Delphi method, 5 experts conducted 3 rounds of evaluation on these 10 points and finally obtained the weight coefficient. The weighted average was performed according to the weight coefficient, and the weighted position estimate was obtained as 179.8 km. Kalman filtering was performed on this estimate, considering the historical fault data and the current measurement value, and the filtered position estimate was obtained as 179.6 km. Finally, 10,000 Monte Carlo simulations were performed to generate the error distribution. The analysis results show that the 95% confidence interval is [179.2 km, 180.0 km]. Therefore, the final precise fault location coordinates were determined to be 179.6 km, and the confidence interval was ±0.4 km. The entire multi-algorithm fusion analysis process takes about 500ms, significantly improving the accuracy and reliability of fault location.

[0041] In a specific embodiment, the process of executing step S106 may specifically include the following steps: (1) Perform geocoding conversion on the precise fault location coordinates to obtain geographic coordinates, and locate them in the geographic information system based on the geographic coordinates to obtain the geographical location of the fault point; (2) Perform terrain analysis on the geographical location of the fault point to obtain terrain feature data, and perform path planning based on the terrain feature data to obtain the target inspection route; (3) Conduct field surveys along the target inspection route to obtain field data, and perform image processing on the field data to obtain fault feature images; (4) Analyze the fault feature image through computer vision algorithm to obtain the fault verification result, and update the fault location information according to the fault verification result to obtain the corrected fault location; (5) Analyze the deviation between the corrected fault location and the initial accurate fault location coordinates to obtain positioning error data, and adjust the parameters of the positioning algorithm based on the positioning error data to obtain optimized parameters; (6) Integrate the optimized parameters and the corrected fault location to obtain the fault location results and system optimization data.

[0042] Specifically, a special coordinate conversion algorithm is usually used to convert the distance points on the line into longitude and latitude coordinates. The earth curvature and projection method need to be considered during the conversion to ensure the accuracy of the conversion. After obtaining the geographic coordinates, the location is performed in the geographic information system (GIS), and the abstract coordinate points are mapped to the actual geographical environment to obtain the specific geographical location of the fault point, including the terrain and surrounding environment. Topographic analysis of the geographical location of the fault point is an important part of preparing for field investigation. Topographic analysis usually includes analysis of factors such as slope, aspect, elevation, and vegetation coverage. These analyses are performed using digital elevation models (DEM) and remote sensing image data, and terrain feature data is extracted through GIS spatial analysis tools. Based on these terrain feature data and combined with the actual direction of the transmission line, the shortest path algorithm (such as Dijkstra algorithm or A* algorithm) is used for path planning. Factors such as terrain difficulty and traffic accessibility need to be considered during planning, and ultimately an optimal target inspection route is obtained. Field investigation along the target inspection route is a direct means of verifying the fault location. During the survey, high-precision GPS equipment is used to record the route, and multimedia data such as on-site images and videos are collected at the same time. The collected on-site data is processed, including image enhancement, noise removal, feature extraction and other steps. Image enhancement can use methods such as histogram equalization or adaptive contrast adjustment to improve image quality. Noise removal can use technologies such as median filtering or wavelet transform. Feature extraction is the key to identifying fault features. Edge detection, texture analysis and other methods can be used to finally obtain an image that highlights the fault features. Analyzing fault feature images through computer vision algorithms is the core step of automatic fault identification. It may involve multiple algorithms, such as deep learning-based target detection algorithms (such as YOLO or Faster R-CNN) for locating faulty components, and convolutional neural networks (CNN) for fault type classification. These algorithms require a large amount of labeled data for training to identify various possible fault features. The analysis results give fault verification results, including information such as fault type and severity. Based on these results, combined with GPS positioning data, the fault location information is updated to obtain the corrected fault location.

[0043] The deviation between the corrected fault location and the initial accurate fault location coordinates is analyzed to calculate the distance difference and direction offset between the two. This analysis obtains the positioning error data, which reflects the deviation between the algorithm positioning result and the actual situation. According to these error data, the positioning algorithm parameters are adjusted. The adjustment process may involve multiple aspects, such as the correction of the traveling wave velocity, the calibration of the impedance parameters, and the update of the Bayesian network weights. These adjustments are aimed at reducing the error of future positioning and improving the accuracy of the algorithm. Finally, a set of optimized parameters are obtained, which will be used for subsequent fault location work. The optimized parameters and the corrected fault location are integrated to form the final fault location results and system optimization data. These data include not only the accurate fault location information, but also the relevant parameters of the algorithm optimization. These results will be used to update the database of the fault location system to further improve the positioning accuracy and reliability of the system.

[0044] For example, a single-phase grounding fault occurred on a 330kV transmission line, and the initial precise fault location coordinates were 150.5km from the starting point. The coordinate conversion algorithm was used to convert this location into longitude and latitude coordinates (120.5678° east longitude, 30.1234° north latitude). After positioning in GIS, it was found that the point was located in a hilly area. Terrain analysis showed that the average slope within 5km around the fault point was 15°, the highest altitude was 500m, and the vegetation coverage rate was 60%. Based on these data, the A* algorithm was used to plan a 12.3km long inspection route, avoiding steep areas and dense vegetation. Field surveys were conducted along the planned route, and 200 high-definition images and 10 minutes of 4K video were collected. The data were preprocessed, including using histogram equalization to enhance image contrast and removing noise through Gaussian filtering. The Canny edge detection algorithm was then used to extract the line and tower contours in the image. The processed image was analyzed using the YOLOv5 target detection algorithm, and a damaged insulator was identified, which was located 151.2 km away from the actual starting point.

[0045] Comparing the corrected fault location (151.2km) with the initial location (150.5km), a positioning error of 0.7km was calculated. Based on this error, the traveling wave velocity was corrected from the original 298m / μs to 297.3m / μs. At the same time, the line parameters in the impedance method were updated, and the resistance per unit length was corrected from 0.028Ω / km to 0.0282Ω / km. These optimized parameters were integrated into the system and are expected to reduce the positioning error by about 20% in similar situations in the future. The final fault location result was determined to be a broken insulator 151.2km away from the starting point, with a positioning error of less than 100m. The entire verification and optimization process took about 4 hours, significantly improving the accuracy of fault location and the reliability of the system.

[0046] The above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the embodiments, a person skilled in the art should understand that the specific implementation modes of the present invention can still be modified or replaced by equivalents, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A two-terminal fault location method based on signal enhancement, characterized in that: include: Collect and preprocess the electrical signals at both ends of the transmission line to obtain preprocessed synchronous electrical signals; Adaptively enhance the pre-processed synchronous electrical signal to obtain an enhanced fault characteristic signal; Perform time-frequency analysis and feature extraction on the enhanced fault feature signal to obtain the fault feature vector and key parameters; Perform machine learning analysis on fault feature vectors and key parameters to obtain fault type identification results and preliminary fault location estimation; Perform multi-algorithm fusion analysis on the preliminary fault location estimation to obtain the precise fault location coordinates; The precise fault location coordinates are mapped and verified on the spot to obtain fault location results and system optimization data.

2. The double-terminal fault location method based on signal enhancement according to claim 1, characterized in that: The step of collecting and preprocessing the electrical signals at both ends of the transmission line to obtain the preprocessed synchronous electrical signals includes: Performing high-frequency sampling on voltage and current signals at both ends of the transmission line to obtain original electrical signal data, and performing time stamp marking on the original electrical signal data to obtain original signal data with time stamp; Performing wavelet denoising on the original signal data with the timestamp to obtain a preliminary denoised signal, and performing adaptive median filtering on the preliminary denoised signal to obtain a filtered signal; Performing outlier detection on the filtered signal to obtain signal data without outliers, and performing interpolation processing on the signal data without outliers to obtain continuous signal data; Performing Fourier transform on the continuous signal data to obtain frequency domain feature data, and performing bandpass filtering on the frequency domain feature data to obtain a target frequency band signal; Performing phase correction on the target frequency band signal by Hilbert transform to obtain a phase-corrected signal, and performing time synchronization processing on the phase-corrected signal to obtain an electrical signal synchronized at both ends; The electrical signals synchronized at both ends are data compressed to obtain a compressed synchronization signal, and the compressed synchronization signal is quantized and encoded to obtain a pre-processed synchronization electrical signal.

3. The double-terminal fault location method based on signal enhancement according to claim 1, characterized in that: The step of performing adaptive enhancement processing on the pre-processed synchronous electrical signal to obtain an enhanced fault characteristic signal comprises: Performing wavelet packet decomposition on the preprocessed synchronous electrical signal to obtain multi-scale signal components, and performing energy distribution analysis on the multi-scale signal components to obtain energy eigenvectors; Adaptively selecting a threshold value for a multi-scale signal component according to the energy eigenvector to obtain a candidate eigencomponent, and sparsely representing the candidate eigencomponent to obtain a sparse coefficient matrix; Performing singular value decomposition on the sparse coefficient matrix to obtain main characteristic components, and performing adaptive gain adjustment on the main characteristic components to obtain a gain-adjusted signal; Decomposing the gain-adjusted signal by empirical mode decomposition to obtain an intrinsic mode function set, and performing correlation analysis on the intrinsic mode function set to obtain a key mode function; Performing adaptive filter design on the key modal function to obtain optimized filter parameters, and filtering the key modal function using the optimized filter parameters to obtain a filtered modal function; The filtered modal function is reconstructed to obtain an enhanced fault characteristic signal.

4. The double-terminal fault location method based on signal enhancement according to claim 1, characterized in that: The step of performing time-frequency analysis and feature extraction on the enhanced fault feature signal to obtain the fault feature vector and key parameters includes: Performing short-time Fourier transform on the enhanced fault characteristic signal to obtain a time-frequency spectrum, and performing peak detection on the time-frequency spectrum to obtain a characteristic frequency point set; Performing regional segmentation on the time-frequency spectrum according to the characteristic frequency point set to obtain time-frequency characteristic regions, and performing energy density calculation on the time-frequency characteristic regions to obtain energy distribution characteristics; Performing principal component analysis on the energy distribution characteristics to obtain main characteristic components, and performing standardization processing on the main characteristic components to obtain normalized characteristic vectors; Performing multi-scale decomposition on the enhanced fault characteristic signal by wavelet packet transform to obtain multi-scale coefficients, and performing statistical feature extraction on the multi-scale coefficients to obtain a statistical feature set; Performing correlation analysis on the statistical feature set to obtain key statistical features, and combining the key statistical features to obtain a statistical feature vector; The normalized feature vector and the statistical feature vector are fused to obtain a fault feature vector, and the fault occurrence time and duration are calculated according to the fault feature vector to obtain the fault feature vector and key parameters.

5. The double-terminal fault location method based on signal enhancement according to claim 1, characterized in that: The step of performing machine learning analysis on the fault feature vector and key parameters to obtain a fault type identification result and a preliminary fault location estimation step includes: Performing dimensionality reduction processing on the fault feature vector to obtain feature data after dimensionality reduction, and performing cluster analysis on the feature data after dimensionality reduction to obtain feature clusters; Constructing a decision tree model according to the feature cluster to obtain a preliminary classification rule, and performing pruning optimization on the preliminary classification rule to obtain an optimized decision rule; The optimized decision rule is trained by a support vector machine algorithm to obtain a fault type classifier, and the fault type classifier is used to classify a new fault feature vector to obtain a fault type identification result; Building a neural network model according to the fault type identification result and key parameters to obtain a position estimation model, and cross-validating the position estimation model to obtain a candidate position estimation model; The candidate location estimation model is optimized by using a particle swarm algorithm to obtain an optimized location estimation model, and the fault feature vector is processed by using the optimized location estimation model to obtain a preliminary fault location estimation.

6. The double-terminal fault location method based on signal enhancement according to claim 1, characterized in that: The step of performing multi-algorithm fusion analysis on the preliminary fault location estimation to obtain accurate fault location coordinates includes: Performing a traveling wave method analysis on a preliminary fault location estimate to obtain a traveling wave method location estimate, and performing an impedance method analysis on the preliminary fault location estimate to obtain an impedance method location estimate; A Bayesian network model is constructed according to the traveling wave method position estimation and the impedance method position estimation to obtain a preliminary fusion result, and fuzzy reasoning is performed on the preliminary fusion result to obtain a fuzzy position interval; Performing genetic algorithm optimization on the fuzzy position interval to obtain a candidate position point set, and performing Delphi method evaluation on the candidate position point set to obtain a weight coefficient; Performing weighted averaging on the candidate position point set according to the weight coefficient to obtain a weighted position estimate, and performing Kalman filtering on the weighted position estimate to obtain a filtered position estimate; The error analysis of the filtered position estimation is performed through Monte Carlo simulation to obtain error distribution, and the confidence interval is calculated based on the error distribution to obtain the precise fault position coordinates.

7. The double-terminal fault location method based on signal enhancement according to claim 6, characterized in that: The steps of mapping and verifying the precise fault location coordinates on the ground to obtain the fault location results and system optimization data include: Performing geocoding conversion on the precise fault location coordinates to obtain geographic coordinates, and locating the fault point in a geographic information system according to the geographic coordinates to obtain the geographic location of the fault point; Performing terrain analysis on the geographical location of the fault point to obtain terrain feature data, and performing path planning based on the terrain feature data to obtain a target inspection route; Conducting field survey along the target inspection route to obtain field data, and performing image processing on the field data to obtain a fault feature image; Analyzing the fault feature image by a computer vision algorithm to obtain a fault verification result, and updating the fault location information according to the fault verification result to obtain a corrected fault location; Performing deviation analysis on the corrected fault location and the initial accurate fault location coordinates to obtain positioning error data, and adjusting parameters of the positioning algorithm according to the positioning error data to obtain optimized parameters; The optimization parameters and the corrected fault location are integrated to obtain the fault location result and system optimization data.

Images