An online cable fault test signal separation method and system

Abstract

The present application relates to the technical field of signal separation, in particular to an online cable fault test signal separation method and system, based on the original cable fault test signal, using fast Fourier transform algorithm, converting the time series signal to the frequency domain, through parallel computing technology, the signal is divided according to the frequency range, and through spectrum analysis, the segmented signal block is generated. The present application improves the accuracy, efficiency and real-time performance of cable fault test signal separation by comprehensively applying fast Fourier transform, principal component analysis, support vector machine classification, fixed threshold determination, dynamic time warping, long short-term memory network model and empirical mode decomposition technology. The use of parallel computing technology shortens the signal processing time, more effectively processes large-scale data sets, and at the same time meets the real-time or near real-time data processing requirements.

Description

Technical Field

[0001] This invention relates to the field of signal separation technology, and in particular to a method and system for separating signals in online cable fault testing. Background Technology

[0002] Signal separation technology encompasses methods and techniques for extracting, identifying, and separating useful information from mixed signals. It is an important branch of electronic engineering and signal processing, widely applied in fields such as communications, biomedical engineering, sound processing, and industrial fault diagnosis. Signal separation technology can process various signals, including analog, digital, and more complex time-frequency signals, and extract valuable data. By using algorithms and mathematical models to improve signal identifiability and reduce noise interference, it enhances the accuracy and efficiency of signal processing.

[0003] The online cable fault testing signal separation method is an application of signal separation technology in power system cable fault diagnosis. This method aims to monitor and analyze the signals transmitted through cables in real time to identify and locate cable faults. Its main purpose is to improve the reliability and safety of power systems by detecting faults promptly through real-time monitoring, thereby reducing the time and scope of power outages. By continuously monitoring cable signals, abnormal signals are separated, and advanced analysis techniques are used to determine the nature and location of the fault, thus achieving rapid response and repair, ensuring the stability and continuity of power supply.

[0004] While existing technologies have made significant progress in signal processing, effectively enhancing the ability to detect and locate cable faults, they still have obvious limitations in handling large-scale datasets, real-time data stream monitoring, and signal feature extraction and classification. These limitations include the difficulty in simultaneously meeting the dual demands of high efficiency and high accuracy in parallel processing and real-time anomaly detection mechanisms, as well as processing speed and real-time response capabilities. Furthermore, for processing complex and nonlinear signals, the adaptability and accuracy of existing technologies in dynamic threshold adjustment, signal similarity analysis, signal filtering, and feature extraction need improvement. The lack of effective adaptive mechanisms and optimized parameter settings limits the flexibility and accuracy of signal processing, impacting the efficiency and reliability of fault detection. Finally, existing technologies are insufficient in their ability to continuously update and learn new fault modes, limiting their adaptability and predictive accuracy in dealing with emerging fault types. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and to propose an online cable fault test signal separation method and system.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an online cable fault test signal separation method, comprising the following steps;

[0007] S1: Based on the original cable fault test signal, the fast Fourier transform algorithm is used to convert the time series signal to the frequency domain. The signal is divided according to the frequency range through parallel computing technology, and the segmented signal blocks are generated through spectrum analysis.

[0008] S2: Based on the segmented signal blocks, principal component analysis is used to extract features and reduce dimensions of multiple signal blocks, and statistically analyze the key frequency components, energy distribution and amplitude feature vectors of the signal to generate a signal feature set;

[0009] S3: Based on the signal feature set, a support vector machine classification model is used to learn and train the signal features. A hyperplane is constructed to perform pattern recognition on the fault signal. The fault signal is detected and classified according to the training results, and a fault type classification result is generated.

[0010] S4: Based on the fault type classification results, the signal is segmented twice using a fixed threshold determination method, and the faulty and non-faulty parts of the signal are separated according to a preset threshold to generate the separated signal.

[0011] S5: Based on the separated signals, a dynamic time warp method is used to perform time series analysis. By measuring and adjusting the similarity between time series, the signals are matched with a predefined signal pattern library to generate matching and similarity analysis results.

[0012] S6: Based on the matching and similarity analysis results, a long short-term memory network model is adopted to identify and predict real-time abnormal signals by analyzing the temporal characteristics and pattern changes of the signals, and to generate abnormal signal identification results by comparing historical data and real-time data.

[0013] S7: Based on the abnormal signal identification results, the empirical mode decomposition method is applied to decompose the signal into intrinsic mode functions. Based on the basic components and variation characteristics of the signal, optimization processing and noise suppression are performed to generate optimized signal features.

[0014] As a further aspect of the present invention, the segmented signal blocks include 0-10Hz signal blocks, 10-50Hz signal blocks, 50-100Hz signal blocks, timestamps, and signal strength values. The signal feature set includes the frequency values ​​of principal components within the frequency range, the corresponding energy ratios, and the direction and magnitude of the amplitude feature vectors. The fault type classification results include electrical insulation damage, conductor breakage, or poor contact, as well as the signal feature parameters of the fault, the energy peak value, and amplitude variation within the target frequency range. The separated signals are specifically the start and end timestamps of the fault signal interval and the normal signal interval. The matching and similarity analysis results are specifically the similarity score and matching degree level corresponding to the signal sequence matching the known fault mode. The abnormal signal identification results include the start time, end time, abnormality type, and abnormality intensity score of the identified abnormal signal. The optimized signal features include the frequency range, average amplitude, and phase angle of each level.

[0015] As a further aspect of the present invention, based on the original cable fault test signal, the fast Fourier transform algorithm is used to convert the time series signal to the frequency domain, the signal is divided according to the frequency range through parallel computing technology, and the segmented signal blocks are generated through spectrum analysis. The specific steps are as follows:

[0016] S101: Based on the original cable fault test signal, perform fast Fourier transform processing, use the numpy.fft.fft() function, set the sampling frequency parameter to fs=1000Hz, convert the time series signal to the frequency domain, analyze its frequency components, and generate frequency domain signal data.

[0017] S102: Based on the frequency domain signal data, parallel computing technology is used to process the data using the multiprocessing library, the signal is divided according to a preset frequency range, the processing speed is optimized and multiple frequency components are captured to generate frequency segmented signal blocks.

[0018] S103: Based on the frequency-segmented signal blocks, perform spectrum analysis using the scipy.signal.spectrogram() function, setting the window function to hann and the NFFT parameter to 256, identify and record the key frequency components and energy distribution of each segmented block, analyze the characteristics of the signal, and generate the segmented signal blocks.

[0019] As a further aspect of the present invention, the specific steps for generating a signal feature set are as follows: Based on the segmented signal blocks, principal component analysis is used to extract features and reduce dimensions of multiple signal blocks, statistically analyze key frequency components, energy distribution and amplitude feature vectors of the signal, and generate a signal feature set.

[0020] S201: Based on the segmented signal blocks, perform statistical analysis, using mean(), std() and amax() from the NumPy library to calculate the mean, standard deviation, maximum and minimum values ​​of each signal block, and extract basic statistical features from the time series to generate basic signal feature data;

[0021] S202: Based on the aforementioned basic signal feature data, principal component analysis is used to reduce dimensionality and extract key features. The data is projected onto the direction with the largest variance, reducing the dimensionality of the dataset while retaining key information, thus generating a dimensionality-reduced signal dataset.

[0022] S203: Based on the reduced-dimensional signal dataset, perform statistical analysis on the key frequency components, energy distribution, and amplitude feature vectors of the signal, statistically analyze the frequency components through vector operations, draw an energy distribution map, display the energy distribution characteristics of the signal, and generate a signal feature set.

[0023] As a further aspect of the present invention, based on the signal feature set, a support vector machine classification model is used to learn and train the signal features, and a hyperplane is constructed to perform pattern recognition on the fault signal. The specific steps for detecting and classifying the fault signal according to the training results and generating the fault type classification result are as follows:

[0024] S301: Based on the signal feature set, select the kernel function of the support vector machine algorithm. The command is SVC(kernel=rbf, C=1, gamma=scale), where the kernel parameter specifies the type of kernel function, C is the penalty parameter, and gamma is the coefficient of the kernel function. By changing the mapping dimension of the original feature space, the kernel function configuration parameters are generated.

[0025] S302: Based on the kernel function configuration parameters, the parameters of the vector machine model are optimized using cross-validation and grid search methods, including the command GridSearchCV(SVC(), param_grid={C: [0.1, 1, 10], gamma:[0.001, 0.0001]}, cv=5), which generates a parameter-optimized model by balancing model complexity and fit.

[0026] S303: Based on the optimized model with the parameters, use the fit and predict methods to train the model and classify the fault signals. The command items include optimizedSVM.fit(X_train, y_train) and optimizedSVM.predict(X_test), construct a classification hyperplane to distinguish cable fault types, and generate fault type classification results.

[0027] As a further aspect of the present invention, based on the fault type classification results, a fixed threshold determination method is used to perform secondary segmentation of the signal, and the faulty part and non-faulty part of the signal are separated according to a preset threshold. The specific steps for generating the separated signal are as follows.

[0028] S401: Based on the fault type classification results, a fixed threshold determination method is used to segment the signal. A target threshold is set, and the normal signal is distinguished from the fault signal by statistical signal characteristics and comparison with the preset threshold, and an initial segmentation signal is generated.

[0029] S402: Based on the initial segmentation signal, by analyzing the time series characteristics of the signal, the fault interval is refined using the moving window technique. The size and step size of the moving window are set, the signal is traversed using the sliding window method, and a preset threshold is applied for judgment to generate a refined segmentation signal.

[0030] S403: Based on the refined segmented signal, using signal reconstruction technology, filter fault signals based on a preset threshold to reduce interference to non-faulty parts, and separate the faulty and non-faulty parts of the signal to generate the separated signal.

[0031] As a further aspect of the present invention, based on the separated signals, a dynamic time warp method is used to perform time series analysis. By measuring and adjusting the similarity between time series, the signals are matched with a predefined signal pattern library to generate matching and similarity analysis results. The specific steps are as follows:

[0032] S501: Based on the separated signal, construct the cost matrix using the dynamic time warp algorithm, calculate the Euclidean distance using the command item numpy.linalg.norm, set the parameter to ord=2, fill the cost matrix, and generate the initial cost matrix;

[0033] S502: Based on the initial cost matrix, calculate the cumulative distance and backtracking path using a dynamic programming algorithm, use the numpy.min function to capture the path with the minimum cumulative cost, set the axis parameter to specify the search direction of the minimum value, and generate the optimal matching path;

[0034] S503: Based on the optimal matching path, the dynamic time warp algorithm is used again to calculate the similarity coefficient between the time series and each pattern, match it with the predefined signal pattern library, evaluate the similarity, and generate matching and similarity analysis results.

[0035] As a further aspect of the present invention, based on the matching and similarity analysis results, a long short-term memory network model is adopted to identify and predict real-time abnormal signals by analyzing the temporal characteristics and pattern changes of the signals. The specific steps for generating abnormal signal identification results by comparing historical data and real-time data are as follows:

[0036] S601: Based on the matching and similarity analysis results, the input layer dimension is set to the number of signal features, the number of hidden layer nodes is set to 128, and the number of output layer nodes corresponds to the number of predicted categories. A long short-term memory network model is constructed, and training parameters are set, including learning rate lr=0.001, batch size of batch_size=32, and number of iterations epochs=100, generating the initial model architecture.

[0037] S602: Based on the initialized model architecture, the Adam optimizer is applied to iteratively optimize the model parameters. By minimizing the CrossEntropyLoss function, the accuracy of the model in classifying abnormal signals is optimized, and the trained and optimized model is generated.

[0038] S603: Based on the trained and optimized model, real-time signal data is input into the model and propagated forward. The model is used to predict its abnormal state. By comparing the output value of the model with a preset threshold, the signal is classified into abnormal levels, and an abnormal signal identification result is generated.

[0039] As a further aspect of the present invention, based on the abnormal signal identification results, the empirical mode decomposition method is applied to decompose the signal into intrinsic mode functions. Based on the basic components and variation characteristics of the signal, optimization processing and noise suppression are performed to generate optimized signal features. The specific steps are as follows:

[0040] S701: Based on the abnormal signal identification results, the empirical mode decomposition method is adopted to decompose the signal step by step through an adaptive process, decomposing it into multiple intrinsic mode functions and residual terms, including the identification of extreme points, the construction of envelopes and the screening criteria of intrinsic mode functions, generating eigenmode decomposition results;

[0041] S702: Based on the intrinsic mode decomposition results, the Hilbert-Huang transform method is used to perform time-frequency analysis, identify noise components, and remove high-frequency noise through energy analysis to generate a noise-suppressed signal;

[0042] S703: Based on the noise-suppressed signal, the signal is smoothed using window function smoothing and frequency filtering techniques to optimize signal characteristics. By adjusting the statistical and spectral characteristics of the signal, key features in the signal are highlighted, and optimized signal characteristics are generated.

[0043] As a further aspect of the present invention, an online cable fault test signal separation system is provided, the system comprising a frequency domain conversion module, a frequency segmentation module, a feature extraction module, a classification model training module, a signal segmentation module, a signal matching module, an abnormal signal identification module, and a signal optimization module;

[0044] The frequency domain conversion module uses the original cable fault test signal and the Fast Fourier Transform algorithm to convert the time series signal to the frequency domain, analyze its frequency components, and generate frequency domain signal data.

[0045] The frequency segmentation module is based on frequency domain signal data and uses parallel computing technology to segment the signal according to a preset frequency range, optimize processing speed and capture multiple frequency components to generate frequency segmented signal blocks.

[0046] The feature extraction module is based on frequency segmentation signal blocks and uses principal component analysis to extract key features, including key frequency components, energy distribution and amplitude feature vectors of the statistical signal, to generate a dimensionality-reduced signal dataset.

[0047] The classification model training module is based on a dimensionality-reduced signal dataset and uses a support vector machine classification model to train and learn the model, construct a classification hyperplane to distinguish cable fault types, and generate fault type classification results.

[0048] The signal segmentation module performs secondary segmentation on the signal based on the fault type classification results and uses a fixed threshold determination method to separate the fault part and the non-fault part, generating the separated signal.

[0049] The signal matching module uses a dynamic time warp method to match the separated signals to a predefined signal pattern library, generating matching and similarity analysis results.

[0050] The abnormal signal identification module, based on the matching and similarity analysis results, uses a long short-term memory network model to identify and predict abnormal signals in real time, and generates abnormal signal identification results.

[0051] The signal optimization module, based on the abnormal signal identification results, applies the empirical mode decomposition method to optimize the signal and suppress noise, including signal decomposition, noise component identification and removal, and signal reconstruction, to generate optimized signal features.

[0052] Compared with existing technologies, the advantages and positive effects of this invention are as follows: By comprehensively applying Fast Fourier Transform, Principal Component Analysis, Support Vector Machine classification, fixed threshold determination, dynamic time warp, Long Short-Term Memory network model, and Empirical Mode Decomposition (EMD), this invention improves the accuracy, efficiency, and real-time performance of cable fault test signal separation. Parallel computing technology shortens signal processing time, enabling more efficient processing of large-scale datasets while meeting real-time or near-real-time data processing requirements. Through dynamic threshold adjustment, enhanced signal similarity analysis, and improved signal filtering and feature extraction strategies, the adaptability and accuracy to complex and nonlinear signal processing are improved. This not only enhances the flexibility and reliability of fault detection but also, through a real-time anomaly detection mechanism, strengthens the system's adaptability and predictive accuracy to newly emerging fault types, ensuring high efficiency and high accuracy in cable fault diagnosis. Attached Figure Description

[0053] Figure 1 The present invention provides a flowchart of an online cable fault test signal separation method;

[0054] Figure 2 This invention provides a detailed flowchart of step S1 in an online cable fault test signal separation method.

[0055] Figure 3 This invention provides a detailed flowchart of step S2 in an online cable fault test signal separation method.

[0056] Figure 4 This invention provides a detailed flowchart of step S3 in an online cable fault test signal separation method.

[0057] Figure 5 This invention provides a detailed flowchart of step S4 in an online cable fault test signal separation method.

[0058] Figure 6 This invention provides a detailed flowchart of step S5 in an online cable fault test signal separation method.

[0059] Figure 7 This invention provides a detailed flowchart of step S6 in an online cable fault test signal separation method.

[0060] Figure 8 This invention provides a detailed flowchart of step S7 in an online cable fault test signal separation method.

[0061] Figure 9 This invention provides a module diagram of an online cable fault test signal separation system. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0063] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0064] Example 1

[0065] Please see Figure 1 The present invention provides a technical solution: an online cable fault test signal separation method. In order to achieve the above objective, the present invention adopts the following technical solution: an online cable fault test signal separation method, comprising the following steps;

[0066] S1: Based on the original cable fault test signal, the fast Fourier transform algorithm is used to convert the time series signal to the frequency domain. The signal is divided according to the frequency range through parallel computing technology, and the segmented signal blocks are generated through spectrum analysis.

[0067] S2: Based on the segmented signal blocks, principal component analysis is used to extract features and reduce dimensions of multiple signal blocks, statistically analyze key frequency components, energy distribution and amplitude feature vectors of the signal, and generate a signal feature set;

[0068] S3: Based on the signal feature set, a support vector machine classification model is used to learn and train the signal features. By constructing a hyperplane, the fault signal is pattern recognized. Based on the training results, the fault signal is detected and classified, and the fault type classification result is generated.

[0069] S4: Based on the fault type classification results, the fixed threshold judgment method is used to perform secondary segmentation of the signal, and the fault part and non-fault part of the signal are separated according to the preset threshold to generate the separated signal;

[0070] S5: Based on the separated signals, a dynamic time warp method is used to perform time series analysis. By measuring and adjusting the similarity between time series, the signals are matched with a predefined signal pattern library to generate matching and similarity analysis results.

[0071] S6: Based on the matching and similarity analysis results, a long short-term memory network model is adopted to identify and predict the behavior of real-time abnormal signals by analyzing the temporal characteristics and pattern changes of the signals. By comparing historical data and real-time data, abnormal signal identification results are generated.

[0072] S7: Based on the abnormal signal identification results, the empirical mode decomposition method is applied to decompose the signal into intrinsic mode functions. According to the basic components and variation characteristics of the signal, optimization processing and noise suppression are performed to generate optimized signal features.

[0073] The segmented signal blocks include 0-10Hz, 10-50Hz, and 50-100Hz signal blocks, along with timestamps and signal strength values. The signal feature set includes the frequency values ​​of the principal components within the frequency range, the corresponding energy ratios, and the direction and magnitude of the amplitude feature vectors. The fault type classification results include electrical insulation damage, conductor breakage, or poor contact, as well as the signal characteristic parameters of the fault, the peak energy and amplitude variations within the target frequency range. The separated signals are specifically the start and end timestamps of the fault signal interval and the normal signal interval. The matching and similarity analysis results are specifically the similarity scores and matching degree levels corresponding to the signal sequences matched with known fault modes. The abnormal signal identification results include the start time, end time, abnormality type, and abnormality intensity score of the identified abnormal signals. The optimized signal features include the frequency range, average amplitude, and phase angle of each level.

[0074] In step S1, the original cable fault test signal is transformed from a time series to the frequency domain and segmented using a Fast Fourier Transform (FFT) algorithm and parallel computing technology. The FFT is employed to analyze the signal's frequency components in detail based on the set sampling frequency parameters. Using a parallel computing framework, the signal is divided into multiple frequency blocks according to a predetermined frequency range. Spectral analysis techniques are used to refine the segmentation, identifying the key frequency components and energy distribution of each frequency block. This improves processing speed while capturing key frequency information of the signal, laying the foundation for subsequent fault analysis and generating frequency-segmented signal blocks that facilitate further analysis.

[0075] In step S2, based on the generated segmented signal blocks, principal component analysis (PCA) is used to extract features and reduce dimensionality. Basic statistical characteristics, including mean, standard deviation, maximum, and minimum values, are calculated for each signal block. PCA captures the projection of the data onto the direction of maximum variance, effectively reducing data dimensionality while preserving key information. This not only simplifies the signal data but also enhances the identification of key frequency components, energy distribution, and amplitude feature vectors in the signal, providing an important data foundation for accurate classification of fault signals and generating a dimensionality-reduced signal dataset with key frequency and energy information.

[0076] In step S3, a support vector machine (SVM) classification model is used to learn and train the extracted signal feature set. By selecting an appropriate kernel function and setting penalty parameters and kernel coefficients, the mapping dimension of the original feature space is adjusted. Cross-validation and grid search techniques are used to optimize model parameters, balancing model complexity and fit. The trained model is then used to classify fault signals. By constructing a classification hyperplane, it distinguishes various types of cable faults, including electrical insulation damage, conductor breakage, or poor contact. Accurate classification of fault types is performed, generating detailed fault type classification results, providing crucial information for precise fault location.

[0077] In step S4, based on the fault type classification results, a fixed threshold method is used to accurately segment the signal, clearly distinguishing between fault signal and normal signal intervals. By setting a target threshold and comparing signal characteristics, the segmented signal is initialized, and time series analysis and moving window techniques are used to refine the fault intervals. Separating the fault signal and normal signal intervals provides a clear foundation for subsequent signal processing and analysis, ensuring minimal interference to non-faulty components.

[0078] In step S5, a dynamic time warp method is used to perform time series analysis on the separated signals, measuring and adjusting the similarity between time series, and matching them against a predefined signal pattern library. A cost matrix is ​​constructed, the Euclidean distance between time series is calculated, and dynamic programming is used to capture the path with the minimum cumulative cost, achieving a high degree of matching between the signals and known fault modes. This improves the accuracy and efficiency of fault diagnosis, generating matching and similarity analysis results with high similarity scores and matching degree levels, providing a precise basis for the identification of real-time abnormal signals.

[0079] In step S6, a Long Short-Term Memory (LSTM) network model is used to analyze the matching and similarity analysis results, providing an in-depth analysis of the signal's temporal characteristics and pattern changes. By setting appropriate network architecture parameters—including an input layer dimension corresponding to the number of signal features, 128 hidden layer nodes, and an output layer node count corresponding to the number of predicted categories—and configuring training parameters such as learning rate, batch size, and number of iterations, the model is iteratively optimized using the Adam optimizer and cross-entropy loss function. This enables the model to predict abnormal signal states based on historical and real-time data, generating abnormal signal identification results including the start time, end time, anomaly type, and anomaly intensity score. This improves the accuracy of fault prediction and real-time response capabilities, providing strong data support for timely fault handling.

[0080] In step S7, based on the identified anomalous signal results, empirical mode decomposition (EMD) is applied to perform in-depth analysis and processing of the signal. By decomposing the signal step-by-step into multiple intrinsic mode functions and residual terms, time-frequency analysis is performed using Hilbert-Huang transform to identify and remove high-frequency noise components. This improves signal clarity. Furthermore, window function smoothing and frequency filtering techniques are used to optimize signal characteristics, emphasizing key information and generating optimized signal features including frequency range, average amplitude, and phase angle for each level. This enhances the accuracy of signal processing, providing high-quality signal data for fault diagnosis and analysis, ensuring the accuracy and efficiency of fault handling.

[0081] Please see Figure 2 Based on the original cable fault test signal, the fast Fourier transform algorithm is used to convert the time series signal to the frequency domain. The signal is divided according to the frequency range through parallel computing technology, and the segmented signal blocks are generated through spectrum analysis. The specific steps are as follows:

[0082] S101: Based on the original cable fault test signal, perform fast Fourier transform processing, use the numpy.fft.fft() function, set the sampling frequency parameter to fs=1000Hz, convert the time series signal to the frequency domain, analyze its frequency components, and generate frequency domain signal data.

[0083] S102: Based on frequency domain signal data, parallel computing technology is used to process the data using the multiprocessing library, the signal is divided according to a preset frequency range, the processing speed is optimized and multiple frequency components are captured to generate frequency segmented signal blocks.

[0084] S103: Based on frequency-segmented signal blocks, perform spectrum analysis using the scipy.signal.spectrogram() function, setting the window function to hann and the NFFT parameter to 256, identify and record the key frequency components and energy distribution of each segmented block, analyze the characteristics of the signal, and generate the segmented signal blocks.

[0085] In substep S101, the `fft` function is called using the `fft` module of the NumPy library to perform a Fast Fourier Transform (FFT) on the original cable fault test signal. The time-series signal is converted to the frequency domain using a sampling frequency parameter `fs` of 1000Hz, and its frequency components are analyzed. The `numpy.fft.fft()` function is imported and used to transform the signal, obtaining frequency domain signal data. This effectively converts the time-series signal into a frequency domain representation, facilitating subsequent frequency component analysis. The generated frequency domain signal data provides the foundation for subsequent steps.

[0086] In sub-step S102, parallel computation is performed using the multiprocessing library to process the frequency domain signal data. This includes segmenting the signal according to a preset frequency range, optimizing processing speed, and capturing multiple frequency components. Tasks are assigned using the Process or Pool functions in the multiprocessing library to process the signal data in parallel within each frequency range. This not only improves data processing speed but also ensures that frequency-segmented signal blocks can be effectively generated, supporting the capture and analysis of multiple frequency components of cable fault test signals.

[0087] In substep S103, the spectrogram function is called from the signal module of the scipy library to perform spectral analysis on the frequency-segmented signal blocks. The window function is set to hann, and the NFFT parameter is 256 to identify and record the key frequency components and energy distribution of each segment. scipy.signal is imported, and the spectrogram function is used to process each signal block and analyze its spectral characteristics. This reveals the key frequency components of the signal block and provides detailed energy distribution information for signal feature analysis.

[0088] Please see Figure 3 The key feature is that, based on the segmented signal blocks, principal component analysis is used to extract features and reduce dimensions of multiple signal blocks, and statistically analyze the key frequency components, energy distribution and amplitude feature vectors of the signal to generate a signal feature set. The specific steps are as follows:

[0089] S201: Based on the segmented signal blocks, perform statistical analysis, using mean(), std() and amax() from the NumPy library to calculate the mean, standard deviation, maximum and minimum values ​​of each signal block, and extract basic statistical features from the time series to generate basic signal feature data;

[0090] S202: Based on basic signal feature data, principal component analysis is used to reduce dimensionality and extract key features. The data is projected onto the direction with the largest variance, reducing the dimensionality of the dataset while retaining key information, thus generating a dimensionality-reduced signal dataset.

[0091] S203: Based on a dimensionality-reduced signal dataset, statistical analysis is performed on the key frequency components, energy distribution, and amplitude feature vectors of the signal. Frequency components are statistically analyzed through vector operations, an energy distribution map is drawn, the energy distribution characteristics of the signal are displayed, and a signal feature set is generated.

[0092] In substep S201, the mean, standard deviation, maximum, and minimum values ​​of each segmented signal block are calculated using the NumPy library to perform statistical analysis. The mean(), std(), and amax() functions are used to extract basic statistical features from the time series, generating basic signal feature data. The calculation of statistics provides a quantitative description of the signal's basic characteristics, laying the foundation for subsequent feature extraction and dimensionality reduction.

[0093] In substep S202, principal component analysis (PCA) is used to reduce the dimensionality of the basic signal feature data and extract key features. This involves projecting the data onto the direction of maximum variance, reducing the dimensionality of the dataset while preserving crucial information. PCA is performed using sklearn.decomposition.PCA or a similar library, selecting an appropriate number of components for dimensionality reduction. This not only reduces data complexity but also highlights the most important features in the signal data, facilitating efficient signal analysis and fault diagnosis.

[0094] In substep S203, statistical analysis is performed based on the dimensionality-reduced signal dataset, including statistics on the key frequency components, energy distribution, and amplitude feature vectors of the signal. Frequency components are statistically analyzed through vector operations, and an energy distribution map is plotted to display the signal's energy distribution characteristics, analyze key signal features, and visually represent the signal's energy characteristics, providing crucial information support for cable fault diagnosis and analysis.

[0095] Please see Figure 4 Based on the signal feature set, a support vector machine classification model is used to learn and train the signal features. By constructing a hyperplane, fault signals are used for pattern recognition. Based on the training results, fault signals are detected and classified, and fault type classification results are generated. The specific steps are as follows:

[0096] S301: Based on the signal feature set, select the kernel function of the support vector machine algorithm. The command is SVC(kernel=rbf, C=1, gamma=scale), where the kernel parameter specifies the type of kernel function, C is the penalty parameter, and gamma is the coefficient of the kernel function. By changing the mapping dimension of the original feature space, the kernel function configuration parameters are generated.

[0097] S302: Based on kernel function configuration parameters, cross-validation and grid search methods are used to optimize the parameters of the vector machine model, including the command GridSearchCV(SVC(), param_grid={C: [0.1, 1, 10], gamma: [0.001,0.0001]}, cv=5), which generates a parameter-optimized model by balancing model complexity and fit.

[0098] S303: Based on the optimized model, the fit and predict methods are used to train the model and classify fault signals. The command items include optimizedSVM.fit(X_train, y_train) and optimizedSVM.predict(X_test), which construct a classification hyperplane to distinguish cable fault types and generate fault type classification results.

[0099] In substep S301, the signal features are learned and trained by selecting the Radial Basis Function (RBF) of the Support Vector Machine (SVM) algorithm as the kernel function. The command SVC(kernel=rbf, C=1, gamma=scale) maps the original feature space to a higher dimension. Here, kernel=rbf determines the use of the radial basis function, C=1 is set as the penalty parameter, representing the degree of penalty for misclassification, and gamma=scale automatically adjusts the γ parameter according to the number of features to optimize the classification boundary. Kernel function configuration parameters are generated to ensure that the model can effectively capture and distinguish features of various fault signals.

[0100] In substep S302, the parameters of the support vector machine model are optimized using cross-validation and grid search. The command `GridSearchCV(SVC(), param_grid={C: [0.1, 1, 10], gamma: [0.001,0.0001]}, cv=5)` systematically searches for various parameter combinations, where `param_grid={C: [0.1, 1, 10],gamma: [0.001, 0.0001]}` specifies the range of the penalty parameter C and the gamma parameter to be optimized, and `cv=5` indicates that 5-fold cross-validation is used to evaluate the performance of each parameter combination. This balances model complexity with good fit, generating a parameter-optimized model. This improves the model's generalization ability, laying the foundation for accurately classifying various types of cable faults.

[0101] In substep S303, the optimized model is trained using the `fit` method and its fault signals are classified using the `predict` method. The commands `optimizedSVM.fit(X_train, y_train)` and `optimizedSVM.predict(X_test)` specifically train the model on the training dataset and predict fault types on the test dataset. A classification hyperplane capable of distinguishing cable fault types is constructed, effectively separating fault signals from normal signals, and fault signal detection and classification are performed based on the model training results. The generated fault type classification results not only improve the accuracy of fault diagnosis but also accelerate fault response, providing a reliable basis for rapid cable fault location and repair.

[0102] Please see Figure 5 Based on the fault type classification results, a fixed threshold judgment method is used to perform secondary segmentation of the signal, and the fault part and non-fault part of the signal are separated according to the preset threshold. The specific steps to generate the separated signal are as follows.

[0103] S401: Based on the fault type classification results, the fixed threshold judgment method is used to segment the signal. The target threshold is set, and the normal signal is distinguished from the fault signal by statistical signal characteristics and comparison with the preset threshold, and an initial segmentation signal is generated.

[0104] S402: Based on the initial segmentation signal, by analyzing the time series characteristics of the signal, the fault interval is refined using the moving window technique. The size and step size of the moving window are set, the signal is traversed in a sliding window manner, and a preset threshold is applied for judgment to generate a refined segmentation signal.

[0105] S403: Based on the refined segmented signal, the signal reconstruction technology is used to filter fault signals based on a preset threshold, reduce interference to non-faulty parts, and separate the faulty and non-faulty parts of the signal to generate the separated signal.

[0106] In sub-step S401, the signal is initially segmented using a fixed threshold method based on the fault type classification results. A target threshold is set, and normal signals are distinguished from fault signals by comparing statistical signal characteristics with the preset threshold. An initial segmentation signal is generated, providing a basis for more refined fault interval location. By accurately separating potential fault signals, the computational burden is reduced for subsequent detailed analysis and processing.

[0107] In sub-step S402, based on the initialized segmented signal, the fault region is refined by analyzing the signal's time-series characteristics and applying a moving window technique. By setting an appropriate moving window size and step size, the entire signal is traversed using a sliding window approach. Detailed judgments are made based on preset thresholds to generate a refined segmented signal. This allows for more accurate identification of the fault signal's specific location, providing more precise input data for subsequent fault analysis and processing.

[0108] In sub-step S403, fault signals determined based on preset thresholds are filtered using signal reconstruction technology to reduce interference with non-faulty components. The faulty and non-faulty components of the signal are separated, and the separated signal is generated. This improves the accuracy of fault signal processing, ensuring that subsequent steps are performed based on clearly defined signals, thereby enhancing the efficiency and accuracy of the entire fault detection and classification system.

[0109] Please see Figure 6 Based on the separated signals, a dynamic time warp method is used to perform time series analysis. By measuring and adjusting the similarity between time series, the signals are matched with a predefined signal pattern library to generate matching and similarity analysis results. The specific steps are as follows:

[0110] S501: Based on the separated signal, a cost matrix is ​​constructed using a dynamic time warp algorithm. The Euclidean distance is calculated using the command item numpy.linalg.norm, with the parameter set to ord=2. The cost matrix is ​​then filled to generate the initial cost matrix.

[0111] S502: Based on the initial cost matrix, the cumulative distance and backtracking path are calculated using a dynamic programming algorithm. The numpy.min function is used to capture the path with the minimum cumulative cost. The axis parameter is set to specify the search direction of the minimum value, and the optimal matching path is generated.

[0112] S503: Based on the optimal matching path, the dynamic time warp algorithm is used again to calculate the similarity coefficient between the time series and each pattern, match it with the predefined signal pattern library, evaluate the similarity, and generate matching and similarity analysis results.

[0113] In substep S501, a cost matrix is ​​constructed using the dynamic time warp algorithm. The Euclidean distance between the separated signals is calculated using `numpy.linalg.norm`, and the parameter `ord=2` is set to fill the cost matrix, generating an initial cost matrix. Signal data is input as an array. The dynamic time warp algorithm calculates the distance between each pair of points and fills it into the cost matrix. Generating the initial cost matrix is ​​a crucial first step in the dynamic time warp process, ensuring that the minimum cost path can be found through dynamic programming, thereby measuring and adjusting the similarity between time series.

[0114] In substep S502, based on the initialized cost matrix, the cumulative distance and backtracking path are calculated using a dynamic programming algorithm. The `numpy.min` function is used to capture the path with the minimum cumulative cost, and the `axis` parameter specifies the search direction for the minimum value, generating the optimal matching path. In this step, the dynamic programming algorithm progressively constructs the cumulative cost matrix and finds the optimal matching path through backtracking. This ensures optimal alignment between time series, providing an accurate foundation for subsequent similarity analysis and pattern matching.

[0115] In substep S503, based on the optimal matching path, the dynamic time warp algorithm is used again to calculate the similarity coefficient between the time series and each pattern, and then matched with a predefined signal pattern library to evaluate the similarity. The calculation of the similarity coefficient provides a quantitative indicator of the degree of matching between each time series and the patterns in the signal pattern library, evaluates the similarity between the time series and known patterns, and provides a basis for identification and classification.

[0116] Please see Figure 7 Based on the matching and similarity analysis results, a long short-term memory network model is adopted to identify and predict real-time abnormal signals by analyzing the temporal characteristics and pattern changes of the signals. The specific steps for generating abnormal signal identification results by comparing historical data and real-time data are as follows:

[0117] S601: Based on the matching and similarity analysis results, the input layer dimension is set to the number of signal features, the number of hidden layer nodes is set to 128, and the number of output layer nodes corresponds to the number of predicted categories. A long short-term memory network model is constructed, and training parameters are set, including learning rate lr=0.001, batch size of batch_size=32, and number of iterations epochs=100, generating the initial model architecture.

[0118] S602: Based on the initialized model architecture, the Adam optimizer is applied to iteratively optimize the model parameters. By minimizing the CrossEntropyLoss function, the accuracy of the model in classifying abnormal signals is optimized, and the trained and optimized model is generated.

[0119] S603: Based on the trained and optimized model, real-time signal data is input into the model and propagated forward. The model is used to predict its abnormal state. By comparing the model's output value with a preset threshold, the signal is classified into abnormal levels, and abnormal signal identification results are generated.

[0120] In substep S601, based on the matching and similarity analysis results, a Long Short-Term Memory (LSTM) network model is constructed. The input layer dimension is set to the number of signal features, the number of hidden layer nodes is set to 128, and the number of output layer nodes corresponds to the number of predicted categories. Training parameters are set, including a learning rate of lr=0.001, a batch size of batch_size=32, and the number of iterations epochs=100, generating an initialized model architecture. This captures the temporal characteristics and pattern changes of the signal, providing a powerful tool for real-time anomaly signal identification and behavior prediction.

[0121] In substep S602, the Adam optimizer is applied to iteratively optimize the initialized model architecture. By minimizing the CrossEntropyLoss function, the accuracy of the model in classifying abnormal signals is improved. Through repeated iterative training, the internal parameters are adjusted cyclically to improve classification accuracy, generating a trained and optimized model. The trained and optimized model can more accurately identify and classify abnormal signals, providing core technical support for real-time monitoring and early warning systems.

[0122] In sub-step S603, real-time signal data is input into the trained and optimized model and propagated forward to predict its abnormal state. By comparing the model's output value with a preset threshold, the signal is classified into anomaly levels, generating anomaly signal identification results. This allows for real-time identification and classification of signal anomalies, providing a basis for timely fault diagnosis and system maintenance.

[0123] Please see Figure 8 Based on the abnormal signal identification results, the empirical mode decomposition method is applied to decompose the signal into intrinsic mode functions. According to the basic components and variation characteristics of the signal, optimization processing and noise suppression are carried out to generate optimized signal features. The specific steps are as follows:

[0124] S701: Based on the abnormal signal identification results, the empirical mode decomposition method is adopted to decompose the signal step by step through an adaptive process, decomposing it into multiple intrinsic mode functions and residual terms, including the identification of extreme points, the construction of envelopes and the selection criteria of intrinsic mode functions, generating eigenmode decomposition results;

[0125] S702: Based on the intrinsic mode decomposition results, the Hilbert-Huang transform method is used to perform time-frequency analysis, identify noise components, and remove high-frequency noise through energy analysis to generate a noise-suppressed signal;

[0126] S703: Based on the noise-suppressed signal, it uses window function smoothing and frequency filtering techniques to smooth the signal, optimize signal characteristics, and highlight key features in the signal by adjusting the statistical and spectral characteristics of the signal, thereby generating optimized signal features.

[0127] In substep S701, the Empirical Mode Decomposition (EMD) method is employed to decompose the signal step-by-step through an adaptive process, breaking it down into multiple intrinsic mode functions and residual terms. This includes the identification of extrema, the construction of the envelope, and the selection criteria for intrinsic mode functions, generating the EMD results. The application of the EMD method provides in-depth analysis of the basic components and variation characteristics of the signal, laying the foundation for subsequent optimization processing and noise suppression.

[0128] In the S702 sub-step, based on the intrinsic mode decomposition results, time-frequency analysis is performed using the Hilbert-Huang transform method to identify noise components. High-frequency noise is then removed using energy analysis to generate a noise-suppressed signal. This process accurately identifies noise components in the signal and effectively removes noise, providing a foundation for signal analysis and applications.

[0129] In substep S703, based on the noise-suppressed signal, window function smoothing and frequency filtering techniques are used to smooth the signal and optimize its characteristics. By adjusting the signal's statistical and spectral characteristics, key features are highlighted, generating optimized signal features. This improves signal quality and provides data support for subsequent signal processing and analysis.

[0130] Please see Figure 9 An online cable fault test signal separation system includes a frequency domain conversion module, a frequency segmentation module, a feature extraction module, a classification model training module, a signal segmentation module, a signal matching module, an abnormal signal identification module, and a signal optimization module.

[0131] The frequency domain conversion module uses the original cable fault test signal and the Fast Fourier Transform algorithm to convert the time series signal to the frequency domain, analyze its frequency components, and generate frequency domain signal data.

[0132] The frequency segmentation module is based on frequency domain signal data and uses parallel computing technology to segment the signal according to a preset frequency range, optimize processing speed and capture multiple frequency components to generate frequency segmented signal blocks.

[0133] The feature extraction module is based on frequency segmentation signal blocks and uses principal component analysis to extract key features, including key frequency components, energy distribution and amplitude feature vectors of the statistical signal, to generate a dimensionality-reduced signal dataset.

[0134] The classification model training module is based on a dimensionality-reduced signal dataset and uses a support vector machine classification model for model training and learning. It constructs a classification hyperplane to distinguish cable fault types and generates fault type classification results.

[0135] Based on the fault type classification results, the signal segmentation module uses a fixed threshold determination method to perform secondary segmentation on the signal, separating the faulty part and the non-faulty part, and generating the separated signal.

[0136] The signal matching module uses a dynamic time warp method to match the separated signals to a predefined signal pattern library, generating matching and similarity analysis results.

[0137] Based on the matching and similarity analysis results, the abnormal signal recognition module uses a long short-term memory network model to identify and predict abnormal signals in real time, and generate abnormal signal recognition results.

[0138] Based on the abnormal signal identification results, the signal optimization module applies the empirical mode decomposition method to optimize the signal and suppress noise, including signal decomposition, noise component identification and removal, and signal reconstruction to generate optimized signal features.

[0139] The frequency domain conversion module uses a Fast Fourier Transform (FFT) algorithm to convert the original cable fault test signal from a time series to the frequency domain. This reveals the frequency components of the signal, improves the accuracy of subsequent processing steps, and enables the system to accurately identify and analyze the frequency characteristics of cable faults, providing crucial preliminary data for fault type determination.

[0140] The frequency segmentation module utilizes parallel computing technology to segment the signal according to a preset frequency range, optimizing processing speed and capturing multiple frequency components in the signal. This improves data processing efficiency and, through the captured key frequency components, provides accurate input for feature extraction and fault classification, ensuring the depth and accuracy of fault analysis.

[0141] The feature extraction module extracts key features from the frequency-segmented signal blocks using principal component analysis, including key frequency components, energy distribution, and amplitude feature vectors. This reduces data complexity while retaining crucial information for fault diagnosis, improving the efficiency and accuracy of classification model training and laying the foundation for accurately identifying various types of cable faults.

[0142] The classification model training module employs a support vector machine (SVM) classification model, training and learning the model based on the extracted signal feature set. By constructing a classification hyperplane, it can effectively distinguish the types of cable faults and generate accurate fault type classification results. This improves the accuracy of fault detection and provides reliable data support for precise fault location and rapid response.

[0143] The signal segmentation module performs secondary segmentation of the signal based on the fault type classification results, using a fixed threshold determination method to effectively separate faulty and non-faulty components. This provides clear target signals for fault analysis, facilitates rapid repair and fault recovery, and reduces system downtime.

[0144] The signal matching module utilizes a dynamic time warp method to match the separated signals with a predefined signal pattern library, generating matching and similarity analysis results. This improves the flexibility and accuracy of signal recognition, enabling the system to effectively identify signals that match known fault modes, providing strong support for the accurate identification of abnormal signals.

[0145] The abnormal signal recognition module analyzes the matching and similarity analysis results using a long short-term memory network model to identify abnormal signals and predict their behavior in real time. It can promptly detect potential fault signals and predict fault development trends, providing a scientific basis for taking preventative measures and reducing potential damage.

[0146] The signal optimization module employs empirical mode decomposition (EMD) to optimize and suppress noise in the abnormal signal identification results. Through signal decomposition, noise component identification and removal, and signal reconstruction, it provides high-quality signal data for accurate fault diagnosis and effective analysis, ensuring the efficient and accurate operation of the system.

[0147] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An online cable fault test signal separation method, characterized by, The method comprises the following steps: Based on the original cable fault test signal, the time series signal is converted to the frequency domain by using the fast Fourier transform algorithm, the signal is segmented according to the frequency range by using the parallel computing technology, and the segmented signal block is generated by spectrum analysis; Based on the segmented signal block, the principal component analysis method is used to extract features and reduce dimensions of multiple signal blocks, and the signal feature set is generated by counting the signal key frequency component, energy distribution and amplitude feature vector; Based on the signal feature set, the support vector machine classification model is used to learn and train the signal features, the pattern recognition of the fault signal is performed by constructing a hyperplane, the fault signal detection and classification are performed according to the training result, and the fault type classification result is generated; Based on the fault type classification result, the fixed threshold decision method is used to perform secondary segmentation on the signal, and the fault part and the non-fault part of the signal are separated according to the preset threshold, and the separated signal is generated; Based on the separated signal, the dynamic time warping method is used for time series analysis, the similarity between time series is measured and adjusted, the signal is matched with the pre-defined signal mode library, and the matching and similarity analysis result is generated; Based on the matching and similarity analysis result, the long short-term memory network model is used to analyze the time sequence characteristics and pattern changes of the signal, and the real-time abnormal signal is recognized and behavior predicted, and the abnormal signal recognition result is generated by comparing the historical data and the real-time data; Based on the abnormal signal recognition result, the empirical mode decomposition method is used to decompose the signal into intrinsic mode functions, and the optimized signal features are generated by optimizing and noise suppressing according to the basic components and change characteristics of the signal.

2. The online cable fault test signal separation method of claim 1, wherein: The segmented signal block includes 0-10Hz signal block, 10-50Hz signal block, 50-100Hz signal block, time stamp and signal intensity value, the signal feature set includes frequency value of principal component in frequency range, corresponding energy ratio and direction and size of amplitude feature vector, the fault type classification result includes signal feature parameters of electrical insulation damage, conductor breakage or poor contact, energy peak value and amplitude change in target frequency range, the separated signal is specifically the start and end time stamps of the fault signal interval and the normal signal interval, the matching and similarity analysis result is specifically the similarity score and matching degree level corresponding to the signal sequence matched with the known fault mode, the abnormal signal recognition result includes the start time, end time, abnormal type and abnormal intensity score of the recognized abnormal signal, and the optimized signal feature includes the frequency range, average amplitude and phase angle of each level.

3. The online cable fault test signal separation method of claim 1, wherein: Based on the original cable fault test signal, the time series signal is converted to the frequency domain by using the fast Fourier transform algorithm, the signal is segmented according to the frequency range by using the parallel computing technology, and the segmented signal block is generated by spectrum analysis; Based on the original cable fault test signal, a fast Fourier transform processing is performed, a numpy.fft.fft() function is used, a sampling frequency parameter fs=1000Hz is set, a time series signal is converted to a frequency domain, and frequency components are analyzed to generate frequency domain signal data; Based on the frequency domain signal data, a parallel computing technology is used, a multiprocessing library is used to process the data, the signal is segmented according to a preset frequency range, the processing speed is optimized, and multiple frequency components are captured to generate a frequency segmented signal block; Based on the frequency segmented signal block, a spectrum analysis is performed, a scipy.signal.spectrogram() function is used, a window function is set to hann and an NFFT parameter is set to 256, key frequency components and energy distribution of each segmented block are identified and recorded, signal characteristics are analyzed, and a segmented signal block is generated.

4. The online cable fault test signal separation method of claim 1, wherein: Based on the segmented signal block, a principal component analysis method is used to extract features and reduce dimensions of multiple signal blocks, key frequency components, energy distribution, and amplitude feature vectors of the signal are counted, and the specific steps of generating a signal feature set are as follows; Based on the segmented signal block, a statistical analysis is performed, mean(), std(), and amax() in the numpy library are used to calculate the mean, standard deviation, maximum value, and minimum value of each signal block, and basic statistical features are extracted from the time series to generate basic signal feature data; Based on the basic signal feature data, principal component analysis is used to reduce dimensions and extract key features, data is projected to the direction with the largest variance, the dimension of the data set is reduced, and key information is retained to generate a reduced dimension signal data set; Based on the reduced dimension signal data set, a statistical analysis is performed on the key frequency components, energy distribution, and amplitude feature vectors of the signal, frequency components are counted through vector operation, an energy distribution diagram is drawn, and the energy distribution characteristics of the signal are displayed to generate a signal feature set.

5. The online cable fault test signal separation method of claim 1, wherein: Based on the signal feature set, a support vector machine classification model is used to learn and train the signal features, a hyperplane is constructed to identify the pattern of the fault signal, and the fault signal is detected and classified according to the training result to generate a fault type classification result. The specific steps are as follows: Based on the signal feature set, a kernel function of the support vector machine algorithm is selected, the command item is SVC(kernel=rbf, C=1, gamma=scale), the kernel parameter specifies the type of kernel function, C is the penalty parameter, and gamma is the coefficient of the kernel function. By changing the mapping dimension of the original feature space, kernel function configuration parameters are generated; Based on the kernel function configuration parameters, a cross-validation and grid search method is used to optimize the parameters of the vector machine model, including the command item GridSearchCV(SVC(), param_grid={C: [0.1, 1, 10], gamma: [0.001, 0.0001]}, cv=5), the model complexity and fitting degree are balanced, and a model with optimized parameters is generated. Based on the parameter-optimized model, the model training and fault signal classification are performed using the fit method and the predict method, and the command items include optimizedSVM.fit(X_train, y_train) and optimizedSVM.predict(X_test), a classification hyperplane distinguishing the cable fault types is constructed, and a fault type classification result is generated.

6. The online cable fault test signal separation method of claim 1, wherein: Based on the fault type classification result, a fixed threshold decision method is used to perform secondary segmentation of the signal, and the fault part and the non-fault part of the signal are separated according to the preset threshold, and the specific steps of the separated signal are as follows: Based on the fault type classification result, a fixed threshold decision method is used to perform segmentation of the signal, a target threshold is set, and the normal signal and the fault signal are distinguished according to the statistical signal characteristics and the comparison of the preset threshold, and an initial segmentation signal is generated. Based on the initial segmentation signal, the time sequence characteristics of the signal are analyzed, the fault interval is refined by using the moving window technology, the size and step of the moving window are set, the signal is traversed in a sliding window manner, the preset threshold is applied for judgment, and a refined segmentation signal is generated. Based on the refined segmentation signal, the signal reconstruction technology is used to filter the fault signal based on the preset threshold, reduce the interference on the non-fault part, and separate the fault part and the non-fault part of the signal, and a separated signal is generated.

7. The online cable fault test signal separation method of claim 1, wherein: Based on the separated signal, a dynamic time warping method is used for time sequence analysis, the similarity between time sequences is measured and adjusted, the signal is matched with a pre-defined signal pattern library, and the specific steps of the matching and similarity analysis result are as follows: Based on the separated signal, a dynamic time warping algorithm is used to construct a cost matrix, a command item numpy.linalg.norm is used to calculate the Euclidean distance, a parameter ord=2 is set, and the cost matrix is filled, and an initial cost matrix is generated. Based on the initial cost matrix, a dynamic programming algorithm is used to calculate the cumulative distance and the backtracking path, a numpy.min function is used to capture the path with the minimum cumulative cost, a parameter axis is set to specify the search direction of the minimum value, and an optimal matching path is generated. Based on the optimal matching path, the dynamic time warping algorithm is used again to calculate the similarity coefficient between the time sequence and each pattern, match with the pre-defined signal pattern library, and evaluate the similarity, and a matching and similarity analysis result is generated.

8. The online cable fault test signal separation method of claim 1, wherein: Based on the matching and similarity analysis result, a long short-term memory network model is used to analyze the time sequence characteristics and pattern changes of the signal, identify the real-time abnormal signal, and predict the behavior, and by comparing the historical data and the real-time data, the specific steps of the abnormal signal identification result are as follows: Based on the matching and similarity analysis results, the input layer dimension is set to the number of signal features, the number of hidden layer nodes is set to 128, and the number of output layer nodes corresponds to the number of predicted classifications. A long short-term memory network model is constructed, and training parameters are set, including a learning rate lr=0.001, a batch size batch_size=32, and an iteration number epochs=100. An initialized model architecture is generated; Based on the initialized model architecture, an Adam optimizer is applied to iteratively optimize the model parameters, minimize the cross-entropy loss function CrossEntropyLoss, and optimize the accuracy of the model in classifying abnormal signals. A trained and optimized model is generated; Based on the trained and optimized model, real-time signal data is input into the model and forward propagation is performed. The model is used to predict the abnormal state of the signal. By comparing the output value of the model with the preset threshold, the signal is classified according to the abnormal level, and an abnormal signal recognition result is generated.

9. The online cable fault test signal separation method of claim 1, wherein: Based on the abnormal signal recognition result, an empirical mode decomposition method is applied to decompose the signal into intrinsic mode functions. According to the basic components and variation characteristics of the signal, optimization processing and noise suppression are performed. The specific steps for generating the optimized signal features are as follows: Based on the abnormal signal recognition result, an empirical mode decomposition method is used to decompose the signal step by step through an adaptive process, and the signal is decomposed into multiple intrinsic mode functions and residual terms, including the identification of extreme points, the construction of envelope lines, and the selection criteria for intrinsic mode functions. The intrinsic mode decomposition result is generated. Based on the intrinsic mode decomposition result, a Hilbert-Huang transform method is used for time-frequency analysis to identify noise components and eliminate high-frequency noise through energy analysis methods. A noise-suppressed signal is generated. Based on the noise-suppressed signal, window function smoothing and frequency filtering techniques are used to smooth the signal and optimize the signal features. By adjusting the statistical and spectral characteristics of the signal, the key features in the signal are highlighted. Optimized signal features are generated.

10. An online cable fault test signal separation system characterized by, The online cable fault test signal separation method according to any one of claims 1-9 is executed, and the system comprises a frequency domain conversion module, a frequency division module, a feature extraction module, a classification model training module, a signal segmentation module, a signal matching module, an abnormal signal recognition module, and a signal optimization module; The frequency domain conversion module converts time series signals to the frequency domain based on the original cable fault test signal using the fast Fourier transform algorithm, analyzes the frequency components, and generates frequency domain signal data; The frequency division module divides the signal according to the preset frequency range based on the frequency domain signal data using parallel computing technology, optimizes the processing speed, and captures multiple frequency components to generate frequency division signal blocks; The feature extraction module extracts key features based on the frequency division signal blocks using principal component analysis, including statistical signal key frequency components, energy distribution, and amplitude feature vectors, to generate a reduced dimension signal data set; The classification model training module trains and learns a model based on the dimensionality reduction signal dataset using a support vector machine classification model, constructs a classification hyperplane that distinguishes cable fault types, and generates a fault type classification result; The signal segmentation module uses a fixed threshold decision method to perform secondary segmentation on the signal based on the fault type classification result, separates the fault part and the non-fault part, and generates a separated signal; The signal matching module uses a dynamic time warping method to match the signal to a pre-defined signal pattern library based on the separated signal, and generates a matching and similarity analysis result; The abnormal signal recognition module uses a long short-term memory network model to identify and predict the behavior of real-time abnormal signals based on the matching and similarity analysis result, and generates an abnormal signal recognition result; The signal optimization module applies an empirical mode decomposition method to optimize and suppress noise of the signal based on the abnormal signal recognition result, including signal decomposition, noise component identification and elimination, and signal reconstruction operation, and generates an optimized signal feature.

Images