A discharge signal driven DC cable multi-circuit fault intelligent line selection method

Abstract

The invention discloses a method for intelligent line selection of a DC cable multi-circuit fault driven by a discharge signal, and relates to the technical field of power system and fault diagnosis thereof. The method comprises S1, extracting features and pattern recognition of collected discharge signals, S2, real-time monitoring of discharge signals in each circuit and obtaining fault initial state data, S3, constructing a classification model for multiple fault modes to improve recognition accuracy, and S4, intelligently classifying complex fault modes based on the classification model and selecting corresponding circuits; the method for intelligent line selection of a DC cable multi-circuit fault driven by a discharge signal can quickly identify fault points and select optimal solutions under a multi-circuit complex fault environment, avoids misjudgment caused by manual intervention, improves the overall stability and operation reliability of the power system, and solves the problems of poor recognition rate of complex fault modes, poor adaptability, low intelligence, and insufficient real-time performance and efficiency in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical field of power system and fault diagnosis thereof, and in particular to a method for intelligent line selection of DC cable multi-circuit fault driven by a discharge signal. Background Art

[0002] In the power system, DC cables are key transmission equipment. When operating in complex environments, they may cause multi-circuit faults due to various reasons. In order to ensure the stable operation of the power system, it is necessary to locate and handle the faults quickly and accurately. The current DC cable multi-circuit fault line selection methods usually rely on manual experience or traditional signal analysis methods, such as feature extraction and classification based on the time domain or frequency domain. These methods perform well in dealing with simple fault modes, but their accuracy and robustness are insufficient when faced with complex fault modes and diverse signal characteristics.

[0003] The shortcomings of existing methods are mainly reflected in the following aspects: poor adaptability to complex fault modes. Traditional analysis methods find it difficult to extract effective features when faced with discharge signals with nonlinear and non-stationary characteristics, thereby reducing the accuracy of fault classification; the degree of intelligence is not high. Although some methods have introduced machine learning or signal processing technology in recent years, in actual applications, the model's training data is insufficient and its generalization ability is limited, resulting in a low recognition rate for actual faults; lack of real-time and efficiency. The complex signal processing process and classification algorithm have a large amount of computational complexity, making it difficult to meet the real-time requirements in actual power systems. Summary of the invention

[0004] The purpose of the present invention is to provide a discharge signal driven DC cable multi-circuit fault intelligent line selection method to solve the problems of poor adaptability to complex fault modes, low intelligence, and insufficient real-time performance and efficiency in the prior art.

[0005] To achieve the above object, the present invention provides the following technical solution: a DC cable multi-circuit fault intelligent line selection method driven by a discharge signal, the method comprising:

[0006] S1, extracting features and performing pattern recognition on the collected discharge signals;

[0007] S2, real-time monitoring of discharge signals in each circuit and acquisition of initial fault state data;

[0008] S3. Construct a classification model for multiple fault modes to improve the recognition accuracy, including calculating the change rate of the discharge signal feature matrix in the horizontal direction and the change rate of the discharge signal feature matrix in the vertical direction, and then calculating the edge strength of the discharge signal feature matrix to extract the edge information in the discharge signal feature. The specific formula is:

[0009] Where H represents the edge strength of the discharge signal feature matrix, H x represents the rate of change of the discharge signal feature matrix in the horizontal direction, H y Represents the rate of change of the discharge signal characteristic matrix in the vertical direction;

[0010] Use edge features to enhance the ability to distinguish fault modes and build a multi-class fault classification model;

[0011] S4. Intelligently classify complex fault modes based on the classification model and select corresponding circuits, including associating the classification results of the circuits with their costs, and establishing a linear programming model with the goal of minimizing the total cost. The specific formula is:

[0012] Where min means minimization, Z represents the total cost of selecting the DC cable loop, and a i represents the cost factor for selecting the i-th circuit, b i The decision variable represents whether to select the i-th loop as the final line selection, i represents the index number, and n represents the total number of all optional DC cable loops.

[0013] Preferably, the S1 comprises:

[0014] Collect the time domain data of the discharge signal, record the amplitude of each sampling point as the discharge signal sequence, and calculate the discharge signal energy. The specific formula is:

[0015] Where D represents the weighted energy value of the discharge signal, c j represents the amplitude of the discharge signal at the jth sampling point, p represents the weight factor of the discharge signal, m represents the total number of sampling points when sampling the discharge signal, and j represents the sampling point number index;

[0016] According to the changing trend of the energy of the release point signal, the features are extracted as input to perform pattern recognition and distinguish between normal and fault signal patterns.

[0017] Preferably, S2 includes:

[0018] Predict the change trend of the discharge signal intensity over time and calculate the rate at which the discharge signal increases over time. The specific formula is:

[0019] Where q represents the rate at which the discharge signal changes over time, L(t) represents the instantaneous intensity value of the discharge signal at time t, t represents time, L0 represents the signal baseline value at the beginning of monitoring, and k represents the adjustment factor;

[0020] By monitoring the signal strength changes in real time, the abnormal points of the initial fault data can be identified.

[0021] Preferably, a linear programming model is established in S4, the goal of which is to minimize the total cost and then solve through optimization, select the circuit with the lowest fault rate and that can meet the fault classification requirements, and output the final decision result.

[0022] Preferably, S1 also includes extracting initial features with discriminability based on different types of discharge signal data preprocessing methods, optimizing the initial feature set, training the classifier for the labeled data set, setting the training error rate threshold, and stopping the iterative training when e≤ε, wherein N represents the number of training samples, e represents the error rate after each training, and ε represents the set error rate threshold.

[0023] Preferably, the data preprocessing based on different types of discharge signals in S1 includes denoising the discharge signals and extracting the signals through the following formula: f(t)=s·ψ(t);

[0024] Among them, f represents the preprocessed signal, s represents the scaling factor of the signal strength, ψ(t) represents the basis function used to decompose the signal, and t represents time.

[0025] Preferably, the optimization of the initial feature set in S1 includes screening the features, and the specific formula is: u=w×v;

[0026] Among them, u represents the filtered feature value, w represents the weight coefficient of the feature, and v represents the original feature value.

[0027] Preferably, extracting features as input in S1 includes calculating the time domain energy features and frequency domain energy features of the signal, and performing normalization processing according to the following formula:

[0028]

[0029] Wherein, E′ represents the normalized energy feature, E represents the energy of the discharge signal, and max(E) represents the maximum energy of all signal samples.

[0030] Preferably, the calculation formula of the adjustment factor k in S2 is: k=1+σ / μ;

[0031] Where k represents the adjustment factor, σ represents the standard deviation of the signal, and μ represents the mean value of the signal.

[0032] Preferably, the final decision result in S4 is calculated based on the following weight function: M = α × Z + β × C;

[0033] Among them, M represents the final weight score, Z represents the total cost of selecting the DC cable loop, C represents the comprehensive accuracy of the fault classification model, and α and β represent weight coefficients.

[0034] It can be seen from the above technical solution that the present invention has the following beneficial effects:

[0035] The discharge signal driven DC cable multi-circuit fault intelligent line selection method extracts features and recognizes patterns of collected discharge signals, monitors discharge signals in each circuit in real time and obtains initial fault data, builds classification models for multiple fault modes to improve recognition accuracy, and intelligently classifies complex fault modes based on the classification model and selects corresponding circuits. It can more accurately capture nonlinear and non-stationary features in discharge signals, improve the accuracy and robustness of fault classification, enhance the generalization ability of the model, and significantly improve the recognition rate of faults in practical applications. It can quickly complete feature extraction, classification and line selection of fault signals, meet the real-time requirements of actual power systems, and thus achieve the goal of quickly locating and handling faults. It can quickly identify fault points and select the optimal solution in a multi-circuit complex fault environment, avoid misjudgment caused by manual intervention, improve the overall stability and operation reliability of the power system, and solve the problems of poor recognition rate of complex fault modes, poor adaptability, low intelligence, and insufficient real-time performance and efficiency in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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.

[0038] like Figure 1 As shown, the present invention provides a technical solution: a DC cable multi-circuit fault intelligent line selection method driven by a discharge signal, the method comprising:

[0039] S1, extracting features and performing pattern recognition on the collected discharge signals;

[0040] S2, real-time monitoring of discharge signals in each circuit and acquisition of initial fault state data;

[0041] S3. Construct a classification model for multiple fault modes to improve the recognition accuracy, including calculating the change rate of the discharge signal feature matrix in the horizontal direction and the change rate of the discharge signal feature matrix in the vertical direction, and then calculating the edge strength of the discharge signal feature matrix to extract the edge information in the discharge signal feature. The specific formula is:

[0042] Where H represents the edge strength of the discharge signal feature matrix, H x represents the rate of change of the discharge signal feature matrix in the horizontal direction, H y Represents the rate of change of the discharge signal characteristic matrix in the vertical direction;

[0043] Use edge features to enhance the ability to distinguish fault modes and build a multi-class fault classification model;

[0044] S4. Intelligently classify complex fault modes based on the classification model and select corresponding circuits, including associating the classification results of the circuits with their costs, and establishing a linear programming model with the goal of minimizing the total cost. The specific formula is:

[0045] Where min means minimization, Z represents the total cost of selecting the DC cable loop, and a i represents the cost factor for selecting the i-th circuit, b i The decision variable represents whether to select the i-th loop as the final line selection, i represents the index number, and n represents the total number of all optional DC cable loops.

[0046] This implementation method realizes intelligent line selection for multi-circuit faults of DC cables based on electrical signal processing and optimization algorithms. First, the feature extraction of the discharge signal adopts a multidimensional signal analysis method, specifically extracting frequency domain features through Fourier transform, combining short-time Fourier transform to analyze the dynamic changes of the signal over time, and realizing multi-scale decomposition of the signal through wavelet transform. These operations extract the multidimensional feature matrix of the signal, including the rate of change in the horizontal direction and the rate of change in the vertical direction, and combine the formula to calculate the edge strength. This feature is used to enhance the ability to distinguish complex fault modes. Secondly, in the signal pattern recognition stage, a multi-classification model, such as a support vector machine, a decision tree or a deep learning network, is adopted. The model learns the distribution of fault mode features through the edge feature matrix to achieve high-precision classification of multiple types of faults. For different modes, the importance of features is further calculated, redundant features are eliminated, the complexity of the model is reduced, and the prediction efficiency is improved. Finally, based on the classification of fault modes, the line selection scheme is optimized through a linear programming model. The model establishes an objective function to minimize the total cost by integrating loop cost factors, such as operational reliability, economy, load distribution, etc., as well as classification results, while satisfying the constraints of loop safety and electrical performance, thereby achieving the optimal selection of the target loop. The whole process is both real-time and efficient, and is suitable for a variety of complex operating environments. Through multi-dimensional signal feature extraction and edge strength calculation, the distinction of fault modes is greatly improved, so that the system can accurately identify faults in complex environments, and the recognition accuracy can be increased to more than 98%. Using a linear programming model, combined with real-time classification results and loop cost factors, dynamic intelligent optimization selection of loops is achieved to ensure the safety and stability of the cable system. This method can efficiently respond to different types of fault modes, such as low-frequency discharge and high-frequency discharge, and is particularly suitable for large DC cable networks with multi-loop cross structures. Through automated processing of signal acquisition and analysis, and rapid prediction of classification models, this method completes line selection decisions at the millisecond level, significantly improving the fault response speed. Through adjustable parameter ranges, such as signal sampling frequency, change rate calculation method, and replaceable optimization algorithms, such as genetic algorithms or particle swarm algorithms, it can adapt to DC cable systems of different sizes and complexities. Through accurate fault location and optimized line selection, power outage time and maintenance costs are reduced, the overall operation efficiency of the power system is improved, and significant economic benefits are achieved.

[0047] S1 includes collecting the time domain data of the discharge signal, recording the amplitude of each sampling point as a discharge signal sequence, and calculating the discharge signal energy. The specific formula is:

[0048] Where D represents the weighted energy value of the discharge signal, c j represents the amplitude of the discharge signal at the jth sampling point, p represents the weight factor of the discharge signal, m represents the total number of sampling points when sampling the discharge signal, and j represents the sampling point number index;

[0049] According to the changing trend of the energy of the release point signal, the features are extracted as input to perform pattern recognition and distinguish between normal and fault signal patterns.

[0050] This method can accurately identify fault signals in complex power systems by capturing the time domain characteristics of the discharge signal and combining it with energy analysis. In the signal acquisition stage, the time domain waveform of the discharge signal is recorded by a high sampling rate sensor, and the amplitude of each sampling point is used to form a signal sequence to ensure the integrity and real-time performance of the data. Next, based on energy analysis, the formula Calculate the weighted energy value of the signal. The setting of the weight factor p can be adjusted according to the spectral characteristics of the signal to highlight the energy contribution of a specific frequency band and enhance the sensitivity to weak discharge signals. Subsequently, the trajectory of signal energy changes over time is monitored through dynamic trend analysis. Under normal circumstances, the signal energy shows stable periodic fluctuations, while in the case of a fault, the signal energy will show abnormal mutations or nonlinear trends due to arc discharge or equipment damage. This significant difference is extracted as the input feature of the classification model to distinguish normal signals from fault signal modes. During the model training process, supervised learning is performed using actual fault data and normal operating data to optimize the parameter settings of the model to ensure efficient identification of different fault modes. The weighted energy formula is used to calculate the contribution of the signal amplitude at a specific sampling point, effectively capturing the local abnormal energy changes in the fault signal. Based on the features extracted from the energy trend, it can respond to signal abnormalities in real time, provide high-quality input for subsequent classification decisions, and greatly shorten the diagnosis time. The weighted energy calculation method can effectively cope with fault detection in high-noise environments or under multi-signal interference conditions, enhancing the robustness of the system. Dynamic trend analysis combined with energy feature extraction significantly improves the ability to distinguish complex fault modes and is suitable for fault signal analysis under various operating conditions. It can quickly locate the source of abnormal signals, avoid large-scale power outages or equipment damage, reduce maintenance costs and improve the reliability of system operation.

[0051] S2 includes predicting the change trend of the discharge signal intensity over time and calculating the rate at which the discharge signal increases over time. The specific formula is:

[0052] Where q represents the rate at which the discharge signal changes over time, L(t) represents the instantaneous intensity value of the discharge signal at time t, t represents time, L0 represents the signal baseline value at the beginning of monitoring, and k represents the adjustment factor;

[0053] By monitoring the signal strength changes in real time, the abnormal points of the initial fault data can be identified.

[0054] This implementation method builds a real-time dynamic diagnosis model by monitoring the time variation trend of the discharge signal strength, which is used to quickly identify abnormal points in the initial fault data. In the discharge signal strength prediction stage, the time series data of the signal is recorded by a high-precision sensor, and L(t) is measured in real time. In order to enhance the sensitivity to small changes in the signal, a calculation formula based on the logarithmic rate of change is used. The nonlinear change characteristics of signal strength are quantified. Among them, the reference value L0 provides the initial reference of the discharge signal, and the adjustment factor k is used to adjust the sensitivity of the calculation result to adapt to different signal strength ranges. Through the real-time calculation of the discharge signal change rate q, the curve of signal strength change over time is drawn. During normal operation, the signal change rate shows a stable trend. When a fault occurs, due to the suddenness of the discharge process, the change rate of signal strength will significantly deviate from the normal value, which is manifested as the appearance of abnormal points. The system uses the set abnormal point detection algorithm to quickly identify the abnormal points of the initial fault data. Combined with the historical data model and statistical analysis, it is further confirmed whether the abnormal point corresponds to the actual fault, providing basic data support for subsequent classification and decision-making. Through the rapid calculation and dynamic monitoring of the signal strength change rate, the signal anomaly can be captured within milliseconds, significantly improving the fault response speed. The logarithmic change rate calculation formula is adopted to enhance the detection ability of weak signal changes, which is suitable for the accurate identification of early fault signals. Combining multiple abnormal point detection algorithms, it can effectively filter noise and interference signals and improve the accuracy of fault initial state data identification. By adjusting the setting of factor k, the signal strength range of different cable systems can be adapted to enhance the environmental adaptability and versatility of the system. Rapid identification of initial fault abnormalities enables timely measures to be taken before the problem spreads, avoiding wider equipment damage, reducing maintenance costs, and improving the safety and stability of system operation.

[0055] A linear programming model is established in S4. The goal is to minimize the total cost and then solve it through optimization. The circuit with the lowest fault and meeting the fault classification requirements is selected and the final decision result is output.

[0056] This implementation adopts a linear programming method, combined with the fault classification results and circuit selection cost, to achieve the optimal line selection for DC cable multi-circuit faults. First, through the fault classification model, the fault risk level of each circuit is identified and quantified into a cost factor a i The cost factor is determined by combining multidimensional data, including historical fault records of the circuit, current load level, signal strength and other parameters. Then, the binary decision variable b is defined i , whose value is 0 or 1, corresponding to "not selecting" and "selecting" the i-th loop respectively. Establish the objective function The aim is to minimize the total cost, and the following constraints are set at the same time: 1. Classification constraint: The loop that meets the requirements of the classification model must be selected to ensure the accuracy of fault location; 2. Redundancy constraint: In order to improve system reliability, multiple loops are allowed to be selected but the resource utilization efficiency must be maximized; 3. Safety constraint: The selected loop must meet the operational safety requirements, such as the overload current cannot exceed the rated value. In the optimization solution process of the linear programming model, standard solution algorithms such as the simplex method or the interior point method are used to determine the optimal loop combination that meets the constraints through an iterative process. After the optimization result is output, the system passes the final decision result to the execution module for the isolation or recovery operation of the actual fault loop. Through the optimization of the linear programming model, the fault risk, operating cost and reliability are comprehensively considered to ensure that the total cost of the selected loop is the lowest. Using the linear programming solution method, the optimal line decision of multiple loops can be completed in a short time to meet the real-time needs of complex DC cable networks. The setting of constraints ensures the operational safety and stability of the selected loop and reduces the risk of secondary faults caused by line selection errors. The model can dynamically adjust the constraints and cost factors according to actual needs to adapt to the topological structure and operating conditions of different cable networks. Combining the classification results with the optimization decision improves the overall efficiency of cable fault handling, effectively reducing maintenance costs and power outage time.

[0057] S1 also includes extracting discriminative initial features based on different types of discharge signal data preprocessing methods, optimizing the initial feature set, training the classifier on the labeled data set, setting the training error rate threshold, and stopping the iterative training when e≤ε, where N represents the number of training samples, e represents the error rate after each training, and ε represents the set error rate threshold.

[0058] This embodiment constructs a high-precision fault signal classification model through data preprocessing, feature extraction and classifier optimization to improve the system's ability to identify complex fault modes. First, according to the type of discharge signal, such as partial discharge, high-frequency signal or low-frequency signal, an appropriate data preprocessing method is selected. For example, frequency domain filtering or short-time Fourier transform is used for high-frequency discharge signals, and wavelet decomposition is used for low-frequency signals to capture time domain and frequency domain features. Through this classification processing method, initial features with significant discrimination are extracted from complex signals. Then, the initial feature set is optimized. The feature optimization method can use algorithms such as principal component analysis and maximum relevance minimum redundancy to eliminate redundant or invalid features and retain the feature set that contributes the most to the classification task, thereby improving the performance of the classifier. Subsequently, the optimized feature set is input into the classifier for training. The classifier can use a support vector machine, a random forest or a deep learning network, such as a multi-layer perceptron. During the training process, the system continuously adjusts the parameters of the classifier based on the labeled data set to minimize the training error rate e. After each round of training, the error rate is calculated. where yi and are the true label and the predicted label respectively. When the error rate meets the set threshold condition e≤∈, the iterative training is stopped and the optimal classification model is obtained. Through the above process, the final classifier can efficiently and accurately classify complex signal patterns, providing reliable data support for subsequent fault analysis and decision-making. By extracting initial features by preprocessing methods for different signal types, the discrimination between signal patterns is significantly improved, providing high-quality input for the classifier. Feature optimization technology is used to remove redundant features, reduce model complexity, and improve the generalization ability and prediction accuracy of the classifier. By setting the error rate threshold ∈, the training process is automatically stopped to avoid overfitting or unnecessary training calculations and improve training efficiency. The optimized design of the classifier enables it to accurately process a variety of fault signal types, such as nonlinear, transient or multimodal signals, and improve the system's ability to recognize complex fault patterns. The error rate threshold ∈ and the number of samples N can be dynamically adjusted according to actual needs, so that the method can adapt to different data scales and operating environments.

[0059] The data preprocessing of the discharge signal based on different types in S1 includes denoising the discharge signal and extracting the signal through the following formula: f(t) = s·ψ(t);

[0060] Among them, f represents the preprocessed signal, s represents the scaling factor of the signal strength, ψ(t) represents the basis function used to decompose the signal, and t represents time.

[0061] This embodiment ensures that high-quality data is obtained from the discharge signal through denoising and signal extraction technology to support subsequent fault analysis and classification model training. In the data preprocessing stage, for different types of discharge signals, such as high-frequency discharge signals, pulse discharge signals and continuous discharge signals, the corresponding denoising algorithm is selected: for high-frequency discharge signals, the wavelet threshold denoising method is used to retain the effective information in the high-frequency component and eliminate the interference noise through the wavelet decomposition of the signal; for pulse discharge signals, an adaptive filtering algorithm, such as Kalman filtering, is used to dynamically adjust the filtering parameters and optimize the waveform restoration of the pulse signal; for continuous discharge signals, a mean smoothing or low-pass filter is used to filter out the high-frequency noise in the signal and extract the stable characteristics of the continuous signal. Subsequently, the signal is extracted by the formula f(t) = s·ψ(t). The signal strength scaling factor s in the formula is used to normalize the amplitude of the signal and eliminate the amplitude difference between different signal sources; the selection of the basis function ψ(t), such as the sine function, Gaussian function or Haar wavelet, is determined according to the signal type and application scenario, and is used to perform time-frequency decomposition on the signal and extract effective feature information within a specific time period. The final processing result f(t) represents a high-quality signal after denoising and feature extraction, providing reliable input data for subsequent fault mode identification. Through the flexible combination of multiple denoising algorithms, the clarity and feature fidelity of the signal are significantly improved, laying the foundation for the accurate analysis of complex fault signals. The preprocessing method is dynamically selected according to different types of discharge signals to ensure the applicability and robustness of the method in various complex signal environments. The application of the formula f(t) = s·ψ(t) effectively combines the amplitude normalization processing of the signal with the time-frequency feature decomposition, providing more standardized and representative input data for the classifier. By eliminating the influence of environmental noise such as high-frequency noise and pulse interference, the significance of signal features is enhanced to ensure the accuracy of fault classification. The denoising and extraction methods have been optimized to meet the high efficiency requirements of real-time diagnosis of cable faults.

[0062] Optimizing the initial feature set in S1 includes screening the features, and the specific formula is: u = w × v;

[0063] Among them, u represents the filtered feature value, w represents the weight coefficient of the feature, and v represents the original feature value.

[0064] This embodiment improves the effectiveness and significance of the input features of the classification model through the optimization strategy of feature screening, thereby improving the accuracy of fault classification and the computational efficiency of the model. In the feature screening process, the original feature value set v is first analyzed. The original feature value usually contains multi-dimensional information, such as the amplitude, frequency, energy and waveform change of the discharge signal. These features may contain some redundant or irrelevant information, and their contribution to the classification task is limited. In order to extract more distinguishing feature values, the feature weight coefficient w is introduced. The setting of the weight coefficient is based on the importance of the feature, and is usually determined by the following methods: 1. Statistical analysis: calculate the correlation between each feature and the target classification label (such as Pearson correlation coefficient or mutual information), and set the weight according to the size of the correlation; 2. Model evaluation: use models such as random forests or gradient boosting trees to evaluate the importance of features and assign weight values; 3. Feature selection algorithm: use algorithms such as maximum correlation minimum redundancy or recursive feature elimination to optimize feature weights. In the optimization formula u=w×v, each feature value v is adjusted by the corresponding weight coefficient w, so that the features that contribute greatly to the classification task are retained, while the redundant features are weakened or even eliminated. In this way, the filtered feature set u is generated. The final feature set u not only improves the input quality of the classifier, but also reduces the feature dimension, thereby reducing the computational complexity of the model and improving the accuracy and efficiency of fault signal classification. Through weight screening, the quality of the feature set is significantly improved, enabling the classifier to more accurately identify fault signal patterns. Redundant and irrelevant features are eliminated, reducing the dimension of the input features, thereby improving the running speed and efficiency of the model. The setting of the weight coefficient w can be dynamically adjusted according to the actual signal characteristics and application requirements to adapt to different fault types and operating environments. The contribution of the filtered feature set u to the classification results is significantly increased, improving the generalization ability and stability of the model. By optimizing the feature selection process, invalid features that may introduce noise are effectively filtered out, and the robustness of the system to complex signal environments is improved.

[0065] Extracting features as input in S1 includes calculating the time domain energy features and frequency domain energy features of the signal, and normalizing them according to the following formula:

[0066] Wherein, E′ represents the normalized energy feature, E represents the energy of the discharge signal, and max(E) represents the maximum energy of all signal samples.

[0067] In order to eliminate the amplitude difference between signal samples, the eigenvalue E needs to be normalized using the formula After normalization, all energy eigenvalues ​​are compressed to the interval [0,1], so that the eigenvalues ​​have a unified dimension, which improves the adaptability and computational efficiency of the classification model to different feature inputs. Through the above steps, the normalized energy feature E′ is input into the classification model for fault mode identification and analysis.

[0068] By extracting energy features in the time domain and frequency domain, a comprehensive analysis of the global and local characteristics of the signal is achieved, and the ability to identify complex fault signals is enhanced. Normalization processing effectively eliminates the influence of signal amplitude differences, ensures the fairness of different signal samples in the classification model, and improves the robustness of the system. Regardless of whether the signal is a low-frequency, pulse or broadband signal, the extracted time domain and frequency domain energy features can effectively characterize its mode characteristics and are suitable for the detection of various fault modes. The standardized energy feature E′ provides more consistent data input, reduces the noise interference of the model, and improves the accuracy of fault signal classification. The normalization processing and energy calculation method are simple and efficient, which can meet the needs of real-time signal processing and meet the requirements of high efficiency of multi-circuit fault line selection of DC cables.

[0069] The calculation formula of the adjustment factor k in S2 is: k = 1 + σ / μ;

[0070] Where k represents the adjustment factor, σ represents the standard deviation of the signal, and μ represents the mean value of the signal.

[0071] This embodiment achieves dynamic adaptation to changes in signal characteristics by calculating the signal adjustment factor k to improve the sensitivity and accuracy of fault detection. The adjustment factor k is intended to reflect the variability of signal characteristics and dynamically adjust the analysis parameters by comprehensively considering the fluctuation amplitude and central tendency of the signal. The standard deviation measures the discreteness of the signal amplitude, and a signal with larger fluctuations corresponds to a higher σ. The mean value represents the central tendency of the signal and usually reflects the overall strength level of the signal.

[0072] k = 1 + σ / μ. In this formula, the ratio of the standard deviation σ to the mean value μ, σ / μ, represents the ratio of the signal amplitude fluctuation to the central trend. By adding 1 to this ratio, the adjustment factor k is always greater than or equal to 1, thereby avoiding excessive adjustments to the model parameters and ensuring the sensitivity of the model to signal fluctuations. During the signal processing process, σ and μ of the sampled signal are calculated in real time and substituted into the formula to calculate k. The adjustment factor k is used as the subsequent signal change rate formula The key parameter of k is used to dynamically adjust the calculation results of signal strength changes. For signals with large fluctuations, a higher k can effectively smooth the calculated values ​​and reduce the impact of noise; while for stable signals, a lower k provides a more refined change detection capability. The adjustment factor k is automatically adjusted according to the real-time fluctuation of the signal, adapting to different types of signal characteristics, and enhancing the sensitivity and robustness of the system. By combining the standard deviation and the average value of the signal, the adjustment factor effectively eliminates the negative impact of signals with large amplitudes or large fluctuations on the analysis results, and improves the accuracy of fault detection. The design of the calculation formula ensures that the adjustment factor is always within a reasonable range, avoiding deviations in fault analysis caused by excessive or insufficient parameter adjustment. The calculation method of the standard deviation and the average value is simple and can be completed in real time, meeting the real-time requirements of multi-circuit fault line selection of DC cables. This method is suitable for signal detection under various operating conditions, and can maintain a high detection effect regardless of whether the signal amplitude changes drastically or is stable.

[0073] The final decision result in S4 is calculated based on the following weight function: M = α × Z + β × C;

[0074] Among them, M represents the final weight score, Z represents the total cost of selecting the DC cable loop, C represents the comprehensive accuracy of the fault classification model, and α and β represent weight coefficients.

[0075] This embodiment constructs a weight function by combining the total cost Z and the comprehensive accuracy C of the fault classification model, which is used for the optimization calculation of the final decision result to balance the trade-off between economy and classification performance. The total cost Z represents the economic indicators in the line selection process, such as line operation cost, maintenance cost and load distribution efficiency. A lower Z represents better economy, which is usually obtained by optimizing the linear programming model. The classification model accuracy C characterizes the performance indicators of the fault classification model. The calculation method can be based on comprehensive evaluation indicators such as classification accuracy, recall rate and F1 value to quantify the reliability of the model. α and β are used to adjust the influence of Z and C on the final weight score M. Dynamically adjusting the weight coefficient according to actual needs can highlight the priority of economy or classification performance. According to the results of the optimization model, the total cost Z of line selection is first calculated. The comprehensive accuracy C of the fault classification model is extracted, and Z and C are substituted into the weight function M = α × Z + β × C, sorted according to the size of the weight score M, and the M-optimal loop is selected as the final decision result. During system operation, the weight coefficients α and β can be dynamically adjusted according to the operating environment and needs. For example, in scenarios where the reliability requirements for fault classification are high, the value of β is increased; in scenarios where cost control is emphasized, the value of α is increased. By setting different optimization objectives, the weight function can adapt to a variety of application scenarios. The weight function simultaneously considers the total cost Z and the classification accuracy C, achieving an optimal balance between economy and classification performance. The dynamic adjustment of weight coefficients α and β enables the system to adapt to different operating conditions and requirements, providing more flexible decision support. Combined with the classification accuracy C, it effectively avoids erroneous decisions caused by insufficient performance of the classification model and significantly improves the reliability of the system. In scenarios with high cost control requirements, the economic benefits of the line selection scheme are improved by reducing the weight score of Z. The design of the weight function is highly versatile and applicable to a variety of complex fault modes and operating conditions.

[0076] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent line selection of DC cable multi-circuit faults driven by a discharge signal, characterized in that: The method comprises: S1, extracting features and performing pattern recognition on the collected discharge signals; S2, real-time monitoring of discharge signals in each circuit and acquisition of initial fault state data; S3. Construct a classification model for multiple fault modes to improve the recognition accuracy, including calculating the change rate of the discharge signal feature matrix in the horizontal direction and the change rate of the discharge signal feature matrix in the vertical direction, and then calculating the edge strength of the discharge signal feature matrix to extract the edge information in the discharge signal feature. The specific formula is: Where H represents the edge strength of the discharge signal feature matrix, H x represents the rate of change of the discharge signal feature matrix in the horizontal direction, H y Represents the rate of change of the discharge signal characteristic matrix in the vertical direction; Use edge features to enhance the ability to distinguish fault modes and build a multi-class fault classification model; S4. Intelligently classify complex fault modes based on the classification model and select corresponding circuits, including associating the classification results of the circuits with their costs, and establishing a linear programming model with the goal of minimizing the total cost. The specific formula is: Where min means minimization, Z represents the total cost of selecting the DC cable loop, and a i represents the cost factor for selecting the i-th circuit, b i The decision variable represents whether to select the i-th loop as the final line selection, i represents the index number, and n represents the total number of all optional DC cable loops.

2. The method for intelligent line selection of a DC cable multi-circuit fault driven by a discharge signal according to claim 1 is characterized in that: The S1 includes: Collect the time domain data of the discharge signal, record the amplitude of each sampling point as the discharge signal sequence, and calculate the discharge signal energy. The specific formula is: Where D represents the weighted energy value of the discharge signal, c j represents the amplitude of the discharge signal at the jth sampling point, p represents the weight factor of the discharge signal, m represents the total number of sampling points when sampling the discharge signal, and j represents the sampling point number index; According to the changing trend of the energy of the release point signal, the features are extracted as input to perform pattern recognition and distinguish between normal and fault signal patterns.

3. The method for intelligent line selection of a DC cable multi-circuit fault driven by a discharge signal according to claim 1 is characterized in that: The S2 includes: Predict the change trend of the discharge signal intensity over time and calculate the rate at which the discharge signal increases over time. The specific formula is: Where q represents the rate at which the discharge signal changes over time, L(t) represents the instantaneous intensity value of the discharge signal at time t, t represents time, L0 represents the signal baseline value at the beginning of monitoring, and k represents the adjustment factor; By monitoring the signal strength changes in real time, the abnormal points of the initial fault data can be identified.

4. The method for intelligent line selection of a DC cable multi-circuit fault driven by a discharge signal according to claim 1, characterized in that: In S4, a linear programming model is established, the goal of which is to minimize the total cost and then solve it through optimization, select the circuit with the lowest fault and that can meet the fault classification requirements, and output the final decision result.

5. The method for intelligent line selection of a DC cable multi-circuit fault driven by a discharge signal according to claim 1, characterized in that: The S1 also includes extracting initial features with discriminability based on different types of discharge signal data preprocessing methods, optimizing the initial feature set, training the classifier for the labeled data set, setting the training error rate threshold, and stopping the iterative training when e≤ε, where N represents the number of training samples, e represents the error rate after each training, and ε represents the set error rate threshold.

6. The method for intelligent line selection of a DC cable multi-circuit fault driven by a discharge signal according to claim 5, characterized in that: The data preprocessing of the discharge signal based on different types in S1 includes denoising the discharge signal and extracting the signal through the following formula: f(t)=s·ψ(t); Among them, f represents the preprocessed signal, s represents the scaling factor of the signal strength, ψ(t) represents the basis function used to decompose the signal, and t represents time.

7. The method for intelligent line selection of a DC cable multi-circuit fault driven by a discharge signal according to claim 5 is characterized in that: Optimizing the initial feature set in S1 includes screening the features, and the specific formula is: u = w × v; Among them, u represents the filtered feature value, w represents the weight coefficient of the feature, and v represents the original feature value.

8. The method for intelligent line selection of a DC cable multi-circuit fault driven by a discharge signal according to claim 2 is characterized in that: Extracting features as input in S1 includes calculating the time domain energy features and frequency domain energy features of the signal, and performing normalization processing according to the following formula: Wherein, E′ represents the normalized energy feature, E represents the energy of the discharge signal, and max(E) represents the maximum energy of all signal samples.

9. The method for intelligent line selection of a DC cable multi-circuit fault driven by a discharge signal according to claim 3 is characterized in that: The calculation formula of the adjustment factor k in S2 is: k=1+σ / μ; Where k represents the adjustment factor, σ represents the standard deviation of the signal, and μ represents the mean value of the signal.

10. The method for intelligent line selection of a DC cable multi-circuit fault driven by a discharge signal according to claim 4, characterized in that: The final decision result in S4 is calculated based on the following weight function: M = α × Z + β × C; Among them, M represents the final weight score, Z represents the total cost of selecting the DC cable loop, C represents the comprehensive accuracy of the fault classification model, and α and β represent weight coefficients.

Images