Power cable fault detection method

Abstract

The invention provides a power cable fault detection method, which comprises the following steps of: obtaining a cable material, a laying environment and an insulation aging degree attribute parameter of each monitoring segment according to a pre-established power cable segment model, determining a correction coefficient of the segment model according to load current magnitude at different operating temperatures, and determining a power cable fault detection result according to the correction coefficient; the correction coefficient is used for adjusting the electrical parameter of each segment to adapt to the actual operation condition; in the fault positioning process, a time sequence analysis method is introduced, the dynamic change trend of current and voltage data before and after a fault occurs is considered, the fault occurrence moment is recognized through an anomaly detection algorithm, and the position coordinates of a fault point are corrected according to the difference between the cable material and the laying environment in combination with a judged fault area.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method for detecting power cable faults. Background Art

[0002] Power cable fault detection is an important guarantee for the safe and stable operation of the power system. Traditional cable fault detection methods mainly rely on manual inspections and regular testing, which have problems such as low detection efficiency and many blind spots. In order to quickly and accurately locate cable faults, a new fault location method based on segment monitoring is proposed. This method divides the power cable into several monitoring segments according to a certain length, installs monitoring devices for current, voltage and other parameters at the beginning and end of each segment, and collects the operating condition data of each segment in real time. When a cable fails, the monitoring device detects abnormal changes in the current and voltage data of the faulty segment, and uploads the abnormal data to the main station. By analyzing and comparing the monitoring data of each segment, the specific location of the fault can be determined, greatly narrowing the scope of the fault search.

[0003] However, in practical applications, due to the influence of factors such as the cable laying environment and operating conditions, there may be large differences in the monitoring data of different segments, resulting in deviations in fault location. How to achieve accurate fault location in a complex and changeable environment is the main technical difficulty faced by this method. In addition, there are various types of cable faults, such as insulation breakdown, external force damage, etc., and the abnormal current and voltage characteristics caused by different types of faults are not the same. A single segmented monitoring data is difficult to fully reflect the fault characteristics, which may lead to misjudgment of the fault type and affect the detection accuracy. Therefore, it is necessary to further study the comprehensive fault analysis method that integrates multi-source monitoring information to improve the reliability and adaptability of fault detection. Summary of the invention

[0004] The present invention provides a method for detecting a power cable fault, which mainly comprises:

[0005] S1. Data collection and parameter correction: collect the cable material, laying environment, insulation aging degree attribute parameters of each monitoring segment, determine the correction coefficient of the segment model according to the load current size under different operating temperatures, and adjust the electrical parameters of each segment to adapt to the actual operating conditions;

[0006] S2. Feature extraction: using wavelet transform algorithm to extract features from real-time collected data of each segment to obtain key parameters including fault current amplitude, fault voltage amplitude and fault duration;

[0007] S3, constructing a fault feature vector, combining the number of joints and branch line number attributes of the segmented model, as well as the cable attribute parameters and correction coefficients, to construct a multidimensional fault feature vector including electrical parameters, physical characteristics and environmental factors;

[0008] S4, fault classification, establish a fault classification model based on the support vector machine algorithm;

[0009] S5, fault location, after determining the fault type, adopt the corresponding fault location strategy;

[0010] For insulation breakdown faults, the steady-state component of the fault is analyzed, and the segment whose estimated fault distance is closest to the actual laying length is found through Euclidean distance calculation and cosine similarity calculation;

[0011] For external force damage faults, analyze the transient component of the fault and determine the most severely damaged segment area through energy distribution comparison;

[0012] S6. Time series analysis and fault point correction. The time series analysis method is introduced to consider the dynamic change trend of current and voltage data before and after the fault occurs. The time of fault occurrence is identified through the anomaly detection algorithm. Combined with the determined fault area, the location coordinates of the fault point are corrected according to the differences in cable material and laying environment.

[0013] S7, Fault verification and correction, integrating multi-source monitoring information such as fault recording and fault indicators to verify the preliminary fault location results;

[0014] When the positioning result is highly consistent with the auxiliary data, it is determined as the final output; when there is a deviation exceeding the setting, the fault location is corrected according to similar cases in the fault knowledge base;

[0015] S8. Update of fault knowledge base. The fault knowledge base contains historical fault data, typical fault characteristics and corresponding positioning results, which are used to improve positioning accuracy and reliability, and are continuously updated according to new fault cases.

[0016] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0017] The present invention discloses a method for intelligently locating power cable faults. First, a cable segmentation model is established to obtain the attribute parameters and operation correction coefficients of each cable segment. Then, wavelet transform feature extraction is performed on real-time collected data. A multi-dimensional fault feature vector is constructed in combination with the cable attributes. A fault classification model is established based on a support vector machine algorithm to distinguish insulation breakdown and external force damage faults. For different fault types, corresponding positioning strategies are adopted to analyze the steady-state or transient components of the fault, and the fault location is determined in combination with time series analysis. Finally, the positioning result is verified by integrating multi-source monitoring information, and the fault knowledge base is used for correction. The method combining multi-dimensional feature analysis and multi-strategy positioning improves the accuracy and reliability of cable fault positioning, providing effective technical support for power system operation and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] For better understanding and implementation, the technical solution of the present application is described in detail below with reference to the accompanying drawings.

[0019] Figure 1 A flow chart of a power cable fault detection method of the present invention;

[0020] Figure 2 One of the schematic diagrams of a power cable fault detection method of the present invention;

[0021] Figure 3 This is a second schematic diagram of a power cable fault detection method according to the present invention. DETAILED DESCRIPTION

[0022] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than 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.

[0023] like Figure 1-3 In this embodiment, a power cable fault detection method may specifically include:

[0024] S1. Data collection and parameter correction: collect the cable material, laying environment, insulation aging degree attribute parameters of each monitoring segment, determine the correction coefficient of the segment model according to the load current size under different operating temperatures, and adjust the electrical parameters of each segment to adapt to the actual operating conditions;

[0025] According to the pre-established power cable segmentation model, the attribute parameters such as cable material, laying environment and insulation aging degree of each monitoring segment are obtained;

[0026] Based on the obtained cable property parameters, a normal temperature and current threshold model of the cable under different operating conditions is established;

[0027] Use temperature sensors and current sensors to monitor the operating temperature and load current of each cable segment in real time, compare the monitoring data with the threshold model, and determine whether the actual operating status of the cable is abnormal. Obtain the corresponding correction coefficient from the preset correction coefficient database for the load current size at different operating temperatures;

[0028] If the actual operating conditions do not match the preset conditions, the Lagrange interpolation algorithm is used to calculate the correction coefficient under the actual conditions based on the temperature, current and correction coefficient under the preset conditions.

[0029] Cactual=Cpreset+k·ΔT+m·ΔI

[0030] C_{\mathrm{actual}} is the correction coefficient under actual working conditions, C_{\mathrm{preset}} is the correction coefficient under preset working conditions, k and m are the correction coefficients of temperature and current;

[0031] The correction coefficient obtained or calculated is multiplied by the electrical parameters such as resistivity and dielectric constant in the cable segment model to obtain the corrected electrical parameters that adapt to the actual operating conditions. According to the corrected electrical parameters, the cable insulation aging model is used to calculate the insulation aging indicators such as insulation loss factor and leakage current of each monitoring segment, and compare them with the preset warning threshold to determine whether the cable insulation status meets the safety operation requirements;

[0032] Specifically, if the insulation aging index exceeds the warning threshold, an early warning signal is triggered to prompt the operation and maintenance personnel to inspect or replace the cable. The actual operating parameters of the cable are associated with the correction coefficient and other data and stored. The least squares method is used to optimize and update the correction coefficient database to improve the accuracy of the correction coefficient.

[0033] At the same time, the support vector machine algorithm is used to train the historical operation data, and a cable insulation aging trend prediction model is established with the cable operation time, load rate, ambient temperature, etc. as input and the insulation aging index as output. The trained insulation aging prediction model is used to predict the change trend of the cable insulation aging index in the next month according to the current cable operation status, and the model is visualized in the form of a curve chart. If the predicted aging index exceeds the warning threshold at some point in the future, an early warning message will be sent to the operation and maintenance personnel by text message or email, providing a decision-making basis for formulating cable maintenance or replacement plans.

[0034] Specifically, the cable condition monitoring and prediction system can effectively evaluate the health of the cable and predict future aging trends, thereby avoiding major accidents;

[0035] First, a cable segmentation model needs to be established. For example, a 10-kilometer-long cable can be divided into 10 1-kilometer-long segments, and the material (such as copper core or aluminum core), laying environment (such as direct underground burial, overhead or tunnel) and initial insulation aging degree of each segment are recorded. This is equivalent to establishing a "file" for each cable segment to record its basic information. Next, a threshold model is established based on the cable attribute parameters. Taking copper core, direct underground burial cable as an example, it is assumed that its normal operating temperature threshold is 70°C and the current threshold is 200A. These thresholds can be determined based on the cable specifications, standards and historical operating data. Cables of different materials and laying environments will have different thresholds. Then, the operating status of the cable is monitored in real time, and temperature sensors and current sensors are installed on each segment. For example, a sensor is installed at a certain distance to transmit the collected temperature and current data to the monitoring center.

[0036] Assuming that the temperature of a segment is 60°C and the current is 150A, by comparing it with the preset threshold, it can be judged that the segment is operating normally. In actual operation, the working condition of the cable may not match the preset working condition. Assuming that the preset working condition is 60°C and the current is 150A, the correction factor is 1.0, and the actual working condition is 65°C and the current is 180A. At this time, it is necessary to find the correction factor that matches the actual working condition from the correction factor database;

[0037] If there is no corresponding data in the database, the Lagrange interpolation algorithm can be used to calculate the correction coefficient. For example, if there are data of temperature 60°C, current 150A, correction coefficient 1.0 and temperature 70°C, current 200A, correction coefficient 1.1 in the database, these two data points can be used to calculate the correction coefficient when the temperature is 65°C and the current is 180A through the Lagrange interpolation algorithm. For example, the calculation result is 1.06. This can more accurately reflect the electrical parameters of the cable under actual working conditions.

[0038] After obtaining the correction coefficient, multiply it with the electrical parameters such as resistivity and dielectric constant in the cable segment model to obtain the corrected electrical parameters;

[0039] For example, suppose the resistivity of a segment is 0.018Ω·mm 2 / m, the correction factor is 1.06, then the corrected resistivity is 0.019Ω·mm 2 / m, the insulation status of the cable can be more accurately evaluated using the corrected electrical parameters. According to the corrected electrical parameters, the insulation aging model of the cable is used to calculate insulation loss factor and leakage current and other insulation aging indicators. For example, the insulation loss factor of a certain segment is calculated to be 0.05 and the leakage current is 1mA. These indicators are compared with the preset warning thresholds (such as insulation loss factor 0.1, leakage current 5mA). If the threshold is exceeded, an early warning signal is triggered to prompt the operation and maintenance personnel to perform maintenance or replacement.

[0040] In order to improve the accuracy of the correction coefficient, it is necessary to continuously optimize the correction coefficient database. For example, the least squares method can be used to fit the actual operating data and update the correction coefficient in the database. At the same time, the actual operating parameters of the cable are associated with the correction coefficient and other data for subsequent analysis and optimization. In order to predict the future aging trend of the cable, the support vector machine algorithm can be used to train the historical operating data and establish an insulation aging trend prediction model. The input parameters may include the cable operating time, load rate, ambient temperature, etc. The output parameter is the insulation aging index. For example, based on the operating data of the past year, the insulation loss factor and leakage current of the cable are predicted in the next month, and the trend is visualized in the form of a curve chart; if the predicted indicator exceeds the warning threshold at some point in the future, an early warning message will be sent to the operation and maintenance personnel via SMS or email so that maintenance or replacement plans can be made in advance;

[0041] Through the above steps, the cable status monitoring and prediction system can realize real-time monitoring, accurate evaluation and trend prediction of the cable status, thereby ensuring the safe and stable operation of the cable and avoiding significant losses caused by cable failures.

[0042] S2. Feature extraction: using wavelet transform algorithm to extract features from real-time collected data of each segment to obtain key parameters including fault current amplitude, fault voltage amplitude and fault duration;

[0043] S3, constructing a fault feature vector, combining the number of joints and branch line number attributes of the segmented model, as well as the cable attribute parameters and correction coefficients, to construct a multidimensional fault feature vector including electrical parameters, physical characteristics and environmental factors;

[0044] According to the preset sampling frequency, the current and voltage data collected in real time in each segment are sampled to obtain discrete time domain signals. The discrete time domain signals are preprocessed to remove power frequency interference and white noise to improve signal quality. The discrete time domain signals are decomposed into multiple scales using the wavelet transform algorithm to obtain high-frequency components reflecting fault characteristics and obtain wavelet coefficients.

[0045] Analyze the amplitude distribution characteristics of wavelet coefficients, determine the amplitude parameters of fault current and voltage by setting thresholds, and obtain the fault duration by threshold judgment. According to the line segment where the fault occurs, obtain the model, length, and cross-sectional area parameters of the cable segment from the pre-established cable attribute database. According to the cable attribute parameters, determine the correction coefficients of the electrical parameters of resistivity and inductance to correct the calculation results of the fault parameters. The fault current amplitude, voltage amplitude, duration, number of joints, and number of branch lines are integrated to construct a multidimensional fault feature vector. The support vector machine algorithm is used to train and classify the fault feature vector to determine the fault type; at the same time, the decision tree algorithm is used to estimate the fault location and complete the fault location. In the algorithm training stage, a training sample set is constructed based on historical operation data and simulation data, and the model parameters are optimized to improve the accuracy and reliability of fault diagnosis.

[0046] Specifically, the core goal of cable fault diagnosis is to quickly and accurately determine the fault type and locate the fault position, so as to ensure the safe and stable operation of the power system. In order to achieve this goal, the collected cable current and voltage signals need to be analyzed and processed. First, the collected current and voltage analog signals need to be digitized. For example, the sampling frequency is set to 10kHz, that is, 10,000 data points are collected per second, and the continuous analog signal is converted into a discrete digital signal for computer processing. However, the collected signals are often mixed with power frequency interference and white noise. These interference signals will affect the extraction of fault characteristics. Therefore, the signal needs to be preprocessed. For example, a digital filter is used to remove 50Hz power frequency interference, and a wavelet threshold denoising method is used to remove white noise, thereby improving signal quality.

[0047] The preprocessed signal can be used for fault feature extraction. Wavelet transform is a powerful signal analysis tool that can decompose the signal into sub-signals of different frequency bands. For example, by using the db4 wavelet basis to decompose the signal into 5 layers, 5 high-frequency components and 1 low-frequency component can be obtained. Since fault signals usually appear as high-frequency oscillations, we can focus on the wavelet coefficients of the high-frequency components and analyze the amplitude distribution characteristics of the wavelet coefficients. For example, the threshold is set to 0.5. When the amplitude of the wavelet coefficient exceeds the threshold, it is considered that a fault has occurred.

[0048] By counting the number of wavelet coefficients exceeding the threshold, the amplitude parameters of the fault current or voltage can be determined. At the same time, by recording the time interval exceeding the threshold, the fault duration can be determined. In order to calculate the fault parameters more accurately, the attribute parameters of the cable itself need to be considered. For example, suppose the fault occurs on a cable of model YJV-10kV-3×240, with a length of 500 meters and a cross-sectional area of ​​240 square millimeters. According to the cable model and specifications, its resistivity can be queried from the database to be 0.018Ω·mm2 / m, inductance is 0.1mH / km;

[0049] Considering the influence of the actual laying environment and operating status, it is necessary to correct the resistivity and inductance. For example, assuming the correction factor is 1.05, the corrected resistivity is 0.0189Ω·mm 2 / m, and the inductance is 0.105mH / km. The corrected electrical parameters can be used to more accurately calculate the impedance of the fault circuit, thereby improving the accuracy of fault parameter calculation. In order to more comprehensively describe the fault characteristics, it is necessary to combine multiple parameters to construct a fault feature vector. For example, the fault current amplitude, voltage amplitude, duration, number of joints, number of branch lines and other attributes are combined into a multidimensional vector, such as [1000A, 5kV, 0.1s, 3, 2]. This vector contains rich fault information and can be used for subsequent fault diagnosis.

[0050] Finally, the machine learning algorithm is used to classify and locate the fault feature vectors. For example, the support vector machine (SVM) algorithm is used to train and classify the fault feature vectors to identify the fault type, such as short circuit, grounding, etc. At the same time, the decision tree algorithm is used to estimate the fault location. For example, according to the fault feature vector, it is judged that the fault occurred 200 meters away from the starting point. In order to improve the accuracy and reliability of diagnosis, a large amount of historical operation data and simulation data is needed to train and optimize the model. For example, various types of fault data that occurred in the past year, as well as fault data generated by simulation software, are collected to construct a training sample set, and the model parameters are continuously adjusted to improve the generalization ability of the model.

[0051] S4, fault classification, establish a fault classification model based on the support vector machine algorithm;

[0052] The original data of the cable fault is obtained, including the signal data such as the current and voltage when the fault occurs, as well as the parameters such as the ambient temperature and humidity when the fault occurs. The obtained original data is preprocessed by using digital signal processing technology, and high-frequency noise interference is removed by low-pass filtering. The key parameters reflecting the fault characteristics are extracted by short-time Fourier transform, including the amplitude, frequency and energy distribution of the fault current, and the fault feature vector is constructed. According to the fault current amplitude, frequency, energy distribution and the attribute parameters such as ambient temperature and humidity extracted from the fault feature vector, the fuzzy C-means clustering algorithm is used to preliminarily classify the fault samples, and several fuzzy clustering centers representing different fault types are obtained. Based on the results of fuzzy clustering, a multi-class support vector machine algorithm is further used to construct an accurate fault classification model. The radial basis kernel function is used to map the samples to a high-dimensional feature space. The optimal classification hyperplane is determined by solving the quadratic programming problem to achieve effective distinction between different types of faults.

[0053] When the Euclidean distance between the feature vector of the fault sample to be judged and the insulation breakdown fault cluster center is the smallest and less than the preset threshold, the fault is judged as an insulation breakdown fault based on the support vector machine classification model; when the Euclidean distance between the feature vector of the fault sample to be judged and the external force damage fault cluster center is the smallest and less than the preset threshold, the fault is judged as an external force damage fault based on the support vector machine classification model. The K-fold cross-validation method is used to evaluate the performance of the support vector machine classification model. The data set is randomly divided into K subsets of similar size, and one of the subsets is selected as the test set each time, and the remaining K-1 subsets are used as training sets. K training and testing are performed, and the average of the K results is taken as the performance indicator of the model. The grid search optimization algorithm is used to optimize the hyperparameters of the support vector machine, such as the kernel function type, kernel function parameters, and penalty coefficient, to improve the accuracy and generalization ability of the fault classification model.

[0054] Specifically, the key to cable fault diagnosis is to accurately identify the fault type and precisely locate the fault location to ensure the safe and stable operation of the power system. To achieve this goal, it is necessary to collect and analyze the current and voltage signals when the fault occurs, as well as parameters such as ambient temperature and humidity; first, obtain the original fault data. For example, in a 10kV distribution system, a cable fault occurs. The sensors installed on the cable line can collect the three-phase current and voltage waveform data when the fault occurs in real time, and simultaneously record the temperature at 25°C and humidity at 70%; next, pre-process the collected original data. Since the original signal often contains high-frequency noise interference, which will affect the extraction of fault features, it is necessary to use digital signal processing technology, such as low-pass filters, to remove these noises. Assuming that a low-pass filter with a cutoff frequency of 1kHz is used to filter the collected current and voltage signals, the noise interference above 1kHz can be effectively removed, and the frequency band where the fault features are located can be retained; then, the fault features are extracted. In order to extract the key parameters reflecting the fault features, methods such as short-time Fourier transform can be used. For example, by performing short-time Fourier transform on the filtered current signal, the frequency and energy distribution of the fault current signal in different time periods can be obtained.

[0055] Assume that the analysis found that at the moment of the fault, the amplitude of the current signal suddenly increased to 5 times the normal value, reaching 5000A. At the same time, there was an obvious energy concentration in the high-frequency part, indicating the presence of high-frequency oscillation components. The fault feature vector was constructed based on the extracted characteristic parameters. For example, a multidimensional fault feature vector can be constructed with parameters such as the fault current amplitude of 5000A, the fault frequency of 500Hz (assuming that the fault frequency is mainly concentrated around 500Hz), energy distribution characteristics (for example, high-frequency energy accounts for 80%), ambient temperature of 25℃ and humidity of 70%: [5000,500,0.8,25,70], and then the preliminary classification of the fault is carried out. The fuzzy C-means clustering algorithm can be used to preliminarily classify the fault samples. Assuming that the fault types are divided into three categories: insulation breakdown, external force damage and others, three cluster centers can be obtained through fuzzy C-means clustering, representing the three fault types respectively. On the basis of the preliminary classification, an accurate fault classification model is constructed;

[0056] Here, a multi-class support vector machine algorithm can be used. Assuming that the radial basis kernel function is used, an optimal classification hyperplane can be obtained through training to distinguish different types of fault samples. The support vector machine can effectively handle nonlinear separable problems by mapping samples to high-dimensional feature space, thereby improving the accuracy of fault classification. To determine the fault type, assume that there is a fault sample to be determined, and its feature vector is [4800, 490, 0.75, 26, 72]. If the Euclidean distance between the vector and the insulation breakdown fault cluster center is the smallest and less than the preset threshold (for example, the threshold is set to 100), then based on the support vector machine classification model, the fault is determined to be an insulation breakdown fault. Similarly, if the Euclidean distance between the vector and the external force damage fault cluster center is the smallest and less than the threshold, it is determined to be an external force damage fault. In order to evaluate and optimize the model, K-fold cross validation is used. For example, the data set is randomly divided into 10 parts, one of which is taken as the test set each time, and the remaining 9 are taken as the training set. Repeat 10 times and calculate the average accuracy as the performance indicator of the model. This can more comprehensively evaluate the generalization ability of the model; finally, through optimization algorithms such as grid search, the kernel function parameters, penalty coefficients and other hyperparameters of the support vector machine are optimized. For example, different kernel functions (linear kernel, polynomial kernel, radial basis kernel, etc.) and different penalty coefficients are tried to find the best hyperparameter combination, thereby improving the accuracy and generalization ability of the fault classification model. By continuously optimizing the model, the accuracy and efficiency of cable fault diagnosis can be improved, ensuring the safe and stable operation of the power system.

[0057] S5, fault location, after determining the fault type, adopt the corresponding fault location strategy;

[0058] For insulation breakdown faults, the steady-state component of the fault is analyzed, and the segment whose estimated fault distance is closest to the actual laying length is found through Euclidean distance calculation and cosine similarity calculation;

[0059] For external force damage faults, analyze the transient component of the fault and determine the most severely damaged segment area through energy distribution comparison;

[0060] According to the fault type attribute, it is determined whether it is an insulation breakdown fault or an external force damage fault, and the corresponding fault location strategy is determined. If it is an insulation breakdown fault, the steady-state component data of the cable is extracted, and the actual laying length segment closest to the estimated fault distance is found by calculating the Euclidean distance and cosine similarity of different segments. According to the Euclidean distance calculation result, the segment with the smallest Euclidean distance is obtained and it is determined as the fault segment. According to the cosine similarity calculation result, the segment with the largest cosine similarity is obtained and it is determined as the fault segment; if it is an external force damage fault, the transient component data of the cable is extracted, and the energy distribution of different segments is compared and analyzed to determine the area with the most concentrated energy distribution. According to the energy distribution comparison result, the segment corresponding to the area with the most concentrated energy is obtained and it is determined as the most severely damaged fault segment. The results of Euclidean distance, cosine similarity and energy distribution comparison are combined to cross-validate the fault segment to improve the accuracy of fault location, and finally output the fault location result.

[0061] Specifically, after the support vector machine model determines the fault type, it is necessary to locate the fault. The fault location strategy varies according to the fault type. The location strategy of insulation breakdown fault is different from that of external force damage fault. If it is an insulation breakdown fault, it is usually manifested as a change in steady-state quantity. Therefore, it is key to extract the steady-state component data of the cable. For example, the fundamental current, fundamental voltage and other data when the fault occurs can be extracted; then, the cable line is divided into several segments, and the Euclidean distance and cosine similarity between the steady-state component of each segment and the estimated distance of the fault point are calculated. For example, assuming that a 10kV cable line is divided into 5 equal-length segments, the fault distance is estimated to be 2.5 kilometers through the preliminary fault distance measurement method, and the Euclidean distance and cosine similarity between the steady-state component of each segment and the steady-state component at 2.5 kilometers are calculated respectively. Assume that the calculation results are as follows: Segment 1: Euclidean distance is 0.8, cosine similarity is 0.92, segment 2: Euclidean distance is 0.5, cosine similarity is 0.95. Segment 3: Euclidean distance is 0.2, cosine similarity is 0.98, segment 4: Euclidean distance is 0.4, cosine similarity is 0.96, segment 5: Euclidean distance is 0.7, cosine similarity is 0.93. According to the Euclidean distance calculation result, segment 3 with the smallest distance is determined to be the fault segment. According to the cosine similarity calculation result, segment 3 with the largest similarity is also determined to be the fault segment. The results of the two methods confirm each other, which improves the reliability of positioning. This is because the Euclidean distance reflects the difference between the two vectors in value, while the cosine similarity reflects the similarity between the two vectors in direction. Combining the two can more comprehensively evaluate the proximity between the segment and the fault point.

[0062] If it is a fault caused by external force, it usually manifests as a change in transient quantities, such as sudden changes in current and voltage. Therefore, it is necessary to extract the transient component data of the cable. For example, the current impact and voltage drop data when the fault occurs can be extracted. Then, the energy distribution of different segments is compared and analyzed to determine the area with the most concentrated energy distribution. For example, a 10kV cable line is also divided into 5 equal-length segments. By analyzing the transient current data of the fault, the energy distribution of each segment can be obtained. Assume that the analysis results are as follows: Segment 1: Energy share 10%, Segment 2: Energy share 15%, Segment 3: Energy share 50%, Segment 4: Energy share 15%, Segment 5: Energy share 10%. According to the energy distribution comparison results, segment 3 corresponding to the area with the most concentrated energy is judged to be the most severely damaged fault segment. This is done because External force damage usually leads to energy concentration near the fault point. The fault location can be accurately located by analyzing the energy distribution. The results of the steady-state component and transient component analysis are combined to cross-validate the fault segmentation. For example, in an insulation breakdown fault, if the Euclidean distance and cosine similarity both point to segment 3, it can be more certain that the fault occurred in segment 3. In an external force damage fault, if the energy distribution analysis also points to segment 3, the fault location can be further confirmed. Through cross-validation, the accuracy of fault location can be improved, and ultimately a reliable fault location result can be output. For example, based on the above two examples, it is finally determined that the fault occurred in segment 3, which provides precise guidance for subsequent fault repairs, thereby reducing power outage time and ensuring the stable operation of the power system.

[0063] S6. Time series analysis and fault point correction. The time series analysis method is introduced to consider the dynamic change trend of current and voltage data before and after the fault occurs. The time of fault occurrence is identified through the anomaly detection algorithm. Combined with the determined fault area, the location coordinates of the fault point are corrected according to the differences in cable material and laying environment.

[0064] The current and voltage time series data within a certain time range before and after the fault occurs are obtained, and a multidimensional time series feature vector is constructed. The isolation forest algorithm is used for anomaly detection. By calculating the anomaly score of each data point, it is determined whether the data point is an anomaly point. The anomaly score of the anomaly point will be significantly higher than that of the normal point. Therefore, a threshold can be set according to the anomaly score, and the point exceeding the threshold is identified as the fault point, so as to determine the time when the fault occurs. According to the result of anomaly detection, the time range of the anomaly is determined, and the time range is used as the fault area. The material parameters and laying environment parameters of the cable in the area, including the resistivity, inductance, capacitance, thermal conductivity, etc. of the cable, as well as factors such as ambient temperature and humidity, are obtained. These parameters are used as features, and the coordinates of the fault point are used as labels to construct a training data set. The gradient boosting decision tree (GBDT) algorithm is used to establish a mapping model between the cable material and laying environment parameters and the fault point location. GBDT combines the prediction results of multiple weak learners (decision trees) and continuously iterates and optimizes to obtain a strong learner.

[0065] The model is adjusted using grid search to find the optimal model parameters. The coordinates of the fault point initially determined in the fault area are input into the trained GBDT model. The model will correct the coordinates of the fault point according to the differences in cable material and laying environment parameters. The cable material parameters reflect the electrical conductivity and insulation performance of the cable, and the laying environment parameters reflect the temperature, humidity and other factors of the cable environment. These factors will affect the location of the fault point. The GBDT model corrects the coordinates of the initially determined fault point by learning the relationship between these parameters and the location of the fault point to obtain a more accurate fault location. The corrected coordinates of the fault point are visualized, the location of the fault point is marked on the cable line map, and the mileage of the fault point is calculated based on the distance between the fault point and the starting point of the cable. The fault location results, the time of the fault, the fault area and other information are integrated to generate a fault diagnosis report, which provides a basis for subsequent fault repair and cause analysis.

[0066] Specifically, the isolation forest algorithm is an anomaly detection algorithm based on the assumption that abnormal data is more likely to be isolated. It isolates data points by constructing multiple isolation trees. In the process of constructing the isolation tree, a feature is randomly selected, and a split point is randomly selected within the value range of the feature to divide the data into two parts. This process is repeated until each data point is isolated or the preset tree height is reached. The path length of anomalies is usually shorter than that of normal points. Therefore, the degree of abnormality can be determined by calculating the average path length of each data point. For example, suppose there is a set of current data, most of which have values ​​of about 10A, but some data points have values ​​exceeding 100A. These data points exceeding 100A are more likely to be quickly isolated in the process of constructing the isolation tree, so their path length will be shorter and the anomaly score will be higher, so they are identified as abnormal points. The advantage of this is that the isolation forest algorithm is more efficient in processing high-dimensional data and data sets containing a lot of noise.

[0067] The current and voltage time series data within a certain time range before and after the fault occurs are obtained. For example, the current and voltage data within 1 second before and after the fault occurs can be obtained with a sampling frequency of 1kHz. In this way, 2000 data points can be obtained. These data points are constructed into a multidimensional time series feature vector. For example, the current value and voltage value at each time point can be used as a two-dimensional vector. Assuming that the current value at a certain time point is 10A and the voltage value is 10kV, the feature vector of the time point is (10, 10000). The anomaly score of each data point is calculated by the isolation forest algorithm. For example, assuming that the anomaly score of a data point is 0.8, and the anomaly scores of other data points are between 0.1 and 0.2, the data point can be judged as an anomaly point. A threshold value, such as 0.5, is set to identify the point exceeding the threshold as a fault point. This can effectively identify the time when the fault occurs, because when a fault occurs, parameters such as current and voltage usually change suddenly, resulting in an increase in the anomaly score. According to the result of anomaly detection, the time range of the anomaly is determined, for example, from the 1000th data point to the 1100th data point. The time range is taken as the fault area, and the material parameters and laying environment parameters of the cable in the fault area are obtained. For example, the resistivity of the cable is 1.7×10^-8Ω·m, the inductance is 1mH / km, the capacitance is 0.1μF / km, the thermal conductivity is 0.9W / (m·K), the ambient temperature is 25℃, and the humidity is 60%. These parameters are used as features, and the coordinates of the fault point are used as labels to construct a training data set. For example, the fault data of multiple cables can be collected, and each data contains the material parameters of the cable, the laying environment parameters, and the coordinates of the fault point. The gradient boosting decision tree (GBDT) algorithm is used to establish a mapping model between the cable material and laying environment parameters and the fault point location. The GBDT algorithm obtains the final prediction result by combining the prediction results of multiple decision trees. Each decision tree will fit a part of the data. The GBDT algorithm continuously iterates and uses the residual of the previous decision tree as the training target of the next decision tree, thereby continuously improving the accuracy of the model. The advantage of this is that the GBDT algorithm can handle nonlinear relationships and has a certain robustness to noise data.

[0068] Use grid search to adjust the model parameters and find the optimal model parameters. For example, you can adjust the number of decision trees, learning rate, maximum depth and other parameters. By comparing the model performance under different parameter combinations, select the parameter combination with the best performance, and input the coordinates of the fault point preliminarily determined in the fault area into the trained GBDT model. The model will correct the coordinates of the fault point according to the differences in cable material and laying environment parameters. For example, assuming that the coordinates of the fault point preliminarily determined are (100, 200), after correction by the GBDT model, the coordinates of the fault point become (105, 195). This can improve the accuracy of fault location, because the cable material and laying environment parameters will affect the location of the fault point. The corrected coordinates of the fault point are visualized, the location of the fault point is marked on the cable line map, and the mileage of the fault point is calculated based on the distance between the fault point and the cable starting point. For example, assuming that the coordinates of the fault point are (105, 195) and the coordinates of the cable starting point are (0, 0), the distance between the fault point and the cable starting point is √(1052+1952)=221.9m, and the mileage of the fault point is 221.9 meters. The fault location results, fault occurrence time, fault area and other information are integrated to generate a fault diagnosis report. For example, the fault diagnosis report can include the following content: the fault occurrence time is 10:00:00 on October 27, 2023, the fault area is from the 1000th data point to the 1100th data point, and the mileage of the fault point is 221.9 meters. This can provide a basis for subsequent fault repair and cause analysis.

[0069] S7, Fault verification and correction, integrating multi-source monitoring information such as fault recording and fault indicators to verify the preliminary fault location results;

[0070] When the positioning result is highly consistent with the auxiliary data, it is determined as the final output; when there is a deviation exceeding the setting, the fault location is corrected according to similar cases in the fault knowledge base;

[0071] S8. Update of fault knowledge base. The fault knowledge base contains historical fault data, typical fault characteristics and corresponding positioning results, which are used to improve positioning accuracy and reliability, and are continuously updated according to new fault cases.

[0072] First, data is acquired from the monitoring equipment and preprocessed, including cleaning, denoising and format unification.

[0073] The preprocessed data will be used for the next step of fault location. The triangulation positioning algorithm is used to calculate the preliminary position coordinates of the fault based on the preprocessed data. Subsequently, auxiliary data such as the equipment operating status and environmental information are introduced to compare with the preliminary fault location and calculate the degree of fit. If the degree of fit is higher than the preset threshold, the preliminary positioning result is the final fault location; otherwise, the pre-established fault feature vector library is used to calculate the similarity between the current fault feature and the feature in the library, and several historical cases with the highest similarity are selected. Based on the positioning results of these cases, a simple weighted average method is used to correct the preliminary fault location to obtain the final fault location coordinates. Finally, the corrected fault location is output as the final positioning result.

[0074] Specifically, monitoring equipment, such as smart sensors, can collect data such as current, voltage, and temperature on the line in real time. These raw data usually contain noise, missing values, and inconsistent formats, and need to be preprocessed. Data cleaning can remove outliers and noise. For example, filtering algorithms can be used to remove high-frequency noise in current and voltage data, and missing values ​​can be filled using interpolation. Format unification refers to converting data collected by different devices into the same format, for example, aligning data with different timestamps. For example, if there is a spike in the current data collected by a sensor that obviously deviates from the normal range, the spike can be judged as noise and removed. After preprocessing, the data can be used for subsequent fault location analysis.

[0075] The triangulation positioning algorithm is a commonly used fault location method. Its principle is to determine the location of the fault point based on the distances from multiple monitoring points to the fault point. Suppose there are three monitoring points A, B, and C, and their distances to the fault point are dA, dB, and dC, respectively. Draw a circle with A, B, and C as the center and dA, dB, and dC as the radius. The intersection of the three circles is the location of the fault point. For example, suppose the coordinates of the three monitoring points A, B, and C are (0, 0), (100, 0), and (50, 87), respectively, and their distances to the fault point are 50, 50, and 87, respectively. Then the coordinates of the fault point can be calculated to be (50, 0). It should be noted that the accuracy of the triangulation positioning algorithm is affected by the location of the monitoring points and the measurement accuracy.

[0076] In addition to using the data from monitoring equipment to locate faults, auxiliary data such as equipment operating status and environmental information can also be introduced to improve positioning accuracy. For example, information such as the load rate of the transformer, ambient temperature, and humidity may be related to the occurrence of a fault. The degree of agreement between the auxiliary data and the preliminary fault location can be calculated. For example, a model can be established based on historical data to predict the location where faults are more likely to occur under specific ambient temperatures and load rates. If the preliminary positioning results match the results predicted by the model, the confidence in the preliminary positioning results can be increased.

[0077] For example, if the initial positioning result is near a transformer, and the transformer load rate is very high and the ambient temperature is also high, these conditions are likely to cause transformer failure, then the initial positioning result can be considered to be relatively reliable. On the contrary, if the degree of agreement is low, the positioning result needs to be further corrected.

[0078] If the preliminary positioning result does not match the auxiliary data, it can be corrected using the pre-established fault feature vector library. The fault feature vector library contains various types of fault features. For example, different types of faults will cause different changes in current and voltage waveforms. By calculating the similarity between the current fault feature and the features in the library, the historical case most similar to the current fault can be found.

[0079] For example, assuming that the feature vector of the current fault is (1, 2, 3), and the fault feature vector library contains feature vectors of three historical cases, namely (1, 2, 4), (2, 3, 4), and (4, 5, 6), then the similarities between the current fault and the three historical cases can be calculated to be 0.9, 0.7, and 0.3, respectively. Select several cases with the highest similarity, and use the weighted average method to correct the preliminary fault location based on the location results of these cases. For example, assuming that the two cases with the highest similarity are selected, their location results are (100, 100) and (110, 110), and the weights are 0.8 and 0.2, respectively. The corrected fault location is (100*0.8+110*0.2, 100*0.8+110*0.2)=(102, 102).

[0080] This can improve the accuracy of fault location by using historical experience. Finally, the corrected fault location coordinates are output as the final location result. For example, the corrected fault location coordinates are (102, 102). This result can be used to guide fault troubleshooting and maintenance.

[0081] What is disclosed above is only a preferred embodiment of the present invention, which certainly cannot be used to limit the scope of rights of the present invention. A person skilled in the art can understand that all or part of the processes of the above embodiments and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A method for detecting a power cable fault, characterized in that: The method comprises: S1. Data collection and parameter correction: collect the cable material, laying environment, insulation aging degree attribute parameters of each monitoring segment, determine the correction coefficient of the segment model according to the load current size under different operating temperatures, and adjust the electrical parameters of each segment to adapt to the actual operating conditions; S2. Feature extraction: using wavelet transform algorithm to extract features from real-time collected data of each segment to obtain key parameters including fault current amplitude, fault voltage amplitude and fault duration; S3, constructing a fault feature vector, combining the number of joints and branch line number attributes of the segmented model, as well as the cable attribute parameters and correction coefficients, to construct a multidimensional fault feature vector including electrical parameters, physical characteristics and environmental factors; S4, fault classification, establish a fault classification model based on the support vector machine algorithm; S5, fault location, after determining the fault type, adopt the corresponding fault location strategy; For insulation breakdown faults, the steady-state component of the fault is analyzed, and the segment whose estimated fault distance is closest to the actual laying length is found through Euclidean distance calculation and cosine similarity calculation; For external force damage faults, analyze the transient component of the fault and determine the most severely damaged segment area through energy distribution comparison; S6. Time series analysis and fault point correction. The time series analysis method is introduced to consider the dynamic change trend of current and voltage data before and after the fault occurs. The time of fault occurrence is identified through the anomaly detection algorithm. Combined with the determined fault area, the location coordinates of the fault point are corrected according to the differences in cable material and laying environment.

2. A method for detecting a power cable fault according to claim 1, characterized in that: The cable material, laying environment, and insulation aging degree attribute parameters of each monitoring segment are collected, and the correction coefficient of the segment model is determined according to the load current size at different operating temperatures to adjust the electrical parameters of each segment, including: According to the pre-established power cable segmentation model, the cable material, laying environment and insulation aging degree attribute parameters of each monitoring segment are obtained; Based on the obtained cable property parameters, a normal temperature and current threshold model of the cable under different operating conditions is established; Use temperature sensors and current sensors to monitor the operating temperature and load current of each cable segment in real time, compare the monitoring data with the threshold model, and determine the abnormality of the actual operating status of the cable; According to the load current size under different operating temperatures, the corresponding correction coefficient is obtained from the preset correction coefficient database; If the actual operating conditions do not match the preset conditions, the Lagrange interpolation algorithm is used to calculate the correction coefficient under the actual conditions based on the temperature, current and correction coefficient under the preset conditions; CactualCpreset+k·△T+m·△I C_{\mathrm{actual}} is the correction coefficient under actual working conditions, C_{\mathrm{preset}} is the correction coefficient under preset working conditions, k and m are the correction coefficients of temperature and current; The correction coefficient obtained or calculated is multiplied by electrical parameters such as resistivity and dielectric constant in the cable segment model to obtain corrected electrical parameters that are suitable for actual operating conditions; According to the corrected electrical parameters, the cable insulation aging model is used to calculate the insulation aging indicators such as insulation loss factor and leakage current of each monitoring segment, and compare them with the preset warning threshold to determine whether the cable insulation status meets the safety operation requirements; If the insulation aging index exceeds the warning threshold, an early warning signal is triggered to prompt the operation and maintenance personnel to inspect or replace the cable; The actual operating parameters of the cable are associated with data such as correction coefficients and stored, and the least squares method is used to optimize and update the correction coefficient database to improve the accuracy of the correction coefficient.

3. A method for detecting a power cable fault according to claim 2, characterized in that: It also includes using support vector machine algorithm to train historical operation data, and establish a cable insulation aging trend prediction model with cable operation time, load rate and ambient temperature as input and insulation aging index as output; Using the trained insulation aging prediction model, according to the current cable operation status, the change trend of the cable insulation aging index in the next month is predicted and visualized in the form of a curve chart; If the predicted aging index exceeds the warning threshold at some point in the future, an early warning message will be sent to the operation and maintenance personnel via SMS or email, providing a decision-making basis for formulating cable maintenance or replacement plans.

4. A method for detecting a power cable fault according to claim 1, characterized in that: The wavelet transform algorithm is used to extract features from the real-time collected data of each segment to obtain key parameters such as fault current amplitude, fault voltage amplitude, and fault duration, including: According to the preset sampling frequency, the current and voltage data collected in real time in each segment are sampled to obtain discrete time domain signals, and the discrete time domain signals are preprocessed to remove power frequency interference and white noise to improve signal quality; The wavelet transform algorithm is used to perform multi-scale decomposition on the discrete time domain signal to obtain the high-frequency components reflecting the fault characteristics and obtain the wavelet coefficients; The amplitude distribution characteristics of the wavelet coefficients are analyzed, the amplitude parameters of the fault current and voltage are determined by setting thresholds, and the fault duration is obtained by threshold judgment.

5. A method for detecting a power cable fault according to claim 1, characterized in that: The multi-dimensional fault feature vector including electrical parameters, physical characteristics and environmental factors is constructed by combining the number of joints, the number of branch lines and the cable attribute parameters and correction coefficients of the segmented model, including: According to the line segment where the fault occurs, the model, length and cross-sectional area parameters of the cable segment are obtained from a pre-established cable attribute database; According to the cable attribute parameters, the correction coefficients of the electrical parameters of resistivity and inductance are determined to correct the calculation results of the fault parameters; The fault current amplitude, voltage amplitude, duration, number of joints and number of branch lines are integrated to construct a multi-dimensional fault feature vector.

6. A power cable fault detection method according to claim 1, characterized in that: The fault classification model is established based on the support vector machine algorithm, including: The support vector machine algorithm is used to train and classify fault feature vectors to determine the fault type; When the fault characteristic vector meets the conditions corresponding to the degree of insulation aging and the fault frequency characteristics are obvious, it is determined to be an insulation breakdown fault; when the fault characteristic vector shows a short-term large current and the fault energy distribution is concentrated, it is determined to be an external force damage fault.

7. A method for detecting a power cable fault according to claim 6, characterized in that: The method of using a support vector machine algorithm to train and classify fault feature vectors includes: Obtain the original data of the cable fault, including signal data such as current and voltage when the fault occurs, as well as parameters such as ambient temperature and humidity when the fault occurs; The acquired raw data is preprocessed using digital signal processing technology, high-frequency noise interference is removed through low-pass filtering, and key parameters reflecting fault characteristics are extracted through short-time Fourier transform, including the amplitude, frequency and energy distribution of the fault current, to construct the fault feature vector; According to the attribute parameters extracted from the fault feature vector, including fault current amplitude, frequency and energy distribution, and ambient temperature and humidity, the fuzzy C-means clustering algorithm is used to preliminarily classify the fault samples and obtain several fuzzy clustering centers representing different fault types. Based on the results of fuzzy clustering, a multi-class support vector machine algorithm is further used to build an accurate fault classification model; The radial basis kernel function is used to map samples to a high-dimensional feature space, and the optimal classification hyperplane is determined by solving the quadratic programming problem to effectively distinguish different types of faults. When the Euclidean distance between the feature vector of the fault sample to be judged and the insulation breakdown fault cluster center is the smallest and is less than a preset threshold, the fault is judged to be an insulation breakdown fault based on the support vector machine classification model; When the Euclidean distance between the feature vector of the fault sample to be judged and the external force damage fault cluster center is the smallest and is less than a preset threshold, the fault is judged to be an external force damage fault based on the support vector machine classification model; The K-fold cross-validation method is used to evaluate the performance of the support vector machine classification model. The data set is randomly divided into K subsets of similar size. One of the subsets is selected as the test set each time, and the remaining K-1 subsets are used as the training set. K training and testing are performed, and the average of the K results is taken as the performance indicator of the model. The grid search optimization algorithm is used to optimize the hyperparameters of the support vector machine, including the kernel function type, kernel function parameters and penalty coefficient, to improve the accuracy and generalization ability of the fault classification model.

8. A method for detecting a power cable fault according to claim 1, characterized in that: The method of analyzing the steady-state component of the insulation breakdown fault and finding the segment whose estimated fault distance is closest to the actual laying length through Euclidean distance calculation and cosine similarity calculation includes: Extract the steady-state component data of the cable, and find the actual laying length segment closest to the estimated fault distance by calculating the Euclidean distance and cosine similarity of different segments; According to the Euclidean distance calculation result, the segment with the smallest Euclidean distance is obtained and determined as the fault segment; According to the cosine similarity calculation result, the segment with the largest cosine similarity is obtained and determined as the fault segment; For the external force damage fault, the transient component of the fault is analyzed, and the most severely damaged segment area is determined by comparing the energy distribution, including: Extract the transient component data of the cable, and determine the area with the most concentrated energy distribution by comparing and analyzing the energy distribution of different segments; According to the energy distribution comparison result, the segment corresponding to the area with the most concentrated energy is obtained, and it is determined as the most severely damaged fault segment; The results of Euclidean distance, cosine similarity and energy distribution comparison are combined to cross-validate the fault segmentation, improve the accuracy of fault location, and finally output the fault location result.

9. A method for detecting a power cable fault according to claim 1, characterized in that: The time series analysis method is introduced to consider the dynamic change trend of current and voltage data before and after the fault occurs, identify the time of fault occurrence through anomaly detection algorithm, and correct the position coordinates of the fault point according to the determined fault area and the differences in cable material and laying environment, including: Obtain the current and voltage time series data within a certain time range before and after the fault occurs, and construct a multi-dimensional time series feature vector; The isolation forest algorithm is used for anomaly detection. By calculating the anomaly score of each data point, it is determined whether the data point is an outlier. The anomaly score of an abnormal point will be significantly higher than that of a normal point. Therefore, a threshold can be set based on the anomaly score, and points exceeding the threshold can be identified as fault points, thereby determining the time when the fault occurs; According to the results of anomaly detection, determine the time range in which the anomaly occurs and use this time range as the fault area; Obtain the material parameters and laying environment parameters of the cables in the area, including the resistivity, inductance, capacitance and thermal conductivity of the cables, as well as factors such as ambient temperature and humidity; These parameters are used as features and the fault point coordinates as labels to construct a training dataset; The gradient boosting decision tree algorithm is used to establish a mapping model between cable material and laying environment parameters and fault point locations; GBDT combines the prediction results of multiple weak learners and continuously iterates and optimizes to obtain a strong learner; Use grid search to adjust the model parameters and find the optimal model parameters; The coordinates of the fault point preliminarily determined in the fault area are input into the trained GBDT model. The model will correct the coordinates of the fault point according to the differences in cable material and laying environment parameters. The cable material parameters reflect the cable's conductivity and insulation properties, and the laying environment parameters reflect factors such as the temperature and humidity of the cable's environment, all of which will affect the location of the fault point; The GBDT model corrects the coordinates of the initially determined fault point by learning the relationship between these parameters and the fault point location to obtain a more accurate fault location; Visualize the corrected coordinates of the fault point, mark the location of the fault point on the cable line map, and calculate the mileage of the fault point based on the distance between the fault point and the starting point of the cable; Integrate the fault location results, fault occurrence time, fault area and other information to generate a fault diagnosis report, providing a basis for subsequent fault repair and cause analysis.

10. A power cable fault detection method according to claim 1, characterized in that: Also includes: S7, Fault verification and correction, integrating multi-source monitoring information such as fault recording and fault indicators to verify the preliminary fault location results; When the positioning result is highly consistent with the auxiliary data, it is determined as the final output; when there is a deviation exceeding the setting, the fault location is corrected according to similar cases in the fault knowledge base; First, data is obtained from monitoring equipment and preprocessed, including cleaning, denoising and format unification; The preprocessed data will be used for the next step of fault location; Calculate the preliminary location coordinates of the fault based on the preprocessed data using a triangulation positioning algorithm; Subsequently, auxiliary data such as equipment operating status and environmental information are introduced to compare with the preliminary fault location and calculate the degree of agreement; If the degree of agreement is higher than the preset threshold, the preliminary positioning result is the final fault location; Otherwise, the pre-established fault feature vector library is used to calculate the similarity between the current fault feature and the features in the library, and several historical cases with the highest similarity are selected; Based on the positioning results of these cases, a simple weighted average method is used to correct the preliminary fault location and obtain the final fault location coordinates; Finally, the corrected fault location is output as the final positioning result; S8. Update of fault knowledge base. The fault knowledge base contains historical fault data, typical fault characteristics and corresponding positioning results, which are used to improve positioning accuracy and reliability, and are continuously updated according to new fault cases.

Images