An intelligent online method for distinguishing feeder terminal faults

Abstract

The present application provides an intelligent online feeder terminal fault identification method, which first collects power parameters, switch states and environmental data of each feeder terminal, converts these data into feature vectors through a feature extraction algorithm, and then uses a pre-trained abnormal probability recognition model to generate an abnormal probability value and calculate its confidence interval. If the lower limit of the confidence interval exceeds a preset threshold, the terminal is judged to be abnormal. For the abnormal terminal, the method determines its adjacent terminals according to the power grid topology structure, and then uses a pre-trained fault classification model to classify the feature vectors of the abnormal terminal and the adjacent terminal respectively to obtain their respective fault classification data. Finally, the method comprehensively considers the fault classification data of the abnormal terminal and the adjacent terminal, generates a comprehensive fault classification result of the abnormal terminal, and realizes intelligent online identification of feeder terminal faults, which helps to improve the accuracy and efficiency of power grid fault diagnosis.

Description

Technical Field

[0001] The present application relates to the electric power field, and in particular to an intelligent feeder terminal fault online determination method. Background Art

[0002] Feeder terminals are key equipment in the power distribution system. They are installed at the end of the distribution line to monitor and control the operating status of the distribution line. They have many important functions, including voltage and current measurement, fault detection, remote control, and data communication. The main function of the feeder terminal is to monitor the operating parameters of the distribution line in real time, detect and locate faults in a timely manner, and improve power supply reliability and operating efficiency.

[0003] However, the existing feeder terminals still have some shortcomings in feeder fault identification. Traditional feeder terminals mainly rely on basic electrical quantities such as current and voltage to judge faults. This method is often difficult to accurately identify and locate faults in complex distribution network environments. In particular, for transient faults, high-resistance faults, or situations where multiple faults coexist, existing technologies often find it difficult to make accurate judgments. In addition, the fault identification algorithm of traditional feeder terminals is relatively simple, and it is impossible to make full use of advanced technologies such as big data and artificial intelligence to improve the accuracy and intelligence level of fault identification.

[0004] These deficiencies limit the intelligent development of the distribution system and affect the further improvement of power supply reliability. Therefore, the development of a new generation of feeder terminals with higher intelligence and stronger fault identification capabilities has become an important direction of current power system research. By introducing advanced signal processing technology, artificial intelligence algorithms, and big data analysis methods, the fault identification accuracy of feeder terminals can be further improved to cope with the increasingly complex distribution network environment and achieve a more intelligent and reliable power supply. Summary of the invention

[0005] The present application provides an intelligent feeder terminal fault online judgment method, comprising the following steps:

[0006] A1, respectively obtaining power parameter data, switch status data and environmental data of each preset feeder terminal within a preset monitoring time window;

[0007] A2, extracting features using a preset feature extraction algorithm based on the power parameter data, switch status data and environmental data to obtain feeder terminal feature vector data;

[0008] A3, generating corresponding abnormal probability value data through a pre-trained feeder abnormal probability recognition model according to the feeder terminal feature vector data;

[0009] A4, calculating the corresponding abnormal probability confidence interval according to the abnormal probability value data and the preset confidence level and determining the corresponding lower limit value of the abnormal probability confidence interval;

[0010] A5, if the lower limit value of the abnormal probability confidence interval of the feeder terminal is greater than the preset abnormal threshold, the corresponding feeder terminal is defined as an abnormal feeder terminal;

[0011] A6, determining a corresponding adjacent feeder terminal according to the abnormal feeder terminal and preset grid topology information;

[0012] A7, generating fault classification data according to the feeder terminal feature vector data of the abnormal feeder terminal using a pre-trained fault classification model and defining the data as abnormal terminal fault classification data;

[0013] A8, generating fault classification data according to the feeder terminal feature vector data of the adjacent feeder terminal corresponding to the abnormal feeder terminal using a pre-trained fault classification model and defining the data as adjacent terminal fault classification data;

[0014] A9, generating abnormal terminal fault comprehensive classification data based on the abnormal terminal fault classification data and the adjacent terminal fault classification data.

[0015] By adopting the above technical scheme, the intelligent feeder terminal fault online identification method can realize the intelligent online identification of feeder faults through the comprehensive application of multi-dimensional data collection, feature extraction and machine learning models. By analyzing the characteristics of adjacent terminals in combination with the power grid topology structure, the accuracy and reliability of fault identification are improved. By further generating fault classification results, fault diagnosis reference information is provided for operation and maintenance personnel, which is helpful to quickly locate and handle faults and improve the safety and stability of power grid operation.

[0016] Optionally, the feature extraction algorithm comprises the following steps:

[0017] B1, generating corresponding power standardization data and environmental standardization data according to the power parameter data and the environmental data using a preset data standardization algorithm;

[0018] B2, generating feature vector data according to the sampling time combination of each sample in the power standardized data, switch state data and environmental standardized data;

[0019] B3, generating reduced dimension feature vector data using a preset feature dimension reduction algorithm according to the feature vector data;

[0020] B4, define the dimension-reduced feature vector data as the feeder terminal feature vector data.

[0021] By adopting the above technical solution, the intelligent feeder terminal fault online identification method can effectively integrate multi-angle data through steps such as data standardization, time series combination and feature dimension reduction, which can not only eliminate the scale differences between different types of data, but also capture the temporal relationship between data. At the same time, the dimension reduction algorithm is used to make the final generated feature data more concise and effective, which not only retains the key information of the original data, but also improves the training efficiency of subsequent models.

[0022] Optionally, the intelligent feeder terminal fault online identification method further includes the following steps for training and generating a feeder abnormality probability recognition model:

[0023] C1, obtaining corresponding historical feeder fault information from a preset power history database;

[0024] C2, obtaining the corresponding feeder fault time from each historical feeder fault information;

[0025] C3, determine the corresponding data backtracking time window according to the feeder fault time and the monitoring time window;

[0026] C4, obtaining historical power parameter data, historical switch status data and historical environmental data of the corresponding feeder terminal from the power history database according to the data backtracking time window;

[0027] C5, performing feature extraction using a feature extraction algorithm based on historical power parameter data, historical switch state data, and historical environment data to obtain corresponding historical feeder terminal feature vector data;

[0028] C6, generates anomaly recognition training data set based on all historical feeder terminal feature vector data and corresponding anomaly label value combinations;

[0029] C7, training a preset first machine learning model according to the abnormality identification training data set to generate a feeder abnormality probability identification model.

[0030] By adopting the above technical solution, the intelligent feeder terminal fault online identification method can obtain feeder fault information and corresponding historical data through the historical power database to train the model to ensure the generalization ability and accuracy of the model, and through the combination of historical feeder terminal feature vector data and corresponding abnormal label values, it can effectively construct an abnormal recognition training data set, and then train an effective recognition model, so that the generated feeder abnormality probability recognition model can more accurately identify potential feeder terminal abnormalities, thereby improving the fault diagnosis efficiency and reliability of the system.

[0031] Optionally, the intelligent feeder terminal fault online identification method further includes the following steps for determining the data backtracking time window:

[0032] D1, determine the corresponding half-value of the time window length according to the monitoring time window;

[0033] D2, determine the corresponding data backtracking start time according to the feeder fault time minus half the time window length;

[0034] D3, determine the corresponding data backtracking termination time according to the feeder fault time plus half the time window length;

[0035] D4, determining the corresponding data backtracking time window according to the data backtracking start time and the data backtracking end time.

[0036] By adopting the above technical solution, the intelligent feeder terminal fault online identification method can determine the time range of data backtracking by taking half of the monitoring time window before and after the fault moment as the center. This symmetrical time window design can not only capture the precursor information before the fault occurs, but also include the evolution process after the fault occurs, thereby providing a comprehensive fault feature data view. This balanced data collection strategy is not only conducive to improving the accuracy of fault analysis, but also provides dynamic information for fault prediction and rapid response, while also optimizing the efficiency of data processing and avoiding the interference of redundant data.

[0037] Optionally, the intelligent feeder terminal fault online identification method further includes the following steps for determining an abnormal label value:

[0038] E1, determining the corresponding sampling time according to each feature vector in the historical feeder terminal feature vector data and defining it as the feature time;

[0039] E2, determining the characteristic moment distance according to the characteristic moment and the corresponding feeder fault moment;

[0040] E3, determine the corresponding abnormal label value according to the characteristic moment distance, data backtracking time window and preset label standard value.

[0041] By adopting the above technical solution, the intelligent feeder terminal fault online identification method can set different label values ​​for the feature vectors at different times, which can effectively distinguish the importance of data at different time points and highlight the key information related to the fault, thereby improving the accuracy of fault detection. It also provides high-quality label data for subsequent machine learning model training and optimization, which helps to improve the performance of the entire fault diagnosis system.

[0042] Optionally, the intelligent feeder terminal fault online identification method further includes the following steps for determining an abnormal threshold:

[0043] F1, obtains the minimum value among all abnormal label values ​​and defines it as the abnormal threshold.

[0044] By adopting the above technical solution, the intelligent feeder terminal fault online identification method can use the minimum abnormal label value of the training data when training the model as the threshold for subsequent judgment of feeder abnormalities, which can be matched with the data during training to ensure the recognition effect and accuracy of feeder abnormalities.

[0045] Optionally, the intelligent feeder terminal fault online identification method further includes the following steps for training a fault classification model:

[0046] G1, obtain corresponding fault classification information from each historical feeder fault information;

[0047] G2, generates corresponding classification label values ​​based on all fault classification information;

[0048] G3, generating a fault classification training set based on the historical feeder terminal feature vector data corresponding to each historical feeder fault information and the corresponding classification label value combination;

[0049] G4, training a preset second machine learning model according to the fault classification training set to generate a corresponding fault classification model.

[0050] By adopting the above technical solution, the intelligent feeder terminal fault online identification method can systematically utilize historical fault data to build an effective fault classification model. By combining historical fault information, feature vector data and classification labels to create a high-quality training set, it can make full use of the experience of known faults, train the model to identify various faults, and improve the accuracy and efficiency of fault diagnosis.

[0051] Optionally, the intelligent feeder terminal fault online determination method further includes the following steps for verifying fault classification information:

[0052] H1, calculate the corresponding mean eigenvector according to the eigenvector data of each historical feeder terminal;

[0053] H2, classify each mean feature vector according to the fault classification information to generate multiple classification feature vector data;

[0054] H3, calculate the corresponding classification center feature vector based on the classification feature vector data;

[0055] H4, calculate the corresponding center distance according to the classification center feature vector and each mean feature vector in the classification feature vector data;

[0056] H5: If the center distance of the mean feature vector is greater than the preset deviation threshold, the fault classification information corresponding to the mean feature vector is defined as abnormal classification information.

[0057] By adopting the above technical solution, the intelligent feeder terminal fault online identification method calculates the mean eigenvector and the classification center eigenvector and compares the distance between them to verify the fault classification information to identify abnormal or erroneous classification information. By setting the deviation threshold, an adjustable judgment standard is achieved, which can capture significant classification errors and tolerate reasonable data fluctuations, thereby improving the accuracy and reliability of the fault classification information and further improving the effectiveness of the training model.

[0058] In summary, the present application includes at least one of the following beneficial technical effects:

[0059] 1. Through the comprehensive application of multi-dimensional data collection, feature extraction and machine learning models, intelligent online identification of feeder faults can be achieved. By combining the grid topology structure to analyze the characteristics of adjacent terminals, the accuracy and reliability of fault identification are improved. By further generating fault classification results, fault diagnosis reference information is provided for operation and maintenance personnel, which helps to quickly locate and handle faults and improve the safety and stability of grid operation.

[0060] 2. Data from multiple angles can be effectively integrated through steps such as data standardization, time series combination and feature dimensionality reduction. This can not only eliminate the scale differences between different types of data, but also capture the temporal relationship between data. At the same time, the dimensionality reduction algorithm can make the final generated feature data more concise and effective, which not only retains the key information of the original data, but also improves the training efficiency of subsequent models.

[0061] 3. The feeder fault information and corresponding historical data can be obtained through the historical power database to train the model to ensure the generalization ability and accuracy of the model. The combination of historical feeder terminal feature vector data and corresponding abnormal label values ​​can effectively construct an abnormality recognition training data set, and then train an effective recognition model, so that the generated feeder abnormality probability recognition model can more accurately identify potential feeder terminal abnormalities, thereby improving the fault diagnosis efficiency and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 The present invention is a schematic diagram of the process of an intelligent feeder terminal fault online determination method. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0064] The embodiments of the present application are further described in detail below in conjunction with the drawings in the specification.

[0065] refer to Figure 1 The present invention provides an online fault identification method for an intelligent feeder terminal, which is used to determine whether the feeder line where the feeder terminal is located has abnormalities and faults based on the power data and related data collected by the intelligent feeder terminal, so as to make corresponding line adjustment actions or perform fault detection in time.

[0066] The intelligent feeder terminal fault online determination method comprises the following steps:

[0067] A1, respectively obtaining power parameter data, switch status data and environmental data of each preset feeder terminal within a preset monitoring time window;

[0068] The monitoring time window is a pre-set time window used to determine the time range for obtaining data;

[0069] The power parameter data is the power-related data collected by the feeder terminal, for example, voltage value, current value, power factor, apparent power, active power, reactive power, power factor, power factor angle and other related power data;

[0070] The switch status data is the status data of each connection switch of the corresponding feeder, for example, the disconnector position, the grounding switch position, the circuit breaker position, the section switch position, the tie switch status, etc., wherein these switch status data can be represented by 0 for disconnection and 1 for closing, so as to facilitate subsequent data processing, or -1 for disconnection and 1 for closing, so as to be suitable for different data processing methods;

[0071] Environmental data refers to the surrounding environment data collected by the feeder terminal, for example, temperature, humidity, rainfall, noise level, dust concentration and other parameter data related to the surrounding environment.

[0072] A2, extracting features using a preset feature extraction algorithm based on the power parameter data, switch status data and environmental data to obtain feeder terminal feature vector data;

[0073] The feature extraction algorithm is a pre-set algorithm used to extract features based on the original data. Commonly used feature extraction algorithms include principal component analysis, linear discriminant analysis, independent component analysis, etc.

[0074] The feeder terminal feature vector data is a data set of feature vectors generated after feature extraction of power parameter data, switch state data and environmental data.

[0075] A3, generating corresponding abnormal probability value data through a pre-trained feeder abnormal probability recognition model according to the feeder terminal feature vector data;

[0076] The feeder abnormality probability recognition model is a pre-trained model, which is used to output the corresponding abnormality probability value according to the input feature vector;

[0077] The abnormal probability value data is a collection of abnormal probability values ​​output by the feeder abnormal probability identification model according to each feature vector in the feeder terminal feature vector data.

[0078] A4, calculating the corresponding abnormal probability confidence interval according to the abnormal probability value data and the preset confidence level and determining the corresponding lower limit value of the abnormal probability confidence interval;

[0079] The confidence level is a pre-set confidence level, for example, 95%, 90%, 70%, etc.;

[0080] The abnormal probability confidence interval is the confidence interval of the abnormal probability value data calculated based on the confidence level;

[0081] The lower limit of the abnormal probability confidence interval is the lower limit of the abnormal probability confidence interval.

[0082] A5, if the lower limit value of the abnormal probability confidence interval of the feeder terminal is greater than the preset abnormal threshold, the corresponding feeder terminal is defined as an abnormal feeder terminal;

[0083] The abnormal threshold is a pre-set reference value used to determine whether the corresponding feeder terminal has an abnormal condition;

[0084] An abnormal feeder terminal is a feeder terminal that may have an abnormal condition.

[0085] A6, determining a corresponding adjacent feeder terminal according to the abnormal feeder terminal and preset grid topology information;

[0086] The power grid topology information is pre-stored distribution network topology information, which can be obtained through communication between the feeder terminal and the corresponding central station;

[0087] The adjacent feeder terminal is another feeder terminal directly connected to the abnormal feeder terminal.

[0088] A7, generating fault classification data according to the feeder terminal feature vector data of the abnormal feeder terminal using a pre-trained fault classification model and defining the data as abnormal terminal fault classification data;

[0089] The fault classification model is a pre-trained model that can output the corresponding recognition probability values ​​of each classification according to the input feature vector;

[0090] The fault classification data is a collection of classification probability values ​​output by the fault classification model according to each feature vector in the feeder terminal feature vector data of the abnormal feeder terminal;

[0091] The abnormal terminal fault classification data refers to the fault classification data of the abnormal feeder terminal.

[0092] A8, generating fault classification data according to the feeder terminal feature vector data of the adjacent feeder terminal corresponding to the abnormal feeder terminal using a pre-trained fault classification model and defining the data as adjacent terminal fault classification data;

[0093] The adjacent terminal fault classification data refers to the fault classification data of the adjacent feeder terminals.

[0094] When an abnormality or failure occurs somewhere in the distribution network, abnormal fluctuations in power signals or data will occur. This fluctuation will be transmitted along the line to both ends. By obtaining the power signals at both ends of the abnormal location, the accuracy of abnormal judgment can be further improved.

[0095] A9, generating abnormal terminal fault comprehensive classification data based on the abnormal terminal fault classification data and the adjacent terminal fault classification data;

[0096] The abnormal terminal fault comprehensive classification data is the comprehensive data of the normal terminal fault classification data and the adjacent terminal fault classification data, which can be generated by setting corresponding weights and calculating with a weighted algorithm;

[0097] Based on the comprehensive classification data of abnormal terminal faults, staff can preliminarily identify and troubleshoot the types of faults to improve the efficiency of fault repair.

[0098] Through the above steps, the intelligent feeder terminal fault online identification method can realize the intelligent online identification of feeder faults through the comprehensive application of multi-dimensional data collection, feature extraction and machine learning models. By combining the grid topology structure to analyze the characteristics of adjacent terminals, the accuracy and reliability of fault identification are improved. By further generating fault classification results, fault diagnosis reference information is provided for operation and maintenance personnel, which is helpful to quickly locate and handle faults and improve the safety and stability of grid operation.

[0099] Furthermore, the feature extraction algorithm comprises the following steps:

[0100] B1, generating corresponding power standardization data and environmental standardization data according to the power parameter data and the environmental data using a preset data standardization algorithm;

[0101] The data normalization algorithm is a pre-set algorithm used to generate corresponding normalized data based on the original data, for example, by cleaning and completing the data, and then normalizing the sample values ​​in the data to between 0 and 1 or -1 and 1, so as to facilitate subsequent model training;

[0102] The power standardized data is the standardized data generated after the power parameter data is standardized;

[0103] Environmental standardized data is standardized data generated after environmental data is standardized.

[0104] B2, generating feature vector data according to the sampling time combination of each sample in the power standardized data, switch state data and environmental standardized data;

[0105] The feature vector data is a collection of feature vectors obtained by combining the power standardized data, switch state data and environmental standardized data based on the sampling time. The samples of the power standardized data, switch state data and environmental standardized data at the same sampling time are combined to obtain the corresponding feature vectors. The collection of all feature vectors is the feature vector data.

[0106] B3, generating reduced dimension feature vector data using a preset feature dimension reduction algorithm according to the feature vector data;

[0107] The feature dimension reduction algorithm is a pre-set algorithm used to perform feature dimension reduction on feature vector data;

[0108] The reduced dimension feature vector data is the feature vector data obtained after the dimension reduction algorithm;

[0109] Because there is a certain intrinsic correlation between the various parameters of power parameter data, switch status data and environmental standardization data, feature dimensionality reduction can be performed. While ensuring that too much feature information is not lost, the dimension of the feature vector can be reduced as much as possible, which can effectively improve the efficiency of model training and subsequent recognition efficiency.

[0110] B4, defining the dimension-reduced feature vector data as feeder terminal feature vector data;

[0111] The feeder terminal feature vector data is the dimension-reduced feature vector data.

[0112] Through the above steps, the intelligent feeder terminal fault online identification method can effectively integrate multi-angle data through steps such as data standardization, time series combination and feature dimension reduction, which can not only eliminate the scale differences between different types of data, but also capture the temporal relationship between data. At the same time, the dimension reduction algorithm is used to make the final generated feature data more concise and effective, which not only retains the key information of the original data, but also improves the training efficiency of subsequent models.

[0113] Furthermore, the intelligent feeder terminal fault online identification method further includes the following steps for training and generating a feeder abnormality probability identification model:

[0114] C1, obtaining corresponding historical feeder fault information from a preset power history database;

[0115] The power history database is a pre-selected database that stores historical data collected by various feeder terminals and various fault information data;

[0116] The historical feeder fault information is the abnormality or fault information of the feeder that has occurred and been confirmed in the past, and may include information such as the fault classification information, the time when the fault occurred, and the duration of the fault.

[0117] C2, obtaining the corresponding feeder fault time from each historical feeder fault information;

[0118] The feeder fault time is the time when the feeder fault occurs in the historical feeder fault information.

[0119] C3, determine the corresponding data backtracking time window according to the feeder fault time and the monitoring time window;

[0120] The data backtracking time window is used to determine the time range for obtaining historical data.

[0121] C4, obtaining historical power parameter data, historical switch status data and historical environmental data of the corresponding feeder terminal from the power history database according to the data backtracking time window;

[0122] The historical power parameter data, historical switch status data and historical environment data are respectively data collected by the feeder terminal corresponding to the historical feeder fault information within the data backtracking time window.

[0123] C5, performing feature extraction using a feature extraction algorithm based on historical power parameter data, historical switch state data, and historical environment data to obtain corresponding historical feeder terminal feature vector data;

[0124] The historical feeder terminal feature vector data is feature vector data extracted based on historical power parameter data, historical switch state data, and historical environment data.

[0125] C6, generates anomaly recognition training data set based on all historical feeder terminal feature vector data and corresponding anomaly label value combinations;

[0126] The abnormal label value is a pre-set label value, which is used to match the feature vector to generate a feature label pair to train the corresponding model;

[0127] The abnormal label value can be set according to the requirements. For example, the abnormal label value corresponding to the feature vector in all historical feeder terminal feature vector data can be set to 1, or it can be set to a value between 0 and 1 according to different situations;

[0128] The anomaly recognition training data set is a data set of feature label pairs generated by matching anomaly label values ​​with feature vectors.

[0129] C7, training a preset first machine learning model according to the abnormality recognition training data set to generate a feeder abnormality probability recognition model;

[0130] The first machine learning model is a pre-set machine learning model, which is used to train and generate a corresponding feeder abnormality probability recognition model based on a training abnormality recognition training data set. The first machine learning model can be a logistic regression model, a support vector machine, a decision tree, a neural network or other models.

[0131] Through the above steps, the intelligent feeder terminal fault online identification method can obtain feeder fault information and corresponding historical data through the historical power database to train the model to ensure the generalization ability and accuracy of the model, and through the combination of historical feeder terminal feature vector data and corresponding abnormal label values, it can effectively construct an abnormal recognition training data set, and then train an effective recognition model, so that the generated feeder abnormality probability recognition model can more accurately identify potential feeder terminal abnormalities, thereby improving the fault diagnosis efficiency and reliability of the system.

[0132] Furthermore, the intelligent feeder terminal fault online identification method further includes the following steps for determining the data backtracking time window:

[0133] D1, determine the corresponding half-value of the time window length according to the monitoring time window;

[0134] The half value of the time window duration is half of the monitoring time window duration. For example, if the monitoring time window is set to 10 minutes, the half value of the time window duration is 5 minutes, that is, half of 10 minutes.

[0135] D2, determine the corresponding data backtracking start time according to the feeder fault time minus half the time window length;

[0136] The data backtracking start time is the time on the time axis that is half the time window length forward of the feeder fault time.

[0137] D3, determine the corresponding data backtracking termination time according to the feeder fault time plus half the time window length;

[0138] The end time of data backtracking is the time on the time axis that is half the length of the time window after the feeder fault time.

[0139] D4, determining the corresponding data backtracking time window according to the data backtracking start time and the data backtracking end time;

[0140] The data backtracking time window is the time window between the start time of the backtracking and the end time of the data backtracking.

[0141] Before an abnormality or fault occurs in a power line, there may be some potential data information revealing that a fault is about to occur. After an abnormality or fault occurs in a power line, the abnormal electrical signal generated by the abnormality or fault will continue to oscillate in the power line for a period of time, such as a fault traveling wave signal. Therefore, by acquiring data within the corresponding time range before and after the fault occurs, the characteristic information data of the fault can be fully obtained to accurately identify its abnormal state.

[0142] Through the above steps, the intelligent feeder terminal fault online identification method can determine the time range of data backtracking by taking half of the monitoring time window before and after the fault moment as the center. This symmetrical time window design can not only capture the precursor information before the fault occurs, but also include the evolution process after the fault occurs, thereby providing a comprehensive fault feature data view. This balanced data collection strategy is not only conducive to improving the accuracy of fault analysis, but also can provide dynamic information for fault prediction and rapid response, while also optimizing the efficiency of data processing and avoiding the interference of redundant data.

[0143] Furthermore, the intelligent feeder terminal fault online identification method further includes the following steps for determining an abnormal label value:

[0144] E1, determining the corresponding sampling time according to each feature vector in the historical feeder terminal feature vector data and defining it as the feature time;

[0145] The characteristic moment is the data sampling moment corresponding to each characteristic vector.

[0146] E2, determining the characteristic moment distance according to the characteristic moment and the corresponding feeder fault moment;

[0147] The characteristic time distance is the time difference between the characteristic time and the feeder fault time. For example, if the characteristic time of a certain characteristic vector is 13:30 and the feeder fault time is 13:33, the characteristic time distance is 3 minutes.

[0148] E3, determine the corresponding abnormal label value according to the characteristic moment distance, data backtracking time window and preset label standard value;

[0149] The label standard value is a pre-set value used to determine the label value corresponding to each feature vector;

[0150] The abnormal label value is the label value corresponding to each feature vector determined by calculation;

[0151] The label value corresponding to the feature vector can be determined by setting a certain calculation method. For example, the label standard value is set to 1. When the feature moment of a feature vector is the same as the fault moment, that is, the feature moment distance is 0, the label value of the feature vector is set to 1. When the feature moment distance of a feature vector is 3 minutes and the data backtracking window is 10 minutes, the corresponding abnormal label value can be 1*(10-3) / 10 = 0.7. In short, the abnormal label value decreases as the feature moment distance increases.

[0152] Through the above steps, the intelligent feeder terminal fault online identification method can set different label values ​​for the feature vectors at different times, which can effectively distinguish the importance of data at different time points and highlight the key information related to the fault, thereby improving the accuracy of fault detection. It also provides high-quality label data for subsequent machine learning model training and optimization, which helps to improve the performance of the entire fault diagnosis system.

[0153] Furthermore, the intelligent feeder terminal fault online identification method further includes the following steps for determining an abnormal threshold:

[0154] F1, obtains the minimum value among all abnormal label values ​​and defines it as the abnormal threshold.

[0155] Through the above steps, the intelligent feeder terminal fault online identification method can use the minimum abnormal label value of the training data during model training as the threshold for subsequent feeder abnormality judgment, which can be matched with the data during training to ensure the recognition effect and accuracy of feeder abnormality.

[0156] Furthermore, the intelligent feeder terminal fault online identification method further includes the following steps for training a fault classification model:

[0157] G1, obtain corresponding fault classification information from each historical feeder fault information;

[0158] The fault classification information is classification information of the historical faults of the feeder in each historical feeder fault information, and can be determined by the staff during the past feeder fault inspection.

[0159] G2, generates corresponding classification label values ​​based on all fault classification information;

[0160] Category label value The label value corresponding to each fault category information.

[0161] G3, generating a fault classification training set based on the historical feeder terminal feature vector data corresponding to each historical feeder fault information and the corresponding classification label value combination;

[0162] The fault classification training set is a collection of historical feeder terminal feature vector data and corresponding classification label values, which is used to train the fault classification model.

[0163] G4, training a preset second machine learning model according to the fault classification training set to generate a corresponding fault classification model;

[0164] The second machine learning model is a pre-set machine learning model, which is used to generate a corresponding fault classification model based on the fault classification training set. The second machine learning model can be a variety of available machine learning models, such as polynomial logistic regression, neural network, naive Bayes classifier and other models.

[0165] Through the above steps, the intelligent feeder terminal fault online identification method can systematically utilize historical fault data to build an effective fault classification model. By combining historical fault information, feature vector data and classification labels, a high-quality training set is created. The experience of known faults can be fully utilized to train the model to identify various faults and improve the accuracy and efficiency of fault diagnosis.

[0166] Furthermore, the intelligent feeder terminal fault online identification method further includes the following steps for verifying the fault classification information:

[0167] H1, calculate the corresponding mean eigenvector according to the eigenvector data of each historical feeder terminal;

[0168] The mean feature vector is a vector generated by combining the average values ​​of each feature in the historical feeder terminal feature vector data.

[0169] H2, classify each mean feature vector according to the fault classification information to generate multiple classification feature vector data;

[0170] The classification feature vector data is a collection of mean feature vectors corresponding to historical feeder terminal feature vector data having the same fault classification information.

[0171] H3, calculate the corresponding classification center feature vector based on the classification feature vector data;

[0172] The classification center feature vector is the mean vector corresponding to the classification feature vector data, which can be determined by calculating the mean of each feature.

[0173] H4, calculate the corresponding center distance according to the classification center feature vector and each mean feature vector in the classification feature vector data;

[0174] The center distance is the distance from each mean feature vector in the classification feature vector data to the classification center feature vector. The center distance can be Euclidean distance, Manhattan distance, Minkowski distance, etc.

[0175] H5, if the center distance of the mean feature vector is greater than the preset deviation threshold, the fault classification information corresponding to the mean feature vector is defined as abnormal classification information;

[0176] The deviation threshold is a pre-set reference value used to determine whether the center distance of the mean eigenvector is too large, and can be set based on experience or data;

[0177] Abnormal classification information is fault classification information that may contain classification errors;

[0178] Usually, the data corresponding to the same fault classification information have certain similarities and will form a data cluster. If the mean feature vector of a certain fault classification information is far away from the corresponding central feature vector, the fault classification information may be incorrectly classified into the fault classification, and further confirmation of the fault classification is required.

[0179] Through the above steps, the intelligent feeder terminal fault online identification method calculates the mean eigenvector and the classification center eigenvector and compares the distance between them to verify the fault classification information to distinguish abnormal or erroneous classification information. By setting the deviation threshold, an adjustable judgment standard is achieved, which can capture significant classification errors and tolerate reasonable data fluctuations, thereby improving the accuracy and reliability of the fault classification information and further improving the effectiveness of the training model.

[0180] The above are all preferred embodiments of the present application, and are not intended to limit the protection scope of the present application. Any feature disclosed in this specification (including the abstract and drawings), unless otherwise stated, can be replaced by other equivalent or alternative features with similar purposes. That is, unless otherwise stated, each feature is only an example of a series of equivalent or similar features.

Claims

1. An intelligent feeder terminal fault online judgment method, characterized in that: The following steps are involved: A1, respectively obtaining power parameter data, switch status data and environmental data of each preset feeder terminal within a preset monitoring time window; A2, extracting features using a preset feature extraction algorithm based on the power parameter data, switch status data and environmental data to obtain feeder terminal feature vector data; A3, generating corresponding abnormal probability value data through a pre-trained feeder abnormal probability recognition model according to the feeder terminal feature vector data; A4, calculating the corresponding abnormal probability confidence interval according to the abnormal probability value data and the preset confidence level and determining the corresponding lower limit value of the abnormal probability confidence interval; A5, if the lower limit value of the abnormal probability confidence interval of the feeder terminal is greater than the preset abnormal threshold, the corresponding feeder terminal is defined as an abnormal feeder terminal; A6, determining a corresponding adjacent feeder terminal according to the abnormal feeder terminal and preset grid topology information; A7, generating fault classification data according to the feeder terminal feature vector data of the abnormal feeder terminal using a pre-trained fault classification model and defining the data as abnormal terminal fault classification data; A8, generating fault classification data according to the feeder terminal feature vector data of the adjacent feeder terminal corresponding to the abnormal feeder terminal using a pre-trained fault classification model and defining the data as adjacent terminal fault classification data; A9, generating abnormal terminal fault comprehensive classification data based on the abnormal terminal fault classification data and the adjacent terminal fault classification data; The intelligent feeder terminal fault online identification method further includes the following steps for training and generating a feeder abnormality probability recognition model: C1, obtaining corresponding historical feeder fault information from a preset power history database; C2, obtaining the corresponding feeder fault time from each historical feeder fault information; C3, determine the corresponding data backtracking time window according to the feeder fault time and the monitoring time window; C4, obtaining historical power parameter data, historical switch status data and historical environmental data of the corresponding feeder terminal from the power history database according to the data backtracking time window; C5, performing feature extraction using a feature extraction algorithm based on historical power parameter data, historical switch state data, and historical environment data to obtain corresponding historical feeder terminal feature vector data; C6, generates anomaly recognition training data set based on all historical feeder terminal feature vector data and corresponding anomaly label value combinations; C7, training a preset first machine learning model according to the abnormality recognition training data set to generate a feeder abnormality probability recognition model; The intelligent feeder terminal fault online identification method further comprises the following steps for determining the abnormal label value: E1, determining the corresponding sampling time according to each feature vector in the historical feeder terminal feature vector data and defining it as the feature time; E2, determining the characteristic moment distance according to the characteristic moment and the corresponding feeder fault moment; E3, determine the corresponding abnormal label value according to the characteristic moment distance, data backtracking time window and preset label standard value.

2. The intelligent feeder terminal fault online identification method according to claim 1 is characterized in that: The feature extraction algorithm comprises the following steps: B1, generating corresponding power standardization data and environmental standardization data according to the power parameter data and the environmental data using a preset data standardization algorithm; B2, generating feature vector data according to the sampling time combination of each sample in the power standardized data, switch state data and environmental standardized data; B3, generating reduced dimension feature vector data using a preset feature dimension reduction algorithm according to the feature vector data; B4, define the dimension-reduced feature vector data as the feeder terminal feature vector data.

3. The intelligent feeder terminal fault online identification method according to claim 2 is characterized in that: The following steps are further included to determine the data backtracking time window: D1, determine the corresponding half-value of the time window length according to the monitoring time window; D2, determine the corresponding data backtracking start time according to the feeder fault time minus half the time window length; D3, determine the corresponding data backtracking termination time according to the feeder fault time plus half the time window length; D4, determining the corresponding data backtracking time window according to the data backtracking start time and the data backtracking end time.

4. The intelligent feeder terminal fault online determination method according to claim 3 is characterized in that: The method further includes the following steps for determining an abnormal threshold: F1, obtains the minimum value among all abnormal label values ​​and defines it as the abnormal threshold.

5. The intelligent feeder terminal fault online determination method according to claim 4 is characterized in that: The following steps are further included for training the fault classification model: G1, obtain corresponding fault classification information from each historical feeder fault information; G2, generates corresponding classification label values ​​based on all fault classification information; G3, generating a fault classification training set based on the historical feeder terminal feature vector data corresponding to each historical feeder fault information and the corresponding classification label value combination; G4, training a preset second machine learning model according to the fault classification training set to generate a corresponding fault classification model.

6. The intelligent feeder terminal fault online determination method according to claim 5 is characterized in that: The following steps are further included for verifying the fault classification information: H1, calculate the corresponding mean eigenvector according to the eigenvector data of each historical feeder terminal; H2, classify each mean feature vector according to the fault classification information to generate multiple classification feature vector data; H3, calculate the corresponding classification center feature vector based on the classification feature vector data; H4, calculate the corresponding center distance according to the classification center feature vector and each mean feature vector in the classification feature vector data; H5: If the center distance of the mean feature vector is greater than the preset deviation threshold, the fault classification information corresponding to the mean feature vector is defined as abnormal classification information.

Images