Fault recording practical training system and method

Abstract

The invention discloses a fault recording practical training system and method, and relates to the technical field of fault recording practical training, and the system comprises an effective value determination unit, a feature vector analysis unit, a feature dimension setting unit, a feature vector screening unit, and a fault recognition training unit. According to the method, the feature vector analysis unit is used for carrying out refined analysis on the data, the multi-dimensional feature vectors are used for expressing related data in a more detailed manner, a subsequent model can accurately analyze the data conveniently, meanwhile, the feature vector screening unit can retrieve the multi-dimensional feature vectors and delete invalid feature vectors, the time needed by model analysis is shortened, and the efficiency of model analysis is improved. The fault identification training unit provides practical training data for the user and compares a judgment result of the user with a model prediction result, so that on one hand, the actual type of the fault can be determined, and on the other hand, the data analysis capability of the user can be improved; therefore, the fault type can be more accurately judged, and facilities can be conveniently maintained and overhauled.

Description

Technical Field

[0001] The present invention relates to the technical field of fault recording training, and specifically provides a fault recording training system and method. Background Technique

[0002] The fault recording training system is a tool specifically used for power system fault simulation, data recording and analysis, which ensures that the operation and maintenance personnel of the power system can improve their skills and cultivate the ability to handle faults. In the patent with the application number 202411232665.5, it is disclosed that "a fault recording method and a fault recording system, which relate to the technical field of battery management. The method includes: obtaining the single-cell state parameters and the total state parameters of multiple energy storage modules; generating the recording data of the battery management system according to the single-cell state parameters and the total state parameters of the multiple energy storage modules, and storing it in the data storage area of the first memory; storing the read recording data in the first buffer area of the first memory; if the target fault level meets the preset fault trigger condition, storing the recording data in the second buffer area of the first memory; reading the historical recording data of the battery management system within a preset historical time period from the first buffer area as the second fault recording data; storing the first fault recording data, the second fault recording data and the fault trigger moment in the second buffer area in the second memory. This application can realize the fault recording of the battery management system, store the corresponding recording data, and provide a data basis for fault tracing."

[0003] The above-mentioned prior art solves the problems of being unable to directly provide DC fault recording and problem analysis for the energy storage battery management system. However, during the operation of the system, since the voltage ratio, current ratio, reclosing result, and quarter are not jointly analyzed, the result of fault type prediction may have a large error, and the feature vectors are not screened, resulting in a long time and a large storage space required for the model during the analysis process. At the same time, the results provided by the user cannot be compared with the model prediction results, and the user's ability to analyze data cannot be improved. Summary of the Invention

[0004] The purpose of the present invention is to provide a fault recording training system and method to solve the problems raised in the above background technique.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A fault recording training system includes a feature vector screening unit and a fault identification training unit.

[0006] The effective value determination unit obtains the voltage, current waveforms and the fault moment of the fault recorder, determines the transformer ratio, analyzes the actual voltage and current data of the first waveform period before and after the fault according to the voltage and current waveforms, calculates the converted current value and voltage value in combination with the transformer ratio, sets the sampling points, and counts the corresponding instantaneous values, calculates the effective values of the voltage and current before and after the fault according to the instantaneous values, and counts the ratio of the effective values of the voltage and current before and after the fault;

[0007] The eigenvector analysis unit determines the reclosing result identifier and the quarter to which each fault occurrence belongs, constructs multiple fault events by using the voltage ratio, current ratio, reclosing identifier and the quarter to which each fault occurrence belongs at the time of each fault occurrence, and analyzes them by using the wavelet decomposition technology to obtain eigenvectors in multiple dimensions;

[0008] The eigen - dimension setting unit, after setting the upper and lower limits of the eigen - dimension, calculates the similarity of the eigenvectors between the fault events, analyzes the adjacent events corresponding to each fault event by using the similarity, calculates the influence value of each adjacent event through the influence value analysis algorithm, determines the correlation function between the number of eigen - features and the influence value, analyzes the optimal eigen - dimension according to the correlation function, classifies the fault types into bird - flock impact, fire, low - resistance contact, high - resistance contact and lightning strike, and adds the corresponding type label to each fault event.

[0009] Preferably, the effective value determination unit includes a voltage value analysis module, a current value analysis module, an effective value analysis module and a ratio analysis module. The voltage value analysis module, after obtaining the voltage and current waveform data recorded in the fault recorder and the fault occurrence moment, determines the voltage transformer ratio and the current transformer ratio, analyzes the actual voltage data corresponding to the first waveform period before and after the fault according to the voltage waveform data and the fault occurrence moment, and calculates the voltage value after being converted by the transformer in combination with the voltage transformer ratio. The current value analysis module analyzes the actual current data corresponding to the first waveform period before and after the fault according to the current waveform data and the fault occurrence moment, and calculates the current value after being converted by the transformer in combination with the current transformer ratio. The effective value analysis module sets the number of sampling points, arranges the sampling points on different waveform periods according to the set number, counts the instantaneous value corresponding to each sampling point, and calculates the effective values of the voltage and current before and after the fault according to the instantaneous value of the sampling point and the number of sampling points. The ratio analysis module, after determining the ratio between the effective values of the voltage before and after each fault occurrence, counts the ratio between the effective values of the current before and after the fault occurrence.

[0010] Preferably, the feature vector analysis unit includes a reclosing identification module, a quarterly division module, and a fault event construction module. The reclosing identification module identifies the reclosing result corresponding to each fault occurrence time. If the reclosing result is in a normal state, the result is marked as one; if the reclosing result is in an abnormal state, the result is marked as zero. The quarterly division module divides all months into four categories, namely the first quarter, the second quarter, the third quarter, and the fourth quarter, and determines the corresponding quarter according to each fault occurrence time. After the fault event construction module determines the voltage ratio, current ratio, reclosing identification, and the corresponding quarter corresponding to each fault occurrence time recorded in the fault recorder, it constructs the corresponding fault event according to the fault occurrence time, voltage ratio, current ratio, reclosing identification, and the corresponding quarter, and analyzes each fault event using the wavelet decomposition technique to obtain feature vectors in multiple dimensions.

[0011] Preferably, the feature dimension setting unit includes a similarity matrix determination module and a proximity matrix analysis module. The similarity matrix determination module sets the upper limit value and the lower limit value of the feature dimension, counts the respective feature vectors of each fault event, calculates the similarity of the feature vectors between all fault events and other fault events, and constructs a similarity matrix of the fault events using the feature vector similarity. The proximity matrix analysis module analyzes multiple adjacent events corresponding to each fault event according to the similarity of the fault event and other fault events in the similarity matrix, counts the similarity between the fault event and its adjacent events, and constructs a proximity matrix corresponding to each fault event according to the similarity.

[0012] Preferably, the feature dimension setting unit further includes an optimal dimension output module and a fault type determination module. The optimal dimension output module analyzes the proximity matrix of the fault events using the influence value analysis algorithm to obtain the influence value of the adjacent events of each fault event, accumulates the influence values of all fault events to obtain the correlation function between the number of features and the influence value, and calculates the optimal feature dimension according to the correlation function. The fault type determination module divides the fault types into five categories, namely bird flock impact fault, fire fault, low-resistance contact fault, high-resistance contact fault, and lightning strike fault. After determining the fault types corresponding to all fault events, it adds the corresponding label information to each fault event according to the fault type. The influence value analysis algorithm is specifically:

[0013]

[0014] Among them, F(s) represents the correlation function, s represents the feature dimension, m represents the number of fault events, n represents the number of adjacent events corresponding to each fault event, i and j represent parameters, and D ij represents the influence value of the jth adjacent event corresponding to the ith fault event.

[0015] Preferably, the feature vector screening unit includes a feature addition module, a feature elimination module, a candidate set determination module, and a model parameter analysis module. After the feature addition module determines the feature vectors corresponding to all fault events, it sets a candidate set and constructs an SVM model, and uses a feature selection algorithm to sequentially add the feature vectors to the candidate set. The accuracy of the SVM model is evaluated through the feature vectors in the candidate set and the type corresponding to each fault event. The feature elimination module arbitrarily selects a feature vector from the current candidate set, calculates the accuracy of the SVM model after removing the current feature vector, compares the difference in accuracy before and after removal. If the difference is less than the threshold, the current feature vector is removed from the candidate set; otherwise, no operation is performed. The candidate set determination module repeats the operation until each feature vector is added to the candidate set, and the dimension of the feature vectors in the candidate set is the optimal feature dimension. After the model parameter analysis module determines the multi-dimensional feature vectors of all fault events according to the feature vectors included in the candidate set, it transmits the multi-dimensional feature vectors of the fault events to the SVM model for analysis to determine the model parameters.

[0016] Preferably, the fault identification training unit includes a real-time data acquisition module, a type prediction module, and a result comparison module. The real-time data acquisition module acquires the real-time fault data recorded in the fault recorder, determines the fault occurrence time, voltage ratio, current ratio, reclosing flag, and the quarter to which it belongs according to the real-time fault data, and transmits them to the user interface to wait for the user's feedback result. The type prediction module analyzes the feature vectors corresponding to the current fault event based on the fault occurrence time, voltage ratio, current ratio, reclosing flag, and the quarter to which it belongs, and transmits them to the SVM model for analysis to predict the type corresponding to the fault event. The result comparison module compares the user's feedback result with the prediction result. If the two are consistent, it prompts the user that the result is correct; otherwise, it returns the prediction result to the user interface.

[0017] The fault recording training method includes the following steps:

[0018] S1. Analyze the effective values of voltage and current: Obtain the voltage and current waveforms of the fault recorder and the fault occurrence time, determine the transformer ratio, analyze the actual voltage and current data of the first waveform cycle before and after the fault according to the voltage and current waveforms, calculate the converted current value and voltage value in combination with the transformer ratio, set the sampling points, and count the corresponding instantaneous values. Calculate the effective values of voltage and current before and after the fault according to the instantaneous values, and count the ratio of the effective values of voltage and current before and after the fault;

[0019] S2. Determine the eigenvector and dimension: Determine the reclosing result identifier and the corresponding quarter for each fault occurrence time. Use the voltage ratio, current ratio, reclosing identifier, and the corresponding quarter at each fault occurrence time to construct multiple fault events. Analyze them using wavelet decomposition technology to obtain eigenvectors in multiple dimensions. After setting the upper and lower limits of the feature dimension, calculate the similarity of the eigenvectors between fault events. Use the similarity to analyze the adjacent events corresponding to each fault event. Calculate the influence value of each adjacent event through the influence value analysis algorithm, and determine the correlation function between the number of features and the influence value. Analyze the optimal feature dimension according to the correlation function, and at the same time add the corresponding fault type label to each fault event;

[0020] S3. Screen the eigenvectors: After determining the eigenvectors of the fault events, add the eigenvectors to the candidate set one by one through the feature selection algorithm, and evaluate the accuracy of the SVM model after adding. Arbitrarily remove one eigenvector. If the difference in the model accuracy before and after removal is less than the threshold, then remove the current eigenvector. Repeat the operation until the dimension of the candidate set reaches the optimal dimension. Transfer the multi-dimensional eigenvectors in the candidate set to the SVM model for analysis and optimize the model parameters;

[0021] S4. Output the fault training result: Obtain the real-time data of the fault recorder, feedback the real-time data to the user interface, and determine the corresponding eigenvector. Input it into the SVM model, compare the predicted fault type with the user feedback. If they are consistent, prompt correctly; if not, return the prediction result.

[0022] Compared with the prior art, the beneficial effects of the present invention are:

[0023] The present invention sets multiple sampling points through the effective value analysis unit, calculates the current effective value and voltage effective value according to the instantaneous values of the sampling points, which is convenient for subsequent identification of the fault type based on the voltage ratio and current ratio before and after the fault. Since different faults have strong correlations with the voltage ratio, current ratio, reclosing result, and quarter, the eigenvector analysis unit refines and analyzes these data, and uses multi-dimensional eigenvectors to describe the relevant data in more detail, which is convenient for the subsequent model to accurately analyze. At the same time, the eigenvector screening unit will retrieve the multi-dimensional eigenvectors, delete the invalid eigenvectors, reduce the time required for model analysis, and further improve the output speed of the fault type result. The fault identification training unit provides training data for users, and compares the user's judgment results with the model prediction results. On the one hand, it can determine the actual type of the fault, and on the other hand, it can improve the user's data analysis ability, enabling them to more accurately judge the fault type, which is convenient for daily maintenance and repair of various facilities. Description of the Drawings

[0024] Figure 1Schematic diagram of the overall system process provided by the embodiments of the present invention;

[0025] Figure 2 Internal module block diagram of the effective value determination unit provided by the embodiments of the present invention;

[0026] Figure 3 Internal module block diagram of the feature vector analysis unit provided by the embodiments of the present invention;

[0027] Figure 4 Internal module block diagram of the feature dimension setting unit provided by the embodiments of the present invention;

[0028] Figure 5 Internal module block diagram of the feature vector screening unit provided by the embodiments of the present invention.

[0029] In the figure: 1. Effective value determination unit; 101. Voltage value analysis module; 102. Current value analysis module; 103. Effective value analysis module; 104. Ratio analysis module; 2. Feature vector analysis unit; 201. Reclosing identification module; 202. Quarter division module; 203. Fault event construction module; 3. Feature dimension setting unit; 301. Similarity matrix determination module; 302. Proximity matrix analysis module; 303. Optimal dimension output module; 304. Fault type determination module; 4. Feature vector screening unit; 401. Feature addition module; 402. Feature elimination module; 403. Candidate set determination module; 404. Model parameter analysis module; 5. Fault identification training unit; 501. Real-time data acquisition module; 502. Type prediction module; 503. Result comparison module. Specific implementation manners

[0030] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0031] Please refer to Figures 1 - 5 , the present invention provides a technical solution: a fault recording training system, including a feature vector screening unit 4 and a fault identification training unit 5;

[0032] The effective value determination unit 1 obtains the voltage and current waveforms and the fault time of the fault recorder, determines the transformer ratio of the instrument transformer, analyzes the actual voltage and current data of the first waveform cycle before and after the fault based on the voltage and current waveforms, calculates the converted current value and voltage value in combination with the transformer ratio of the instrument transformer, sets the sampling points, and counts the corresponding instantaneous values, calculates the effective values of the voltage and current before and after the fault based on the instantaneous values, and counts the ratio of the effective values of the voltage and current before and after the fault;

[0033] The eigenvector analysis unit 2 determines the reclosing result identifier and the quarter to which each fault occurrence time belongs, constructs multiple fault events by using the voltage ratio, current ratio, reclosing identifier and the quarter to which each fault occurrence time belongs, and analyzes them by using the wavelet decomposition technology to obtain eigenvectors in multiple dimensions;

[0034] The feature dimension setting unit 3 calculates the similarity of the eigenvectors between the fault events after setting the upper and lower limits of the feature dimensions, analyzes the adjacent events corresponding to each fault event by using the similarity, calculates the influence value of each adjacent event through the influence value analysis algorithm, determines the correlation function between the number of features and the influence value, analyzes the optimal feature dimension according to the correlation function, divides the fault types into bird group impact, fire, low-resistance contact, high-resistance contact and lightning strike, and adds the corresponding type labels to each fault event.

[0035] The effective value determination unit 1 includes a voltage value analysis module 101, a current value analysis module 102, an effective value analysis module 103 and a ratio analysis module 104. The voltage value analysis module 101 determines the transformer ratio of the voltage transformer and the current transformer after obtaining the voltage and current waveform data recorded in the fault recorder and the fault occurrence time, analyzes the actual voltage data corresponding to the first waveform cycle before and after the fault based on the voltage waveform data and the fault occurrence time, and calculates the voltage value after conversion by the instrument transformer in combination with the transformer ratio of the voltage transformer. The current value analysis module 102 analyzes the actual current data corresponding to the first waveform cycle before and after the fault based on the current waveform data and the fault occurrence time, and calculates the current value after conversion by the instrument transformer in combination with the transformer ratio of the current transformer. The effective value analysis module 103 sets the number of sampling points, arranges the sampling points on different waveform cycles according to the set number, counts the instantaneous value corresponding to each sampling point, and calculates the effective values of the voltage and current before and after the fault based on the instantaneous value of the sampling point and the number of sampling points. The ratio analysis module 104 determines the ratio between the effective values of the voltage before and after each fault occurrence, and counts the ratio between the effective values of the current before and after the fault occurrence;

[0036] The eigenvector analysis unit 2 includes a reclosing identification module 201, a quarterly division module 202, and a fault event construction module 203. The reclosing identification module 201 identifies the reclosing result corresponding to the occurrence time of each fault. If the reclosing result is in the normal state, the result is marked as one; if the reclosing result is in the abnormal state, the result is marked as zero. The quarterly division module 202 divides all months into four categories, namely the first quarter, the second quarter, the third quarter, and the fourth quarter, and determines the corresponding quarter according to the occurrence time of each fault. After the fault event construction module 203 determines the voltage ratio, current ratio, reclosing identification, and the corresponding quarter corresponding to the occurrence time of each fault recorded in the fault recorder, it constructs the corresponding fault event according to the occurrence time of the fault, voltage ratio, current ratio, reclosing identification, and the corresponding quarter, and analyzes each fault event using the wavelet decomposition technology to obtain eigenvectors in multiple dimensions;

[0037] The feature dimension setting unit 3 includes a similarity matrix determination module 301 and a proximity matrix analysis module 302. The similarity matrix determination module 301 sets the upper limit value and the lower limit value of the feature dimension. After counting the eigenvectors of each fault event, it calculates the similarity of the eigenvectors between all fault events and other fault events, and constructs a similarity matrix of fault events using the eigenvector similarity. The proximity matrix analysis module 302 analyzes multiple adjacent events corresponding to each fault event according to the similarity of the fault event and other fault events in the similarity matrix, counts the similarity between the fault event and its adjacent events, and constructs a proximity matrix corresponding to each fault event according to the similarity;

[0038] The feature dimension setting unit 3 further includes an optimal dimension output module 303 and a fault type determination module 304. The optimal dimension output module 303 analyzes the proximity matrix of fault events using the influence value analysis algorithm to obtain the influence value of adjacent events of each fault event, accumulates the influence values of all fault events to obtain the correlation function between the number of features and the influence value, and calculates the optimal feature dimension according to the correlation function. The fault type determination module 304 divides the fault types into five categories, namely bird flock impact fault, fire fault, low-resistance contact fault, high-resistance contact fault, and lightning strike fault. After determining the fault types corresponding to all fault events, it adds the corresponding label information to each fault event according to the fault type. The influence value analysis algorithm is specifically:

[0039]

[0040] Among them, F(s) represents the correlation function, s represents the feature dimension, m represents the number of fault events, n represents the number of adjacent events corresponding to each fault event, i and j represent parameters, and D ij represents the influence value of the jth adjacent event corresponding to the ith fault event;

[0041] The feature vector screening unit 4 includes a feature addition module 401, a feature elimination module 402, a candidate set determination module 403, and a model parameter analysis module 404. After the feature addition module 401 determines the feature vectors corresponding to all fault events, it sets a candidate set and constructs an SVM model. The feature selection algorithm is used to sequentially add the feature vectors to the candidate set, and the accuracy of the SVM model is evaluated by the feature vectors in the candidate set and the type corresponding to each fault event. The feature elimination module 402 randomly selects a feature vector from the current candidate set, calculates the accuracy of the SVM model after removing the current feature vector, compares the difference in accuracy before and after removal. If the difference is less than the threshold, the current feature vector is removed from the candidate set; otherwise, no operation is performed. The candidate set determination module 403 repeats the operation until each feature vector is added to the candidate set, and the dimension of the feature vectors in the candidate set is the optimal feature dimension. After the model parameter analysis module 404 determines the multi-dimensional feature vectors of all fault events according to the feature vectors included in the candidate set, it transmits the multi-dimensional feature vectors of the fault events to the SVM model for analysis to determine the model parameters. The feature selection algorithm is specifically:

[0042] w′=argMCCR(W m +w)

[0043] where w′ represents the accuracy, MCCR(·) represents the maximum accuracy expression of the set, W m represents the candidate set, and w represents the intermediate variable in the forward process;

[0044] The fault identification training unit 5 includes a real-time data acquisition module 501, a type prediction module 502, and a result comparison module 503. The real-time data acquisition module 501 acquires the real-time fault data recorded in the fault recorder, determines the fault occurrence time, voltage ratio, current ratio, reclosing flag, and the quarter to which it belongs according to the real-time fault data, and transmits it to the user interface to wait for the user to feedback the result. The type prediction module 502 analyzes the feature vector corresponding to the current fault event according to the fault occurrence time, voltage ratio, current ratio, reclosing flag, and the quarter to which it belongs, and transmits it to the SVM model for analysis to predict the type corresponding to the fault event. The result comparison module 503 compares the user feedback result with the prediction result. If the two are the same, it prompts the user that the result is correct; otherwise, it returns the prediction result to the user interface;

[0045] A fault recording training method includes the following steps:

[0046] S1. Analyze the effective values of voltage and current: Obtain the voltage and current waveforms of the fault recorder and the fault occurrence time, determine the transformer ratio, analyze the actual voltage and current data of the first waveform cycle before and after the fault based on the voltage and current waveforms, calculate the converted current value and voltage value in combination with the transformer ratio, set the sampling points, and count the corresponding instantaneous values. Calculate the effective values of voltage and current before and after the fault based on the instantaneous values, and count the ratio of the effective values of voltage and current before and after the fault;

[0047] S2. Determine the feature vectors and dimensions: Determine the reclosing result identifier and the quarter to which each fault occurrence time belongs, construct multiple fault events using the voltage ratio, current ratio, reclosing identifier, and the quarter to which each fault occurrence time belongs, analyze them using wavelet decomposition technology to obtain feature vectors in multiple dimensions. After setting the upper and lower limits of the feature dimensions, calculate the similarity of the feature vectors between fault events, analyze the adjacent events corresponding to each fault event using the similarity, calculate the influence value of each adjacent event through the influence value analysis algorithm, and determine the correlation function between the number of features and the influence value. Analyze the optimal feature dimension according to the correlation function, and add the corresponding fault type label to each fault event;

[0048] S3. Screen the feature vectors: After determining the feature vectors of the fault event, add the feature vectors to the candidate set one by one through the feature selection algorithm, and evaluate the accuracy of the SVM model after adding. Arbitrarily remove a feature vector. If the difference in the model accuracy before and after removal is less than the threshold, then remove the current feature vector. Repeat the operation until the dimension of the candidate set reaches the optimal dimension. Transmit the multi-dimensional feature vectors in the candidate set to the SVM model for analysis and optimize the model parameters;

[0049] S4. Output the fault training result: Obtain the real-time data of the fault recorder, feedback the real-time data to the user interface, and determine the corresponding feature vectors. Input them into the SVM model, compare the predicted fault type with the user feedback. If they are the same, prompt correctly. If they are not the same, return the prediction result.

[0050] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0051] Although embodiments of the present invention have been shown and described, those of ordinary skill in the art will appreciate that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A fault recording training system, comprising a feature vector screening unit (4) and a fault identification training unit (5), characterized in that: An effective value determination unit (1), the effective value determination unit (1) obtains the voltage, current waveforms and fault time of a fault recorder, determines the transformer ratio of the current transformer, analyzes the actual voltage and current data of the first waveform period before and after the fault according to the voltage and current waveforms, calculates the converted current value and voltage value in combination with the transformer ratio, sets sampling points, and counts the corresponding instantaneous values, calculates the effective values of voltage and current before and after the fault according to the instantaneous values, and counts the ratio of the effective values of voltage and current before and after the fault; A feature vector analysis unit (2), the feature vector analysis unit (2) determines the reclosing result identifier and the quarter to which it belongs at the moment of each fault occurrence, constructs a plurality of fault events by using the voltage ratio, current ratio, reclosing identifier and the quarter to which it belongs at the moment of each fault occurrence, and analyzes them by using wavelet decomposition technology to obtain feature vectors in multiple dimensions; A feature dimension setting unit (3), after the feature dimension setting unit (3) sets the upper and lower limits of the feature dimension, calculates the similarity of the feature vectors between the fault events, analyzes the adjacent events corresponding to each fault event by using the similarity, calculates the influence value of each adjacent event through an influence value analysis algorithm, determines the correlation function between the number of features and the influence value, analyzes the optimal feature dimension according to the correlation function, divides the fault types into bird flock impact, fire, low-resistance contact, high-resistance contact and lightning strike, and adds the corresponding type label to each fault event.

2. The fault recording training system according to claim 1, characterized in that: The effective value determination unit (1) includes a voltage value analysis module (101), a current value analysis module (102), an effective value analysis module (103) and a ratio analysis module (104). The voltage value analysis module (101) obtains the voltage, current waveform data recorded in the fault recorder and the fault occurrence time, determines the voltage transformer ratio and the current transformer ratio, analyzes the actual voltage data corresponding to the first waveform period before and after the fault according to the voltage waveform data and the fault occurrence time, and calculates the voltage value converted by the transformer in combination with the voltage transformer ratio. The current value analysis module (102) analyzes the actual current data corresponding to the first waveform period before and after the fault according to the current waveform data and the fault occurrence time, and calculates the current value converted by the transformer in combination with the current transformer ratio. The effective value analysis module (103) sets the number of sampling points, arranges the sampling points on different waveform periods according to the set number, counts the instantaneous value corresponding to each sampling point, and calculates the effective values of voltage and current before and after the fault according to the instantaneous value of the sampling point and the number of sampling points. The ratio analysis module (104) determines the ratio between the effective values of voltage before and after each fault occurrence, and counts the ratio between the effective values of current before and after the fault occurrence.

3. The fault recording training system according to claim 1, characterized in that: The feature vector analysis unit (2) includes a reclosing identification module (201), a quarterly division module (202), and a fault event construction module (203). The reclosing identification module (201) identifies the reclosing result corresponding to each fault occurrence time. If the reclosing result is in the normal state, its result is identified as one; if the reclosing result is in the abnormal state, its result is identified as zero. The quarterly division module (202) divides all months into four categories, namely the first quarter, the second quarter, the third quarter, and the fourth quarter, and determines the corresponding quarter to which each fault occurrence time belongs. After the fault event construction module (203) determines the voltage ratio, current ratio, reclosing identification, and the corresponding quarter corresponding to each fault occurrence moment recorded in the fault recorder, it constructs the corresponding fault event according to the fault occurrence moment, voltage ratio, current ratio, reclosing identification, and the corresponding quarter, and uses the wavelet decomposition technology to analyze each fault event to obtain feature vectors in multiple dimensions.

4. A fault recording training system according to claim 1, characterized in that: The feature dimension setting unit (3) includes a similarity matrix determination module (301) and a proximity matrix analysis module (302). The similarity matrix determination module (301) sets the upper limit value and the lower limit value of the feature dimension. After counting the respective feature vectors of each fault event, it calculates the similarity of the feature vectors between all fault events and other fault events, and constructs a similarity matrix of fault events using the feature vector similarity. The proximity matrix analysis module (302) analyzes multiple adjacent events corresponding to each fault event according to the similarity of the fault event and other fault events in the similarity matrix, counts the similarity between the fault event and its adjacent events, and constructs a proximity matrix corresponding to each fault event according to the similarity.

5. A fault recording training system according to claim 4, characterized in that: The feature dimension setting unit (3) further includes an optimal dimension output module (303) and a fault type determination module (304). The optimal dimension output module (303) analyzes the proximity matrix of fault events using the influence value analysis algorithm to obtain the influence value of the adjacent events of each fault event, accumulates the influence values of all fault events to obtain the correlation function between the number of features and the influence value, and calculates the optimal feature dimension according to the correlation function. The fault type determination module (304) divides the fault types into five types, namely bird flock impact fault, fire fault, low-resistance contact fault, high-resistance contact fault, and lightning strike fault. After determining the fault types corresponding to all fault events, it adds the corresponding label information to each fault event according to the fault type.

6. The fault recording training system according to claim 1, characterized in that: The feature vector screening unit (4) includes a feature addition module (401), a feature elimination module (402), a candidate set determination module (403), and a model parameter analysis module (404). After the feature addition module (401) determines the feature vectors corresponding to all fault events, it sets a candidate set and constructs an SVM model. The feature selection algorithm is used to sequentially add the feature vectors to the candidate set, and the accuracy of the SVM model is evaluated by the feature vectors in the candidate set and the type corresponding to each fault event. The feature elimination module (402) arbitrarily selects a feature vector from the current candidate set, calculates the accuracy of the SVM model after removing the current feature vector, compares the difference in accuracy before and after removal. If the difference is less than the threshold, the current feature vector is removed from the candidate set; otherwise, no operation is performed. The candidate set determination module (403) repeats the operation until each feature vector is added to the candidate set, and the dimension of the feature vectors in the candidate set is the optimal feature dimension. The model parameter analysis module (404) determines the multi-dimensional feature vectors of all fault events according to the feature vectors included in the candidate set, and transmits the multi-dimensional feature vectors of the fault events to the SVM model for analysis to determine the model parameters.

7. A fault recording training system according to claim 1, characterized in that: The fault identification training unit (5) includes a real-time data acquisition module (501), a type prediction module (502), and a result comparison module (503). The real-time data acquisition module (501) acquires the real-time fault data recorded in the fault recorder, determines the fault occurrence time, voltage ratio, current ratio, reclosing flag, and the quarter to which it belongs according to the real-time fault data, and transmits them to the user interface to wait for the user to feedback the result. The type prediction module (502) analyzes the feature vectors corresponding to the current fault event according to the fault occurrence time, voltage ratio, current ratio, reclosing flag, and the quarter to which it belongs, and transmits them to the SVM model for analysis to predict the type corresponding to the fault event. The result comparison module (503) compares the user feedback result with the prediction result. If the two are consistent, it prompts the user that the result is correct; otherwise, it returns the prediction result to the user interface.

8. A fault recording training method, characterized in that The fault recording training method is applicable to a fault recording training system according to any one of claims 1-7, and includes the following steps: S1. Analyze the effective values of voltage and current: Obtain the voltage and current waveforms of the fault recorder and the fault occurrence time, determine the transformer ratio of the transformer, analyze the actual voltage and current data of the first waveform cycle before and after the fault according to the voltage and current waveforms, calculate the converted current value and voltage value in combination with the transformer ratio, set the sampling points, and count the corresponding instantaneous values. Calculate the effective values of voltage and current before and after the fault according to the instantaneous values, and count the ratio of the effective values of voltage and current before and after the fault; S2. Determine the eigenvector and dimension: Determine the reclosing result identifier and the corresponding quarter for each fault occurrence time. Use the voltage ratio, current ratio, reclosing identifier, and the corresponding quarter at each fault occurrence time to construct multiple fault events. Analyze them using wavelet decomposition technology to obtain eigenvectors in multiple dimensions. After setting the upper and lower limits of the feature dimension, calculate the similarity of the eigenvectors between fault events. Use the similarity to analyze the adjacent events corresponding to each fault event. Calculate the influence value of each adjacent event through the influence value analysis algorithm, and determine the correlation function between the number of features and the influence value. Analyze the optimal feature dimension according to the correlation function, and at the same time add the corresponding fault type label to each fault event; S3. Screen the eigenvector: After determining the eigenvector of the fault event, add the eigenvector to the candidate set one by one through the feature selection algorithm, and evaluate the accuracy of the SVM model after adding. Arbitrarily remove one eigenvector. If the difference in the model accuracy before and after removal is less than the threshold, then remove the current eigenvector. Repeat the operation until the dimension of the candidate set reaches the optimal dimension. Transmit the multi-dimensional eigenvector in the candidate set to the SVM model for analysis to optimize the model parameters; S4. Output the fault training result: Obtain the real-time data of the fault recorder, feedback the real-time data to the user interface, and determine the corresponding eigenvector. Input it into the SVM model, compare the predicted fault type with the user feedback. If they are consistent, it indicates correct. If they are inconsistent, return the prediction result.

Images