Fault recording data fault feature extraction method

Abstract

The invention relates to the technical field of power grid system engineering, and discloses a fault recording data fault feature extraction method, which comprises the following steps: data acquisition and integration: based on a regulation and control cloud platform, carrying out fault feature extraction on fault recording data; fault recording data, geographic information, meteorological information, lightning positioning system data and power grid line fault data over the years are collected, deep fusion is carried out on the multi-source data, and an omnibearing data fusion framework is constructed. According to the fault recording data fault feature extraction method, in order to improve the accuracy of fault feature extraction and provide a more accurate basis for power grid fault diagnosis and processing, the multi-level data fusion analysis architecture is adopted to comprehensively understand the operation state of primary and secondary equipment when a fault occurs, so that deeper and more accurate extraction of fault features is realized, and the fault diagnosis accuracy is improved. Subsequent fault diagnosis and processing can be carried out based on more accurate feature information, and the overall accuracy and effectiveness are improved.

Description

Technical Field

[0001] The invention relates to the technical field of power grid system engineering, and in particular to a method for extracting fault characteristics from fault recording data. Background Art

[0002] In power grid system engineering, for AC transmission line and power grid AC line faults, it is necessary to integrate information from multiple data sources to fully understand the fault situation. For example, it is necessary to integrate multi-source data such as fault recording, geographic information, meteorological information, lightning location system, and power grid line faults over the years, so as to study the characteristic quantities of different types of faults such as wildfires, lightning strikes, icing, bird damage, wind deviation, etc.

[0003] Among them, fault feature extraction is the key technical difficulty, because fault features are the specific representation of faults and are closely related to the causes of faults. Therefore, it is necessary to count and classify fault features based on accumulated experience, and to mine feature information that can represent faults. To this end, it is necessary to integrate multi-source data (such as lightning positioning systems, wildfire warning systems, and geographic information system data) according to the fault characteristics of different typical fault scenarios, study fast and accurate fault identification algorithms, make full use of professional source information of relay protection, and establish a standardized and shareable fault feature database to improve the accuracy and speed of fault analysis calculations, and meet the needs of intelligent diagnosis and analysis sharing of typical faults and complex faults. Study fast and accurate fault identification algorithms, make full use of professional source information of relay protection, and establish a standardized and shareable fault feature database to improve the accuracy and speed of fault analysis calculations, and meet the needs of intelligent diagnosis and analysis sharing of typical faults and complex faults.

[0004] However, although a variety of technologies are currently used to pre-process the fault data of multiple systems in the power grid, the data quality may still be affected by various factors, which cannot improve the accuracy of fault feature extraction, provide a more accurate basis for power grid fault diagnosis and processing, improve the accuracy and efficiency of fault identification, quickly determine the fault type and related processing measures, shorten the fault processing time, and comprehensively consider various factors to achieve all-round and multi-angle judgment of AC line faults and make disaster prevention and mitigation preparations in advance. In view of this, we propose a fault feature extraction method for fault recording data. Summary of the invention

[0005] The purpose of the present invention is to provide a method for extracting fault characteristics from fault recording data to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for extracting fault characteristics from fault recording data comprises the following steps:

[0008] S1. Data collection and integration: Based on the control cloud platform, fault recording data, geographic information, meteorological information, lightning location system data, and power grid line fault data over the years are collected, and the above multi-source data are deeply integrated to build a comprehensive data fusion framework. The fusion process uses an intelligent fusion algorithm, which can automatically learn and identify the correlation and complementarity between different data sources, realize the coordinated use of data in different fields, and mine the potential correlation between different data sources to fully describe the situation when the fault occurs, providing a rich and comprehensive basis for fault feature extraction;

[0009] S2. Data preprocessing: Use time series data processing technology, big data analysis methods and machine learning algorithms to preprocess the collected and integrated multi-system fault data of the power grid, deal with the noise, outliers and inconsistent data formats that may exist in the data, and ensure that the data quality reaches a level that can be used for subsequent accurate fault feature extraction. Different technologies complement each other and play their respective advantages to achieve comprehensive data processing;

[0010] S3. Multi-level data fusion analysis: a multi-level data fusion analysis framework of data model fusion, information fusion and knowledge fusion is used to process the pre-processed data. Specifically, firstly, data model fusion is performed to integrate data from different data sources according to a certain model; then information fusion is performed to further extract effective information from the data; finally, through knowledge fusion, professional knowledge is combined with the fused information to fully understand the operating status of primary and secondary equipment when the fault occurs, so as to achieve a deeper and more accurate extraction of fault characteristics;

[0011] S4. Fault feature extraction and database application. According to the fault features of different typical fault scenarios, a specially studied fast and accurate fault identification algorithm is used to extract fault features from the fused and analyzed data. The fault identification algorithm can identify different types of fault scenarios such as wildfires, lightning strikes, ice cover, bird damage and wind deviation according to their unique fault features. At the same time, make full use of the professional source information of relay protection to establish a standardized and shareable fault feature database, compare and reference the extracted fault features with the standard features in the database, and continuously improve the accuracy and speed of fault analysis calculations to meet the needs of intelligent diagnosis and analysis sharing of typical and complex faults.

[0012] S5. Feature quantity sorting and optimization: A feature quantity sorting algorithm based on the Fisher score method is used to sort the importance of various extracted features and screen out the most representative key features. In addition, a dynamic feature quantity optimization mechanism is designed in combination with the Fisher score sorting. According to the changes in real-time data, the selection and weight of feature quantities are dynamically adjusted to adapt to changes in different fault scenarios and complexity, thereby improving the accuracy and efficiency of fault identification.

[0013] S6. Comprehensive judgment and intelligent diagnosis: Build a multi-dimensional comprehensive judgment model, comprehensively consider various fault characteristics and auxiliary information, such as meteorological conditions, lightning activities, and geographical environment, to conduct all-round and multi-angle judgment on AC line faults; Combine with the dispatching cloud platform to achieve real-time intelligent diagnosis and early warning of AC line faults. When abnormal data is detected, the system can quickly start the diagnosis program, output the fault type and time with the highest correlation degree, and at the same time provide handling principles and suggestions, shorten the fault handling time, and improve the reliability and safety of power grid operation.

[0014] Preferably, in the data acquisition step, the collected geographical information at least includes the longitude and latitude, altitude, and topographic and geomorphic information of the location where the line is located; the collected meteorological information at least includes temperature, humidity, wind speed, wind direction, rainfall, air pressure, and weather phenomenon type information; the data collected by the lightning positioning system at least includes lightning occurrence time, location, and intensity information; the historical line fault data of the power grid at least includes past fault occurrence time, fault type, fault location, and handling measures information.

[0015] Preferably, the intelligent fusion algorithm includes but is not limited to the convolutional neural network (CNN) or recurrent neural network (RNN) in deep learning algorithms, and the multi-layer perceptron (MLP) in neural network algorithms. Through these algorithms, automatic learning and recognition of the correlation and complementarity between different data sources are realized.

[0016] Preferably, in the data preprocessing step: The time series data processing technology is used to process the time series-related characteristics in the data, including but not limited to smoothing processing, differencing processing, and seasonal adjustment processing of time series data to remove seasonal fluctuations and trend changes in the data, and factors affecting data quality; The big data analysis method is used to process massive data, including but not limited to using data mining techniques to conduct clustering analysis, association rule mining, and classification analysis on the data to discover potential patterns and rules in the data and improve data quality; The machine learning algorithm is used to achieve automatic classification and pattern recognition of data, including but not limited to using support vector machine (SVM), decision tree, and naive Bayes algorithm to classify the data to identify outliers and noise factors affecting data quality.

[0017] Preferably, in the multi-level data fusion analysis step: the models adopted in the data model fusion include but are not limited to linear regression models, logistic regression models, and principal component analysis models, making the integrated data more convenient for subsequent analysis and processing; in the information fusion step, the methods for refining effective information include but are not limited to information entropy calculation and mutual information calculation methods to quantify and extract the effective information in the data for further in-depth analysis; in the knowledge fusion step, the professional knowledge combined includes but is not limited to the operation principles of power systems, fault diagnosis principles, and relay protection principles. By combining this professional knowledge with the fused information, the operating states of primary and secondary equipment during a fault can be understood more accurately, and thus the fault characteristics can be extracted more deeply.

[0018] Preferably, the fault recognition algorithms for different types of fault scenarios include: wildfires, lightning strikes, icing, bird damage, and wind deflection.

[0019] Preferably, the fault feature database includes the following parts:

[0020] Standard fault feature table: storing the standard feature information of various typical faults, including fault types, fault occurrence conditions, and fault feature quantities, providing a comparison reference standard for the extracted fault features; historical fault case table: recording the information of past fault cases, including the fault occurrence time, location, type, treatment process, and results, for analyzing the fault development law and summarizing experience and lessons; feature quantity association table: storing the association relationships between different fault types and related feature quantities to quickly search for and analyze the feature quantity information related to specific fault types.

[0021] Preferably, the feature quantity ranking algorithm based on the Fisher score method ranks the importance degree of feature quantities according to the following formula:

[0022] Let the sample set be X = {x1, x2, …, x n}, and the category set be C = {c1, c2, …, c m}. For feature f, its Fisher score calculation formula is:

[0023]

[0024] where, n i is the number of samples in category c i , μ i,f is the mean value of feature in category c i , μ f is the mean value of feature f in the entire sample set, is the variance of feature in category c iThe variance of the feature f. The feature quantities are sorted according to the calculated Fisher score F(f). The higher the Fisher score of a feature quantity, the higher its degree of importance.

[0025] Preferably, the dynamic feature quantity optimization mechanism dynamically adjusts the selection and weight of feature quantities according to the following rules:

[0026] Let the current time be t. At time t, the Fisher scores F t (f) of each feature quantity are calculated based on real-time data, and are compared with the Fisher scores F t-1 (f) of each feature quantity calculated at the previous time t - 1. If the absolute value of the difference between F t (f) and F t-1 (f) is greater than the set threshold ΔF, then the weight w t (f) of the feature quantity f is adjusted according to the following formula:

[0027] w t (f) = w t-1 (f) + α(F t (f) - F t-1 (f))

[0028] where w t-1 (f) is the weight of the feature quantity f at the previous time, α is the adjustment coefficient, and the value range is: 0 < α < 1;

[0029] Meanwhile, according to the change situation of the weights of each feature quantity, the selection of feature quantities is adjusted. When the weight of a certain feature quantity is lower than the set lower limit value w min , this feature quantity is temporarily excluded from the subsequent fault feature extraction and analysis process; when the weight of a certain feature quantity is higher than the set upper limit value w max , this feature quantity is taken as the key attention object, and its participation degree in the subsequent analysis is increased.

[0030] Preferably, in the comprehensive research and judgment and intelligent diagnosis step: after the real-time intelligent diagnosis and warning system outputs the fault type and time with the maximum correlation degree, it provides treatment principles and suggestions according to the following rules: for different fault types, different treatment principle and suggestion templates are set, and the corresponding template is matched according to the current fault type, and specific fault information is filled in the template to generate specific treatment principle and suggestion content.

[0031] Compared with the prior art, the present invention provides a method for extracting fault features from fault recording data, and has the following beneficial effects:

[0032] 1. This method for extracting fault characteristics from fault recording data aims to improve the accuracy of fault characteristic extraction, provide a more accurate basis for power grid fault diagnosis and treatment, comprehensively understand the operating states of primary and secondary equipment during a fault by adopting a multi-level data fusion analysis framework, thereby achieving a more in-depth and accurate extraction of fault characteristics, enabling subsequent fault diagnosis and treatment to be carried out based on more accurate characteristic information, and enhancing the overall accuracy and effectiveness.

[0033] 2. This method for extracting fault characteristics from fault recording data aims to enhance the accuracy and efficiency of fault recognition, quickly determine the fault type and related treatment measures, and shorten the fault handling time. By using a feature quantity ranking algorithm based on the Fisher score method to rank and screen key features according to the importance degree of the extracted features, and combining with a dynamic feature quantity optimization mechanism to dynamically adjust the feature quantity selection and weight according to real-time data changes, and at the same time adopting a fault recognition algorithm specifically studied for different typical fault scenarios for targeted recognition, and using a fault characteristic database containing a standard fault characteristic table, a historical fault case table, and a feature quantity association table for comparison and reference, thereby improving the accuracy and efficiency of fault recognition, being able to quickly output the fault type and time with the highest correlation degree, and generating specific content based on the matching treatment principles and recommended templates, effectively shortening the fault handling time and improving the reliability of power grid operation.

[0034] 3. This method for extracting fault characteristics from fault recording data aims to comprehensively consider various factors, achieve an all-round and multi-angle judgment of AC line faults, make preparations for disaster prevention and mitigation in advance, and enhance the safety of power grid operation. By constructing a multi-dimensional comprehensive judgment model, comprehensively considering factors such as meteorological conditions, lightning activities, geographical environment fault characteristics, and auxiliary information, and combining with the control cloud platform to realize real-time intelligent diagnosis and early warning of AC line faults, thereby providing strong support for disaster prevention and mitigation and ensuring the safety and stability of power grid operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is the overall method flow chart of the present invention;

[0036] Figure 2 It is the flow chart of the data collection and integration steps of the present invention;

[0037] Figure 3 It is the flow chart of the multi-level data fusion analysis steps of the present invention;

[0038] Figure 4 It is the flow chart of the fault characteristic extraction and database application steps of the present invention;

[0039] Figure 5 It is the flow chart of the comprehensive judgment and intelligent diagnosis steps of the present invention;

[0040] Figure 6It is a block diagram of the data collection steps of the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] See also Figure 1 - Figure 6 , the present invention provides a technical solution:

[0043] A method for extracting fault characteristics from fault recording data comprises the following steps:

[0044] S1. Data collection and integration: Based on the control cloud platform, we collect fault recording data, geographic information, meteorological information, lightning location system data, and power grid line fault data over the years, deeply integrate the above multi-source data, and build a comprehensive data fusion framework. The fusion process uses an intelligent fusion algorithm, which can automatically learn and identify the correlation and complementarity between different data sources, realize the coordinated use of data in different fields, and dig out the potential correlation between different data sources to fully describe the situation when the fault occurs, providing a rich and comprehensive basis for fault feature extraction;

[0045] S2. Data preprocessing: Use time series data processing technology, big data analysis methods and machine learning algorithms to preprocess the collected and integrated multi-system fault data of the power grid, deal with the noise, outliers and inconsistent data formats that may exist in the data, and ensure that the data quality reaches a level that can be used for subsequent accurate fault feature extraction. Different technologies complement each other and play their respective advantages to achieve comprehensive data processing;

[0046] S3. Multi-level data fusion analysis: a multi-level data fusion analysis framework of data model fusion, information fusion and knowledge fusion is used to process the pre-processed data. Specifically, firstly, data model fusion is performed to integrate data from different data sources according to a certain model; then information fusion is performed to further extract effective information from the data; finally, through knowledge fusion, professional knowledge is combined with the fused information to fully understand the operating status of primary and secondary equipment when the fault occurs, so as to achieve a deeper and more accurate extraction of fault characteristics;

[0047] S4. Fault Feature Extraction and Database Application: According to the fault features of different typical fault scenarios, a specially developed fast and accurate fault identification algorithm is used to extract fault features from the data after fusion analysis. The fault identification algorithm can target different types of fault scenarios such as wildfires, lightning strikes, icing, bird damage, and wind deflection, and conduct targeted identification based on their respective unique fault features. At the same time, make full use of the source-end information of the relay protection profession to establish a standardized and shareable fault feature database, compare and reference the extracted fault features with the standard features in the database, continuously improve the accuracy and speed of fault analysis and calculation, and meet the intelligent diagnosis, analysis, and sharing requirements of typical faults and complex faults.

[0048] S5. Feature Quantity Ranking and Optimization: Use a feature quantity ranking algorithm based on the Fisher score method to rank the importance of various extracted features, and screen out the most representative key features. And in combination with the Fisher score ranking, design a dynamic feature quantity optimization mechanism to dynamically adjust the selection and weight of feature quantities according to the changes in real-time data, so as to adapt to the changes in different fault scenarios and complexities, and improve the accuracy and efficiency of fault identification.

[0049] S6. Comprehensive Judgment and Intelligent Diagnosis: Build a multi-dimensional comprehensive judgment model, comprehensively consider various fault features and auxiliary information, such as meteorological conditions, lightning activities, and geographical environment, to conduct a comprehensive and multi-angle judgment on AC line faults. Combine with the control cloud platform to realize real-time intelligent diagnosis and early warning of AC line faults. When abnormal data is detected, the system can quickly start the diagnosis program and output the fault type and time with the highest correlation, and at the same time provide treatment principles and suggestions, shorten the fault handling time, improve the reliability and safety of power grid operation. In case of typhoons and abnormal weather, it can also give early risk linkage prompts for related line faults, providing strong support for disaster prevention and mitigation.

[0050] In an embodiment of the present invention, in the data acquisition step, the collected geographical information at least includes the longitude, latitude, altitude, and topographical and geomorphic information of the location where the line is located; the collected meteorological information at least includes temperature, humidity, wind speed, wind direction, rainfall, air pressure, and weather phenomenon type information; the collected lightning location system data at least includes lightning occurrence time, location, and intensity information; the collected historical line fault data of the power grid at least includes previous fault occurrence time, fault type, fault location, and treatment measure information.

[0051] In addition, the intelligent fusion algorithm includes, but is not limited to, the convolutional neural network (CNN) or recurrent neural network (RNN) in deep learning algorithms, and the multi-layer perceptron (MLP) in neural network algorithms, and realizes the automatic learning and recognition of the correlation and complementarity between different data sources through these algorithms.

[0052] In addition, in the data preprocessing step: time series data processing techniques are used to process the time series related characteristics in the data, including but not limited to smoothing, differencing, and seasonal adjustment of time series data to remove seasonal fluctuations and trend changes in the data, which are factors affecting data quality; big data analysis methods are used to process massive data, including but not limited to using data mining techniques for clustering analysis, association rule mining, and classification analysis of the data to discover potential patterns and regularities in the data and improve data quality; machine learning algorithms are used to achieve automatic classification and pattern recognition of data, including but not limited to using support vector machines (SVMs), decision trees, and naive Bayes algorithms to classify the data to identify outliers and noise, which are factors affecting data quality.

[0053] In one embodiment of the present invention, in the multi-level data fusion analysis step:

[0054] The models adopted for data model fusion include but are not limited to linear regression models, logistic regression models, and principal component analysis models, making the integrated data more convenient for subsequent analysis and processing; in the information fusion step, the methods for refining effective information include but are not limited to calculating information entropy and mutual information to quantify and extract the effective information in the data for further in-depth analysis; in the knowledge fusion step, the professional knowledge combined includes but is not limited to the operating principles of power systems, fault diagnosis principles, and relay protection principles. By combining this professional knowledge with the fused information, the operating states of primary and secondary equipment during a fault can be understood more accurately, and thus the fault characteristics can be extracted more deeply, so as to achieve a more in-depth and accurate extraction of fault characteristics, enabling subsequent fault diagnosis and processing to be carried out based on more accurate characteristic information and improving the overall accuracy and effectiveness.

[0055] In addition, the specific identification methods of the fault identification algorithm for different types of fault scenarios are shown in the following table:

[0056]

[0057] In addition, the fault feature database includes the following parts: standard fault feature table: stores the standard feature information of various typical faults, including fault types, fault occurrence conditions, and fault feature quantities, providing a comparison reference standard for the extracted fault features; historical fault case table: records the information of past fault cases, including the time, location, type, processing process, and results of the faults, for analyzing the development laws of faults and summarizing experience and lessons; feature quantity association table: stores the association relationships between different fault types and related feature quantities to facilitate quick search and analysis of feature quantity information related to specific fault types.

[0058] In an embodiment of the present invention, the feature quantity ranking algorithm based on the Fisher score method ranks the importance degree of feature quantities according to the following formula:

[0059] Let the sample set be X = {x1, x2, …, x n}, and the category set be C = {c1, c2, …, c m}. For feature f, its Fisher score calculation formula is:

[0060]

[0061] where n i is the number of samples in category c i , μ i,f is the mean value of the feature in category c i , μ f is the mean value of feature f in the entire sample set, is the variance of feature f in category c i . Rank the feature quantities according to the calculated Fisher score F(f). The higher the Fisher score of a feature quantity, the higher its importance degree.

[0062] In addition, the dynamic feature quantity optimization mechanism dynamically adjusts the selection and weight of feature quantities according to the following rules:

[0063] Let the current time be t. At time t, calculate the Fisher score F t (f) of each feature quantity based on real-time data, and compare it with the Fisher score F t-1 (f) of each feature quantity calculated at the previous time t - 1. If the absolute value of the difference between F t (f) and F t-1 (f) is greater than the set threshold ΔF, then adjust the weight w t (f) of feature f according to the following formula:

[0064] w t (f) = w t-1 (f) + α(F t (f) - F t-1 (f))

[0065] where w t-1 (f) is the weight of feature f at the previous time, and α is the adjustment coefficient, and its value range is: 0 < α < 1;

[0066] Meanwhile, adjust the selection of feature quantities according to the change of the weights of each feature quantity. When the weight of a certain feature quantity is lower than the set lower limit value w minAt that time, temporarily exclude this characteristic quantity from the subsequent fault characteristic extraction and analysis process; when the weight of a certain characteristic quantity is higher than the set upper limit value w max At that time, take this characteristic quantity as the key object of concern and increase its participation in the subsequent analysis.

[0067] In addition, in the comprehensive judgment and intelligent diagnosis step:

[0068] The multi-dimensional comprehensive judgment model calculates the correlation degree R of each fault type according to the following formula j :

[0069] Let F ij be the eigenvalue of the i-th fault characteristic for the j-th fault type, W ij be the weight of the i-th fault characteristic for the j-th fault type, and N be the total number of fault characteristics, then

[0070]

[0071] According to the calculated correlation degree R j Sort each fault type. The fault type with the highest correlation degree is the currently most likely fault type. After the real-time intelligent diagnosis and early warning system outputs the fault type and time with the largest correlation degree, it provides handling principles and suggestions according to the following rules: For different fault types, set different handling principle and suggestion templates, match the corresponding template according to the current fault type, and fill in specific fault information such as the fault location and the fault occurrence time in the template to generate specific handling principle and suggestion content.

[0072] Furthermore, improve the accuracy and efficiency of fault identification, can quickly output the fault type and time with the largest correlation degree, and generate specific content based on the matched handling principle and suggestion template, effectively shortening the fault handling time and improving the reliability of power grid operation.

[0073] The above has generally described the present invention in detail, but based on the present invention, some modifications or improvements can be made, which are obvious to those of ordinary skill in the art. Therefore, modifications or improvements without departing from the spirit of the present invention are within the protection scope of the present invention.

Claims

1. A method for extracting fault features from fault recording data, characterized in that: The following steps are involved: S1. Data collection and integration: Based on the control cloud platform, fault recording data, geographic information, meteorological information, lightning location system data, and power grid line fault data over the years are collected, and the above multi-source data are deeply integrated to build a comprehensive data fusion framework. The fusion process uses an intelligent fusion algorithm, which can automatically learn and identify the correlation and complementarity between different data sources, realize the coordinated use of data in different fields, and mine the potential correlation between different data sources; S2. Data preprocessing: Use time series data processing technology, big data analysis methods and machine learning algorithms to preprocess the collected and integrated multi-system fault data of the power grid, deal with the noise, outliers and inconsistent data formats that may exist in the data, and ensure that the data quality reaches a level that can be used for subsequent accurate fault feature extraction; S3. Multi-level data fusion analysis: a multi-level data fusion analysis framework of data model fusion, information fusion and knowledge fusion is used to process the pre-processed data. Specifically, firstly, data model fusion is performed to integrate data from different data sources according to a certain model; then information fusion is performed to further extract effective information from the data; finally, through knowledge fusion, professional knowledge is combined with the fused information to fully understand the operating status of primary and secondary equipment when the fault occurs, so as to achieve a deeper and more accurate extraction of fault characteristics; S4. Fault feature extraction and database application. According to the fault features of different typical fault scenarios, a specially studied fast and accurate fault identification algorithm is used to extract fault features from the fused and analyzed data. The fault identification algorithm can identify different types of fault scenarios such as wildfires, lightning strikes, ice cover, bird damage and wind deviation according to their unique fault features. At the same time, make full use of the professional source information of relay protection to establish a standardized and shareable fault feature database, compare and reference the extracted fault features with the standard features in the database, and continuously improve the accuracy and speed of fault analysis calculations. S5. Feature quantity sorting and optimization: A feature quantity sorting algorithm based on the Fisher score method is used to sort the importance of various extracted features and select the most representative key features. In addition, a dynamic feature quantity optimization mechanism is designed in combination with the Fisher score sorting to dynamically adjust the selection and weight of feature quantities according to the changes in real-time data. S6. Comprehensive analysis and intelligent diagnosis. Construct a multi-dimensional comprehensive analysis model. Consider a variety of fault characteristics and auxiliary information, such as meteorological conditions, lightning activities and geographical environment, to conduct a comprehensive and multi-angle analysis of AC line faults. Combined with the control cloud platform, it can realize real-time intelligent diagnosis and early warning of AC line faults. When abnormal data is detected, the system can quickly start the diagnostic program and output the most correlated fault type and time. At the same time, it provides processing principles and suggestions to shorten the fault handling time and improve the reliability and safety of power grid operation.

2. The method for extracting fault characteristics from fault recording data according to claim 1, characterized in that: In the data collection step, the collected geographic information includes at least the longitude and latitude, altitude, and topographic information of the geographical location of the line; the collected meteorological information includes at least temperature, humidity, wind speed, wind direction, rainfall, air pressure, and weather phenomenon type information; the collected lightning positioning system data includes at least lightning occurrence time, location, and intensity information; the collected power grid line fault data over the years includes at least previous fault occurrence time, fault type, fault location, and treatment measures information.

3. The method for extracting fault characteristics from fault recording data according to claim 1, characterized in that: The intelligent fusion algorithm includes but is not limited to convolutional neural network (CNN) or recurrent neural network (RNN) in deep learning algorithms, and multi-layer perceptron (MLP) in neural network algorithms. These algorithms are used to realize automatic learning and recognition of the correlation and complementarity between different data sources.

4. The method for extracting fault characteristics from fault recording data according to claim 1, characterized in that: In the data preprocessing step: The time series data processing technology is used to process the time series related characteristics in the data, including but not limited to smoothing, difference processing, and seasonal adjustment processing of the time series data to remove seasonal fluctuations and trend changes in the data, factors that affect data quality; The big data analysis method is used to process massive amounts of data, including but not limited to using data mining technology to perform cluster analysis, association rule mining, and classification analysis on the data to discover potential patterns and rules in the data and improve data quality; The machine learning algorithm is used to realize automatic classification and pattern recognition of data, including but not limited to using support vector machine (SVM), decision tree, and naive Bayes algorithm to classify data to identify outliers and noise factors in the data that affect data quality.

5. The method for extracting fault characteristics from fault recording data according to claim 1, characterized in that: In the multi-level data fusion analysis step: The models used in the data model fusion include but are not limited to linear regression models, logistic regression models, and principal component analysis models, so that the integrated data is more convenient for subsequent analysis and processing; In the information fusion step, the method of extracting effective information includes but is not limited to quantifying and extracting effective information in the data through information entropy calculation and mutual information calculation methods for further in-depth analysis; In the knowledge fusion step, the combined professional knowledge includes but is not limited to the power system operation principle, fault diagnosis principle, and relay protection principle. By combining these professional knowledge with the fused information, it is possible to more accurately understand the operating status of primary and secondary equipment when a fault occurs, thereby extracting fault characteristics more deeply.

6. The method for extracting fault characteristics from fault recording data according to claim 1, characterized in that: The fault identification algorithm targets different types of fault scenarios including: wildfire, lightning strike, icing, bird damage and wind deviation.

7. The method for extracting fault characteristics from fault recording data according to claim 1, characterized in that: The fault feature database includes the following parts: Standard fault feature table: stores standard feature information of various typical faults, including fault type, fault occurrence conditions, and fault feature quantity, providing a comparative reference standard for the extracted fault features; Historical fault case table: records the fault case information that has occurred in the past, including the time, location, type, handling process and results of the fault, which is used to analyze the fault development law and summarize the experience and lessons; Feature quantity association table: stores the association between different fault types and related feature quantities, so as to quickly find and analyze the feature quantity information related to a specific fault type.

8. The method for extracting fault characteristics from fault recording data according to claim 1, characterized in that: The feature quantity ranking algorithm based on the Fisher score method ranks the importance of the feature quantities according to the following formula: Assume that the sample set is X = {x1, x2, ..., x n }, the category set is C = {c1,c2,…,c m }, for feature f, its Fisher score calculation formula is: Among them, n i For category c i The number of samples in , μ i,f For category c i The mean of the features in f is the mean of feature f in the entire sample set, For category c i The variance of feature f in is calculated, and the feature quantities are sorted according to the calculated Fisher score F(f). The feature quantity with a higher Fisher score has a higher degree of importance.

9. The method for extracting fault characteristics from fault recording data according to claim 1, characterized in that: The dynamic feature quantity optimization mechanism dynamically adjusts the selection and weight of feature quantities according to the following rules: Assume that the current time is t. At time t, the Fisher score F of each feature is calculated based on the real-time data. t (f) The Fisher score F of each feature value calculated at the previous time t-1 t-1 (f) Compare, if F t (f) and F t-1 If the absolute value of the difference between (f) and (f) is greater than the set threshold ΔF, the weight w of the feature value f is adjusted according to the following formula: t (f): w t (f)=w t-1 (f)+α(F t (f)-F t-1 (f)) Among them, w t-1 (f) is the weight of the feature f at the previous moment, α is the adjustment coefficient, and its value range is: 0<α<1; At the same time, according to the changes in the weights of each feature, the selection of the feature is adjusted. When the weight of a feature is lower than the set lower limit w min When the weight of a feature value is higher than the set upper limit w, the feature value is temporarily excluded from the subsequent fault feature extraction and analysis process; max When analyzing the feature quantity, the feature quantity is taken as the focus to increase its participation in subsequent analysis.

10. The method for extracting fault characteristics from fault recording data according to claim 1, characterized in that: In the comprehensive analysis and intelligent diagnosis step: after the real-time intelligent diagnosis and early warning system outputs the most correlated fault type and time, it provides processing principles and suggestions according to the following rules: for different fault types, different processing principles and suggestion templates are set, and the corresponding template is matched according to the current fault type, and the specific fault information is filled in the template to generate specific processing principles and suggestion content.

Images