A multi-source information fusion power grid anomaly identification method

Abstract

The application discloses a power grid anomaly identification method based on multi-source information fusion, and relates to the technical field of power grid anomaly identification.The method generates anomaly identification data by intercepting abnormal data segments from corresponding real-time data streams every preset identification period for each data source, respectively performs automatic identification and manual verification on the anomaly identification data, determines the extension degree of abnormal data segments of each data source in each anomaly type and each cause based on the data source used for assisting verification by analyzing the extension degree of abnormal data segments of each data source by verification personnel in several manual verification processes, provides extended data segments based on abnormal data segments in subsequent anomaly identification processes, realizes intelligent extension of abnormal data segments, avoids the problem of unreasonable data retrieval time range caused by personal experience difference of verification personnel, ensures the comprehensiveness of data retrieval, and improves the accuracy of power grid anomaly identification result verification.

Description

Technical Field

[0001] This invention relates to the field of power grid anomaly identification technology, specifically a power grid anomaly identification method based on multi-source information fusion. Background Technology

[0002] With the rapid development of power systems towards intelligence and large scale, the structure of power grids is becoming increasingly complex, and the number of various power equipment is increasing significantly. The multi-source information generated during operation, such as power equipment measurement data, equipment status data, and environmental monitoring data, is growing explosively. As a core infrastructure related to the national economy and people's livelihood, the safe and stable operation of the power grid directly depends on the normal operating conditions of various power equipment. Therefore, timely and accurate identification of abnormal conditions of power equipment in the power grid has become a key link in ensuring the safety of the power grid. Currently, AI models have been widely adopted in the field of power equipment anomaly identification in power grids. Among them, models such as deep learning, Bayesian networks, YOLO, and Faster R-CNN have demonstrated excellent performance in power equipment anomaly identification due to their powerful data processing and feature extraction capabilities. They have significantly improved the automation level and efficiency of power equipment anomaly identification in power grids and provided important technical support for power grid operation and maintenance. However, it cannot be ignored that the power equipment anomaly identification results of AI models are not absolutely reliable. Due to quality issues such as noise interference and data loss in multi-source data of power equipment, as well as objective factors such as the complex and ever-changing power grid operation scenarios and the continuous emergence of new types of power equipment faults, AI models inevitably have the risk of false alarms and are prone to misjudging the normal operation fluctuations of power equipment as anomalies. Because the requirements for the safe operation of the power grid are extremely high, any misjudgment of abnormalities in power equipment may trigger a chain of failures, causing serious consequences such as large-scale power outages and equipment damage, which directly affect the national economy and people's livelihood. Therefore, all power equipment anomaly identification results output by the AI ​​model must be manually reviewed by operation and maintenance personnel with professional knowledge. This is a necessary link to ensure the accuracy of power equipment anomaly identification in the power grid and avoid safety risks. The core purpose of manual review is to verify and identify the abnormal results of power equipment output by the AI ​​model one by one, eliminate false anomalies reported by the model, ensure the accuracy of power equipment anomaly judgment, and fundamentally avoid safety accidents caused by model misjudgment. However, in existing technologies, the manual review process for power equipment anomaly identification results output by AI models requires maintenance personnel to retrieve a large amount of relevant auxiliary data to verify the accuracy of the results. Furthermore, the time frame for data retrieval is entirely determined by the maintenance personnel's personal experience, lacking unified standards and effective support. Specifically, maintenance personnel typically first retrieve data from a short time range before and after the anomaly segment, such as the two minutes before or after the anomaly segment. If the data from this time period is insufficient to accurately determine the authenticity of the anomaly, they then retrieve data from a longer time range, such as the anomaly count... The data segment is either 5 minutes before or 5 minutes after the current segment. In addition, when the data segment of a single parameter is insufficient to support the judgment, the operation and maintenance personnel also need to manually retrieve the data segments of other relevant parameters of the abnormal power equipment to further assist in the verification. This series of data retrieval processes all require the operation and maintenance personnel to actively retrieve data one by one from multiple independent business systems. The whole process is cumbersome and time-consuming. Furthermore, it is easy to affect the efficiency and accuracy of the verification of the abnormal power equipment results output by the AI ​​model due to incomplete or untimely data retrieval, or unreasonable time range and parameter types due to differences in personal experience. This cannot meet the needs of rapid handling of abnormal power equipment in the power grid. To address the above problems, this invention proposes a solution. Summary of the Invention

[0003] The purpose of this invention is to provide a power grid anomaly identification method based on multi-source information fusion, in order to solve the problems mentioned in the background art.

[0004] This invention provides a method for identifying power grid anomalies based on multi-source information fusion, comprising the following steps: S1. Every preset identification period, for each data source, according to its preset interception strategy, abnormal data segments are intercepted from the corresponding real-time data stream and abnormal identification data for the identification period is generated for the target power equipment to identify abnormalities. Here, one data source corresponds to one monitoring parameter of the target power equipment. S2. For each identification cycle of abnormal identification data, the abnormality of the power grid is automatically identified based on the data, and the power grid identification data of the identification cycle is generated based on the automatic identification results. The power grid identification data contains a status signal, and the status signal is selected from the number 0 or 1. S3. For each power grid identification data with a status signal of 1 generated, the power grid identification data is displayed to the verification personnel, who then manually verify it. During the verification process, the verification personnel retrieve several feedback data segments by entering several retrieval commands to verify whether the anomaly type in the power grid identification data is correct. After the verification is completed, manual verification data for the corresponding identification period is generated. Each retrieval command can only retrieve one feedback data segment. S4. When the amount of stored manually verified data reaches a preset fixed amount, analyze all manually verified data and generate verification auxiliary information that adapts to several abnormal combinations. S5. When manually verifying the power grid identification data for each identification period, the corresponding abnormal combination is matched according to the abnormal type and cause in the power grid identification data. Based on the verification auxiliary information of the abnormal combination, the power grid identification data is updated, and data segments from several data sources are automatically extracted as auxiliary retrieval data for the identification period for the verification personnel to refer to, thereby completing the efficient verification and identification of power grid abnormalities.

[0005] Furthermore, in step S1, for any data source, according to the preset interception time of the data source, starting from the end time of the previous identification period, all monitoring values ​​collected within the interception time are obtained and arranged in the order of collection time to form a corresponding abnormal data segment.

[0006] Furthermore, in step S2, if the status signal is selected as number 1, it indicates that the target power equipment has identified an anomaly within the corresponding identification period. At this time, the power grid identification data also includes the anomaly type, the cause of the anomaly, and several abnormal data segments and their corresponding data sources that identify the anomaly type. The abnormal data segment of any data source is a data segment extracted from the data segment of the data source that can represent the anomaly type. If the status signal is selected as number 0, it indicates that the target power equipment has not identified an anomaly within the corresponding identification period.

[0007] Furthermore, in step S4, the analysis content for generating verification auxiliary information that adapts to several abnormal combinations is as follows: S41: Label all stored manually verified data as A1, A2, ..., Aa, where a is a preset fixed amount of manually verified data to be stored; S42: Combine and label the anomaly types and causes contained in the manually checked data A1 to obtain the anomaly combination label D1=(B1,C1) of the manually checked data A1, where B1 refers to the anomaly type in the manually checked data A1 and C1 refers to the cause in the manually checked data A1; similarly, obtain the anomaly combination labels D2=(B2,C2), D3=(B3,C3), ..., Da=(Ba,Ca) of the manually checked data A2, A3, ..., Aa. S43: Classify all the obtained abnormal combination labels, group the manually checked data with the same abnormal type and the same cause into the same category, and record them as the same abnormal combination cluster. Several abnormal combination clusters can be obtained. Mark all the obtained abnormal combination clusters as E1, E2, ..., Ee, 1≤e≤a; The abnormal combination cluster has a corresponding relationship with the abnormal combination obtained by combining the abnormal type and the cause of its occurrence; S44: Obtain all invoked objects contained in the abnormal combination cluster E1, and determine whether each invoked object satisfies the preset first and second marking conditions. Mark all invoked objects that satisfy the preset first marking condition as F1, F2, ..., Ff, f≥1; mark all invoked objects that satisfy the preset second marking condition as M1, M2, ..., Mm, m≥1. S45: Calculate the first and second extension thresholds for the retrieved objects F1, F2, ..., Ff: S46: Calculate the retrieval frequency of the retrieved objects M1, M2, ..., Mm, and select the retrieved objects whose values ​​are greater than or equal to the preset retrieval frequency threshold P1 as the auxiliary retrieval objects of the abnormal combination corresponding to the abnormal combination cluster E1; S47: Based on the first and second extension thresholds of the retrieved objects F1, F2, ..., Ff, the retrieval frequency of the retrieved objects M1, M2, ..., Mm, and all auxiliary retrieved objects of the abnormal combination corresponding to the abnormal combination cluster E1, generate and store the verification auxiliary information of the abnormal combination corresponding to the abnormal combination cluster E1. S48: Generate and store the verification auxiliary information of the abnormal combination corresponding to the abnormal combination clusters E2, ..., Ee, respectively, according to S42 to S47.

[0008] Furthermore, in S44, any retrieved object that satisfies the preset first marking condition also contains the abnormal data segment of the retrieved object in the manual verification data containing the retrieved object; any retrieved object that satisfies the preset second marking condition does not contain the abnormal data segment of the retrieved object in the manual verification data containing the retrieved object.

[0009] Furthermore, in S45, the calculation of the first and second extension thresholds for the retrieved objects F1, F2, ..., Ff is as follows: S451: Using the formula Calculate the average retrieval delay K1 of object F1 within the abnormal combination cluster E1, where G1 is the total number of manually checked data containing object F1 within the abnormal combination cluster E1 that simultaneously contains object F1 and its abnormal data segment. H1-j and H2-j represent the start and end times of the retrieval of object F1 in the retrieval instruction of each manually checked data containing object F1 and its abnormal data segment within the abnormal combination cluster E1; I1-j and I2-j represent the start and end times of the abnormal data segment of object F1 in each manually checked data containing object F1 and its abnormal data segment within the abnormal combination cluster E1. S452: For each call duration interval in the call instruction containing the call object F1 within the abnormal combination cluster E1, determine its corresponding call duration. Then, sort all the determined call durations in ascending order. Among all the sorted call durations, take the call duration located in the middle of the sort as the median call duration H1 of the abnormal combination cluster E1. The call duration of any call duration interval is the call end time of the call duration interval minus its call start time. S453: Based on the median retrieval duration H1, and combined with the average retrieval extension duration K1, calculate the first extension threshold L1 and the second extension threshold L2 for the retrieved object F1 in the abnormal combination cluster E1. The formula for calculating the first extension threshold L1 is L1=(K1+G+H1) / 2×ɑ1, and the formula for calculating the second extension threshold L2 is L2=ɑ2×L1. In the above formulas, ɑ1 is a preset first extension coefficient, ɑ2 is a preset second extension coefficient, ɑ1ϵ(1,1.5], ɑ2ϵ(1.5,3], and ɑ1<ɑ2, K is the average of the average retrieval extension durations of the retrieved objects F1, F2, ..., Ff in the abnormal combination cluster E1. S454: Calculate and obtain the first and second extension thresholds of the objects F2, F3, ..., Ff in the abnormal combination cluster E1 in sequence according to S451 to S453.

[0010] Furthermore, in S46, the calculation of the call frequency of the call objects M1, M2, ..., Mm is as follows: The retrieval frequency R1 of object M1 within the abnormal combination cluster E1 is calculated using the formula R1=N1 / Q1. In the formula, N1 is the total number of retrieval instructions containing object M1 in the manually checked abnormal data segments within the abnormal combination cluster E1 that do not contain the retrieval object M1, and Q1 is the total number of all retrieval instructions in the manually checked abnormal data segments within the abnormal combination cluster E1 that do not contain the retrieval object M1. Similarly, the retrieval frequencies of objects M2, M3, ..., Mm are calculated respectively.

[0011] Furthermore, in step S5, the specific details are as follows: Extract the anomaly type and its cause from the power grid identification data, match the corresponding anomaly combination according to the anomaly type and its cause in the power grid identification data, and obtain the verification auxiliary information of the stored anomaly combination; Based on the verification auxiliary information, the extended data segment and prefetched extended data segment of the data source corresponding to each abnormal data segment in the power grid identification data are determined. Each abnormal data segment in the power grid identification data is replaced with the extended data segment of its corresponding data source to complete the update of the power grid identification data. The prefetched extended data segment of the data source corresponding to each abnormal data segment is prefetched into the cache. When it is detected that the retrieval command entered by the verification personnel during the manual verification process contains the corresponding data source, the prefetched extended data segment of the data source is retrieved from the cache and displayed to the verification personnel for viewing. Based on the verification auxiliary information, data segments from several data sources are automatically extracted as auxiliary data for the identification cycle and simultaneously displayed to the verification personnel for analysis and viewing during manual verification.

[0012] Furthermore, in step S5, the contents of the extended data segment and the prefetched extended data segment corresponding to any abnormal data segment in the power grid identification data are determined as follows: If the verification auxiliary information contains a first extension threshold and a second extension threshold of the data source corresponding to the abnormal data segment, then the first truncation extension interval and the second truncation extension interval are determined sequentially according to the end time of the abnormal data segment in the power grid identification data. The left endpoint of the first truncation extension interval is the end time of the abnormal data segment, and the right endpoint is the end time plus the first extension threshold. The left endpoint of the second truncation extension interval is the end time plus the first extension threshold, and the right endpoint is the end time plus the first extension threshold plus the second extension threshold. According to the first intercepted extended interval, the corresponding data segment is intercepted from the real-time data stream of the data source, and the data segment is concatenated with the abnormal data segment to obtain the extended data segment of the data source; According to the second truncation extension interval, the corresponding data segment is extracted from the real-time data stream of the data source, and the data segment is concatenated with the abnormal data segment to obtain the prefetch extension data segment of the data source; If the verification auxiliary information does not contain the data source corresponding to the abnormal data segment, then the abnormal data segment will not be processed.

[0013] Furthermore, in step S5, the data segments from several data sources are automatically extracted based on the verification auxiliary information as the content of the auxiliary data to be retrieved during the identification period, as follows: Each auxiliary retrieval object in the verification auxiliary information is obtained. Based on the preset fixed retrieval time of the auxiliary retrieval object, starting from the end of the previous identification cycle, all monitoring values ​​collected within the preset fixed retrieval time are obtained and arranged in the order of collection time to form the corresponding auxiliary data segment. The auxiliary retrieval data for the recognition period is generated based on the auxiliary data segment of the auxiliary retrieval object.

[0014] Compared with existing technologies, it has the following advantages: This invention generates abnormal identification data for each data source by extracting abnormal data segments from the corresponding real-time data stream at preset identification cycles. This data is used to identify anomalies in target power equipment during the specified identification cycle. The data is then automatically identified and manually verified. By analyzing the extension of abnormal data segments from each data source during several manual verification processes, and based on the data sources used to assist in the verification, the extension of abnormal data segments for each anomaly type and its cause is determined. In subsequent anomaly identification processes, extended data segments and pre-fetched extended data segments are provided based on the abnormal data segments. The pre-fetched extended data segments are quickly retrieved and displayed from the cache. This method achieves intelligent extension of abnormal data segments and avoids the problem of unreasonable data retrieval time ranges due to differences in the personal experience of verification personnel, ensuring the comprehensiveness of data retrieval and improving the accuracy of power grid anomaly identification result verification. Furthermore, the pre-fetched extended data segments are not directly displayed but are first pre-fetched into the cache and quickly retrieved and displayed when a verification personnel request them. This establishes a fast retrieval channel, allowing verification personnel to quickly obtain data after entering a command. This invention determines the auxiliary data retrieval objects for each anomaly type and its cause based on the frequency of retrieval of feedback data segments from various data sources during manual verification. Simultaneously, auxiliary data segments of each auxiliary object are retrieved during the power grid identification data display process shown to the verification personnel. This method achieves automatic retrieval of auxiliary verification data segments, significantly reducing the manual data retrieval operations of verification personnel, lowering the time consumption of a single verification, meeting the needs of rapid power grid anomaly handling, and avoiding the security risks associated with frequent identity verification while reducing cumbersome operations. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 This application provides a method for identifying power grid anomalies based on multi-source information fusion, comprising the following steps: S1. Every preset identification period, for each data source, abnormal data segments are extracted from the corresponding real-time data stream according to the preset interception strategy of the data source; based on all the extracted abnormal data segments and the data source corresponding to each abnormal data segment, abnormal identification data for the identification period is generated for the abnormal identification of the target power equipment, wherein one data source corresponds to one monitoring parameter of the target power equipment. In step S1, for any data source, according to the preset interception time of the data source, starting from the end time of the previous identification period, all monitoring values ​​collected within the interception time are obtained and arranged in the order of collection time to form the corresponding abnormal data segment. In step S1, the monitoring parameters are selected by the management personnel based on the anomaly identification requirements of the target power equipment; S2. For each identification cycle, anomaly identification data is generated, and power grid anomalies are automatically identified based on the data. Power grid identification data for the identification cycle is generated based on the results of the automatic identification. The power grid identification data includes a status signal. In this application, the status signal is selected from the number 0 or 1. If the status signal is selected as the number 1, it indicates that the target power equipment has identified an anomaly within the corresponding identification period. At this time, the power grid identification data also includes the anomaly type, the cause of the anomaly, and several abnormal data segments and their corresponding data sources that identify the anomaly type. The abnormal data segment of any data source is a data segment extracted from the data segment of the data source that can represent the anomaly type. If the status signal is set to 0, it indicates that no abnormality was detected in the target power equipment within the corresponding identification period; In step S2, the process of automatically identifying anomalies in the target power equipment is as follows: The generated anomaly identification data is input into a pre-trained power grid anomaly identification model. The power grid anomaly identification model performs feature extraction and multi-source information fusion analysis on the anomaly identification data to determine whether there is an anomaly in the target power equipment. If an anomaly is determined to exist, the anomaly type and cause are output, and several anomaly data segments that can characterize the anomaly type and their corresponding data sources are marked. Finally, power grid identification data containing a state signal of 1 is generated. If no anomaly is identified, power grid identification data containing a state signal of 0 is generated directly. S3. For each power grid identification data with a status signal of 1 generated, the power grid identification data is displayed to the verification personnel, who then manually verify it. During the verification process, the verification personnel retrieve several feedback data segments by entering several retrieval commands to verify whether the anomaly type in the power grid identification data is correct. After the verification is completed, manual verification data for the corresponding identification period is generated. Each retrieval command can only retrieve one feedback data segment. In step S3, any retrieval instruction includes a retrieval object and a retrieval duration interval. The retrieval object refers to the monitoring parameter, which is selected by the inspector from all monitoring parameters of the target power equipment based on the anomaly type and cause in the power grid identification data. The left and right endpoints of the retrieval duration interval are the retrieval start time and retrieval end time, respectively. The retrieval start time and retrieval end time are determined by the inspector with reference to the start time and end time of the abnormal data segment in the power grid identification data, satisfying that the retrieval start time ≤ the start time and the retrieval end time ≥ the end time. In step S3, the content of the feedback data segment of any retrieval instruction is as follows: based on the retrieval object, the retrieval start time and the retrieval end time of the retrieval duration interval in the retrieval instruction, all monitoring values ​​of the monitoring parameter within the retrieval duration interval are extracted from the real-time data stream of the corresponding data source and arranged in chronological order to form the feedback data segment. By comparing and analyzing all feedback data segments with abnormal data segments in the power grid identification data, the inspectors verify whether the abnormality type and cause marked in the power grid identification data are accurate, determine whether the abnormality is real or false, and record all retrieval instructions and inspection results during the inspection process, including whether the inspection result is real or false. Based on the power grid identification data and verification results, a manual verification report for the identification period is generated and stored. The manual verification data for the identification period is generated and stored based on the power grid identification data, all retrieval instructions and corresponding feedback data segments; S4. When the amount of stored manually verified data reaches a preset fixed amount, analyze all manually verified data and generate verification auxiliary information that adapts to several abnormal combinations. In step S4, the analysis content for generating verification auxiliary information that adapts to several abnormal combinations is as follows: S41: Label all stored manually verified data as A1, A2, ..., Aa, where a is a preset fixed amount of manually verified data to be stored; S42: Combine and label the anomaly types and causes contained in the manually checked data A1 to obtain the anomaly combination label D1=(B1,C1) of the manually checked data A1, where B1 refers to the anomaly type in the manually checked data A1 and C1 refers to the cause in the checked data A1. Similarly, the abnormal combination labels D2=(B2,C2), D3=(B3,C3), ...,Aa of manually checked data A2, A3, ...,Aa are obtained sequentially, where B2 and Ba are the abnormal types extracted from manually checked data A2 and Aa respectively, and C2 and Ca are the causes extracted from manually checked data A2 and Aa respectively. S43: Classify all the obtained abnormal combination labels, group the manually checked data with the same abnormal type and the same cause into the same category, and record them as the same abnormal combination cluster. Several abnormal combination clusters can be obtained. Mark all the obtained abnormal combination clusters as E1, E2, ..., Ee, 1≤e≤a; Since the anomaly types and causes are the same in all manually checked data within any anomaly cluster, the correspondence is as follows: the anomaly cluster corresponds to the anomaly combination obtained by combining the anomaly types and causes. S44: Obtain all invoked objects contained in the abnormal combination cluster E1, and determine whether each invoked object satisfies the preset first and second marking conditions. Mark all invoked objects that satisfy the preset first marking condition as F1, F2, ..., Ff, f≥1; mark all invoked objects that satisfy the preset second marking condition as M1, M2, ..., Mm, m≥1. Any retrieved object that meets the preset first marking condition also contains the abnormal data segment of the retrieved object in the manually checked data containing the retrieved object. Any retrieved object that meets the preset second marking condition does not contain the abnormal data segment of the retrieved object in the manually checked data containing the retrieved object. That is, the retrieved object was selected by the checker from the data source corresponding to other real-time data streams that did not identify abnormal features. S45: Calculate the first and second extension thresholds for the retrieved objects F1, F2, ..., Ff. The specific calculation details are as follows: S451: Using the formula Calculate the average retrieval delay K1 of object F1 within the abnormal combination cluster E1, where G1 is the total number of manually checked data containing object F1 within the abnormal combination cluster E1 that simultaneously contains object F1 and its abnormal data segment. H1-j and H2-j represent the start and end times of the retrieval in the retrieval instruction containing the retrieval object F1 within each manually checked data segment of the abnormal combination cluster E1, which simultaneously contains the retrieval object F1 and its abnormal data segment. I1-j and I2-j represent the start and end times of the abnormal data segment of object F1 in each manually checked data that simultaneously contains object F1 and its abnormal data segment within the abnormal combination cluster E1. It should be noted that when j=1, the manually checked data included in H1-1 and H2-1 are consistent with the manually checked data included in I1-j and I2-j, and so on for j=2, 3, ..., G1. It should also be noted that all variables in the formula have been dimensionless. S452: For each call duration interval in the call instruction containing the call object F1 within the abnormal combination cluster E1, determine its corresponding call duration. Then, sort all the determined call durations in ascending order. Among all the sorted call durations, take the call duration located in the middle of the sort as the median call duration H1 of the abnormal combination cluster E1. The call duration of any call duration interval is the call end time of the call duration interval minus its call start time. S453: Based on the median retrieval duration H1, calculate the first extension threshold L1 and the second extension threshold L2 for the retrieved object F1 in the abnormal combination cluster E1, combined with the average retrieval extension duration K1. The formula for calculating the first extension threshold L1 is L1=(K1+G+H1) / 2×ɑ1, and the formula for calculating the second extension threshold L2 is L2=ɑ2×L1. In the above formulas, ɑ1 is a preset first extension coefficient, ɑ2 is a preset second extension coefficient, ɑ1ϵ(1,1.5], ɑ2ϵ(1.5,3], and ɑ1<ɑ2. K is the average of the average retrieval extension durations of the retrieved objects F1, F2, ..., Ff in the abnormal combination cluster E1. The average retrieval extension durations of the retrieved objects F2, ..., Ff in the abnormal combination cluster E1 are calculated according to step S451. It should be noted that all variables in the formula have been dedimensionalized. S454: Calculate and obtain the first and second extended thresholds of the objects F2, F3, ..., Ff in the abnormal combination cluster E1 in sequence according to S451 to S453; S46: Calculate the call frequency of the retrieved objects M1, M2, ..., Mm. The calculation is as follows: The retrieval frequency R1 of object M1 within the abnormal combination cluster E1 is calculated using the formula R1=N1 / Q1. In the formula, N1 is the total number of retrieval instructions containing object M1 in the manually checked abnormal data segments within the abnormal combination cluster E1 that do not contain the retrieval object M1, and Q1 is the total number of all retrieval instructions in the manually checked abnormal data segments within the abnormal combination cluster E1 that do not contain the retrieval object M1. Similarly, the retrieval frequencies of objects M2, M3, ..., Mm are calculated respectively. Select the objects whose values ​​among the objects M1, M2, ..., Mm are greater than or equal to the preset retrieval frequency threshold P1 as the auxiliary retrieval objects for the abnormal combination corresponding to the abnormal combination cluster E1. S47: Based on the first and second extension thresholds of the retrieved objects F1, F2, ..., Ff, generate verification auxiliary information for the abnormal combination corresponding to the abnormal combination cluster E1 based on the retrieval frequency of the retrieved objects M1, M2, ..., Mm, and store it. The verification auxiliary information also includes all auxiliary retrieved objects of the abnormal combination corresponding to the abnormal combination cluster E1. S48: Generate and store the verification auxiliary information of the abnormal combination corresponding to the abnormal combination clusters E2, ..., Ee respectively according to S42 to S47; S5. When manually verifying the power grid identification data for each identification cycle, match the corresponding abnormal combination according to the abnormal type and cause in the power grid identification data, update the power grid identification data according to the verification auxiliary information of the abnormal combination, and automatically extract data segments from several data sources as auxiliary retrieval data for the identification cycle for the verification personnel to refer to, so as to complete the efficient verification and identification of power grid abnormalities. Step S5 contains the following details: Extract the anomaly type and its cause from the power grid identification data, match the corresponding anomaly combination according to the anomaly type and its cause in the power grid identification data, and obtain the verification auxiliary information of the stored anomaly combination; Based on the verification auxiliary information, the extended data segment and prefetched extended data segment of the data source corresponding to each abnormal data segment in the power grid identification data are determined. Each abnormal data segment in the power grid identification data is replaced with the extended data segment of its corresponding data source to complete the update of the power grid identification data. The prefetched extended data segment of the data source corresponding to each abnormal data segment is prefetched into the cache. When it is detected that the retrieval command entered by the verification personnel during the manual verification process contains the corresponding data source, the prefetched extended data segment of the data source is retrieved from the cache, extracted, and displayed to the verification personnel for viewing. After the verification personnel finish verifying the power grid identification data, it is deleted from the cache. In step S5, the contents of the extended data segment and the prefetched extended data segment corresponding to any abnormal data segment in the power grid identification data are determined as follows: If the verification auxiliary information contains a first extension threshold and a second extension threshold of the data source corresponding to the abnormal data segment, then the first truncation extension interval and the second truncation extension interval are determined sequentially according to the end time of the abnormal data segment in the power grid identification data. The left endpoint of the first truncation extension interval is the end time of the abnormal data segment, and the right endpoint is the end time plus the first extension threshold. The left endpoint of the second truncation extension interval is the end time plus the first extension threshold, and the right endpoint is the end time plus the first extension threshold plus the second extension threshold. According to the first intercepted extended interval, the corresponding data segment is intercepted from the real-time data stream of the data source, and the data segment is concatenated with the abnormal data segment to obtain the extended data segment of the data source; According to the second truncation extension interval, the corresponding data segment is extracted from the real-time data stream of the data source, and the data segment is concatenated with the abnormal data segment to obtain the prefetch extension data segment of the data source; If the verification auxiliary information does not contain the data source corresponding to the abnormal data segment, then the abnormal data segment will not be processed. Based on the verification auxiliary information, several data segments from various data sources are automatically extracted as auxiliary data retrieval content for the identification period, as follows: Each auxiliary retrieval object in the verification auxiliary information is obtained. Based on the preset fixed retrieval time of the auxiliary retrieval object, starting from the end of the previous identification cycle, all monitoring values ​​collected within the preset fixed retrieval time are obtained and arranged in the order of collection time to form the corresponding auxiliary data segment. The auxiliary retrieval data for the recognition period is generated based on the auxiliary data segment of the auxiliary retrieval object.

[0018] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0019] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for identifying power grid anomalies through multi-source information fusion, characterized in that, Includes the following steps: S1. Every preset identification period, for each data source, according to its preset interception strategy, abnormal data segments are intercepted from the corresponding real-time data stream and abnormal identification data for the identification period is generated for the target power equipment to identify abnormalities. Here, one data source corresponds to one monitoring parameter of the target power equipment. S2. For each identification cycle of abnormal identification data, the abnormality of the power grid is automatically identified based on the data, and the power grid identification data of the identification cycle is generated based on the automatic identification results. The power grid identification data contains a status signal, and the status signal is selected from the number 0 or 1. S3. For each power grid identification data with a status signal of 1 generated, the power grid identification data is displayed to the verification personnel, who then manually verify it. During the verification process, the verification personnel retrieve several feedback data segments by entering several retrieval commands to verify whether the anomaly type in the power grid identification data is correct. After the verification is completed, manual verification data for the corresponding identification period is generated. Each retrieval command can only retrieve one feedback data segment. S4. When the amount of stored manually verified data reaches a preset fixed amount, analyze all manually verified data and generate verification auxiliary information that adapts to several abnormal combinations. S5. When manually verifying the power grid identification data for each identification period, the corresponding abnormal combination is matched according to the abnormal type and cause in the power grid identification data. Based on the verification auxiliary information of the abnormal combination, the power grid identification data is updated, and data segments from several data sources are automatically extracted as auxiliary retrieval data for the identification period for the verification personnel to refer to, thereby completing the efficient verification and identification of power grid abnormalities.

2. The power grid anomaly identification method based on multi-source information fusion according to claim 1, characterized in that, In step S1, for any data source, according to the preset interception time of the data source, starting from the end time of the previous identification period, all monitoring values ​​collected within the interception time are obtained and arranged in the order of collection time to form the corresponding abnormal data segment.

3. The power grid anomaly identification method based on multi-source information fusion according to claim 1, characterized in that, In step S2, if the status signal is selected as number 1, it indicates that the target power equipment has identified an anomaly within the corresponding identification period. At this time, the power grid identification data also includes the anomaly type, the cause of the anomaly, and several abnormal data segments and their corresponding data sources that identify the anomaly type. The abnormal data segment of any data source is a data segment that can represent the anomaly type and is extracted from the data segment of the data source. If the status signal is set to 0, it indicates that no abnormality was detected in the target power equipment within the corresponding identification period.

4. The power grid anomaly identification method based on multi-source information fusion according to claim 1, characterized in that, In step S4, the analysis content for generating verification auxiliary information that adapts to several abnormal combinations is as follows: S41: Label all stored manually verified data as A1, A2, ..., Aa, where a is a preset fixed amount of manually verified data to be stored; S42: Combine and label the anomaly types and causes contained in the manually checked data A1 to obtain the anomaly combination label D1=(B1,C1) of the manually checked data A1, where B1 refers to the anomaly type in the manually checked data A1 and C1 refers to the cause in the manually checked data A1; similarly, obtain the anomaly combination labels D2=(B2,C2), D3=(B3,C3), ..., Da=(Ba,Ca) of the manually checked data A2, A3, ..., Aa. S43: Classify all the obtained abnormal combination labels, group the manually checked data with the same abnormal type and the same cause into the same category, and record them as the same abnormal combination cluster. Several abnormal combination clusters can be obtained. Mark all the obtained abnormal combination clusters as E1, E2, ..., Ee, 1≤e≤a; The abnormal combination cluster has a corresponding relationship with the abnormal combination obtained by combining the abnormal type and the cause of its occurrence; S44: Obtain all invoked objects contained in the abnormal combination cluster E1, and determine whether each invoked object satisfies the preset first and second marking conditions. Mark all invoked objects that satisfy the preset first marking condition as F1, F2, ..., Ff, f≥1; mark all invoked objects that satisfy the preset second marking condition as M1, M2, ..., Mm, m≥1. S45: Calculate the first and second extension thresholds for the retrieved objects F1, F2, ..., Ff: S46: Calculate the retrieval frequency of the retrieved objects M1, M2, ..., Mm, and select the retrieved objects whose values ​​are greater than or equal to the preset retrieval frequency threshold P1 as the auxiliary retrieval objects of the abnormal combination corresponding to the abnormal combination cluster E1; S47: Based on the first and second extension thresholds of the retrieved objects F1, F2, ..., Ff, the retrieval frequency of the retrieved objects M1, M2, ..., Mm, and all auxiliary retrieved objects of the abnormal combination corresponding to the abnormal combination cluster E1, generate and store the verification auxiliary information of the abnormal combination corresponding to the abnormal combination cluster E1. S48: Generate and store the verification auxiliary information of the abnormal combination corresponding to the abnormal combination clusters E2, ..., Ee, respectively, according to S42 to S47.

5. The power grid anomaly identification method based on multi-source information fusion according to claim 4, characterized in that, In S44, any retrieved object that satisfies the preset first marking condition also contains the abnormal data segment of the retrieved object in the manual verification data containing the retrieved object; any retrieved object that satisfies the preset second marking condition does not contain the abnormal data segment of the retrieved object in the manual verification data containing the retrieved object.

6. The power grid anomaly identification method based on multi-source information fusion according to claim 4, characterized in that, S45, the calculation of the first and second extension thresholds for retrieved objects F1, F2, ..., Ff is as follows: S451: Using the formula Calculate the average retrieval delay K1 of object F1 within the abnormal combination cluster E1, where G1 is the total number of manually checked data containing object F1 within the abnormal combination cluster E1 that simultaneously contains object F1 and its abnormal data segment. H1-j and H2-j represent the start and end times of the retrieval of object F1 in the retrieval instruction of each manually checked data containing object F1 and its abnormal data segment within the abnormal combination cluster E1; I1-j and I2-j represent the start and end times of the abnormal data segment of object F1 in each manually checked data containing object F1 and its abnormal data segment within the abnormal combination cluster E1. S452: For each call duration interval in the call instruction containing the call object F1 within the abnormal combination cluster E1, determine its corresponding call duration. Then, sort all the determined call durations in ascending order. Among all the sorted call durations, take the call duration located in the middle of the sort as the median call duration H1 of the abnormal combination cluster E1. The call duration of any call duration interval is the call end time of the call duration interval minus its call start time. S453: Based on the median retrieval duration H1, and combined with the average retrieval extension duration K1, calculate the first extension threshold L1 and the second extension threshold L2 for the retrieved object F1 in the abnormal combination cluster E1. The formula for calculating the first extension threshold L1 is L1=(K1+G+H1) / 2×ɑ1, and the formula for calculating the second extension threshold L2 is L2=ɑ2×L1. In the above formulas, ɑ1 is a preset first extension coefficient, ɑ2 is a preset second extension coefficient, ɑ1ϵ(1,1.5], ɑ2ϵ(1.5,3], and ɑ1<ɑ2, K is the average of the average retrieval extension durations of the retrieved objects F1, F2, ..., Ff in the abnormal combination cluster E1. S454: Calculate and obtain the first and second extension thresholds of the objects F2, F3, ..., Ff in the abnormal combination cluster E1 in sequence according to S451 to S453.

7. The power grid anomaly identification method based on multi-source information fusion according to claim 4, characterized in that, S46, The calculation of the call frequency of the call objects M1, M2, ..., Mm is as follows: The retrieval frequency R1 of object M1 within the abnormal combination cluster E1 is calculated using the formula R1=N1 / Q1. In the formula, N1 is the total number of retrieval instructions containing object M1 in the manually checked abnormal data segments within the abnormal combination cluster E1 that do not contain the retrieval object M1, and Q1 is the total number of all retrieval instructions in the manually checked abnormal data segments within the abnormal combination cluster E1 that do not contain the retrieval object M1. Similarly, the retrieval frequencies of objects M2, M3, ..., Mm are calculated respectively.

8. The power grid anomaly identification method based on multi-source information fusion according to claim 1, characterized in that, Step S5 contains the following details: Extract the anomaly type and its cause from the power grid identification data, match the corresponding anomaly combination according to the anomaly type and its cause in the power grid identification data, and obtain the verification auxiliary information of the stored anomaly combination; Based on the verification auxiliary information, the extended data segment and prefetched extended data segment of the data source corresponding to each abnormal data segment in the power grid identification data are determined. Each abnormal data segment in the power grid identification data is replaced with the extended data segment of its corresponding data source to complete the update of the power grid identification data. The prefetched extended data segment of the data source corresponding to each abnormal data segment is prefetched into the cache. When it is detected that the retrieval command entered by the verification personnel during the manual verification process contains the corresponding data source, the prefetched extended data segment of the data source is retrieved from the cache and displayed to the verification personnel for viewing. Based on the verification auxiliary information, data segments from several data sources are automatically extracted as auxiliary data for the identification cycle and simultaneously displayed to the verification personnel for analysis and viewing during manual verification.

9. The power grid anomaly identification method based on multi-source information fusion according to claim 8, characterized in that, In step S5, the contents of the extended data segment and the prefetched extended data segment corresponding to any abnormal data segment in the power grid identification data are determined as follows: If the verification auxiliary information contains a first extension threshold and a second extension threshold of the data source corresponding to the abnormal data segment, then the first truncation extension interval and the second truncation extension interval are determined sequentially according to the end time of the abnormal data segment in the power grid identification data. The left endpoint of the first truncation extension interval is the end time of the abnormal data segment, and the right endpoint is the end time plus the first extension threshold. The left endpoint of the second truncation extension interval is the end time plus the first extension threshold, and the right endpoint is the end time plus the first extension threshold plus the second extension threshold. According to the first intercepted extended interval, the corresponding data segment is intercepted from the real-time data stream of the data source, and the data segment is concatenated with the abnormal data segment to obtain the extended data segment of the data source; According to the second truncation extension interval, the corresponding data segment is extracted from the real-time data stream of the data source, and the data segment is concatenated with the abnormal data segment to obtain the prefetch extension data segment of the data source; If the verification auxiliary information does not contain the data source corresponding to the abnormal data segment, then the abnormal data segment will not be processed.

10. The power grid anomaly identification method based on multi-source information fusion according to claim 9, characterized in that, In step S5, the data segments from several data sources are automatically extracted based on the verification auxiliary information as the content of the auxiliary retrieval data for the identification period, as follows: Each auxiliary retrieval object in the verification auxiliary information is obtained. Based on the preset fixed retrieval time of the auxiliary retrieval object, starting from the end of the previous identification cycle, all monitoring values ​​collected within the preset fixed retrieval time are obtained and arranged in the order of collection time to form the corresponding auxiliary data segment. The auxiliary retrieval data for the recognition period is generated based on the auxiliary data segment of the auxiliary retrieval object.

Images