Power grid information operation and maintenance active early warning method and system based on data analysis

Abstract

This invention discloses a proactive early warning method and system for power grid information operation and maintenance based on data analysis, belonging to the field of power grid operation and maintenance technology. The method includes the following steps: acquiring a set of alarm data output by the power grid operation system; diagnosing and extracting initial root cause features from the alarm data set; comparing power grid business indicator data with preset recovery target thresholds; extracting actual effective operation data from manual intervention records; generating a knowledge conflict record set; updating the recommended priority sequence in the handling strategy library based on the actual effective operation data; and applying the corrected causal weight matrix and the updated handling strategy library for early warning. This invention solves the problems of existing early warning models being unable to adapt to business changes, lacking self-correction mechanisms, and prone to secondary faults due to misjudgment by setting up a handling effect observation window to identify invalid operations and tracing back to effective manual operations.

Description

Technical Field

[0001] This invention relates to the field of power grid operation and maintenance technology, and in particular to a proactive early warning method and system for power grid information operation and maintenance based on data analysis. Background Technology

[0002] With the deepening of the digital transformation of the power grid, the scale of information operation and maintenance of power grid enterprises continues to expand. Monitoring the operational status of various business systems, network equipment, and servers has become a key link in ensuring the safe and stable operation of the power grid. Traditional operation and maintenance models mainly rely on fixed alarm thresholds and manual experience judgment, which are difficult to cope with the increasingly complex business scenarios and the real-time processing needs of massive amounts of data.

[0003] Against this backdrop, proactive early warning technology based on big data analytics has emerged. Existing proactive early warning systems typically collect operational indicators of various hardware devices, application services, and network links in the power grid information system. They establish dynamic baselines using historical data, combine correlation analysis algorithms to detect anomalies and pinpoint root causes in real-time monitoring data, and execute automated handling operations based on a pre-set handling strategy library, thereby achieving early warning and rapid recovery from faults.

[0004] While existing technologies have improved operational efficiency to some extent, they still have significant limitations in practical applications. Power grid information systems typically face frequent business version upgrades, configuration parameter adjustments, and changes in the operating environment. When these changes occur, the system's operational characteristics often change accordingly. Existing early warning models, lacking the ability to perceive and adaptively adjust to these changes, still rely on historical baselines from before the changes for anomaly assessment, easily generating a large number of false positive alarms that do not reflect the normal operating state after the changes. When such false positive alarms occur simultaneously with real network jitter or other faults, the root cause analysis algorithm may incorrectly identify the application itself as the source of the fault due to the dominance of false positive alarms, while ignoring the real network problem. More seriously, if maintenance personnel perform actions such as restarting the application based on incorrect root cause diagnoses, not only will they fail to solve the fundamental problem, but they may also interrupt normal business processes, leading to longer service interruptions when subsequent faults worsen. Therefore, existing root cause analysis models lack the ability to learn and correct from erroneous handling, and the recommendation logic of the handling strategy library is relatively rigid, making it difficult to continuously optimize their diagnostic accuracy during actual operation and maintenance. Summary of the Invention

[0005] This application provides a proactive early warning method and system for power grid information operation and maintenance based on data analysis. It solves the problems in the prior art where the early warning model cannot adapt to business changes and lacks the ability to learn and correct from error handling. It realizes root cause diagnosis and strategy optimization based on reverse tracing of actual handling effects, continuously improves the accuracy of early warning and avoids business interruption caused by error handling.

[0006] This application provides a data analysis-based proactive early warning method for power grid information operation and maintenance, including: acquiring a set of alarm data output by the power grid operation system, diagnosing and extracting initial root cause features from the alarm data set, and matching and executing initial handling operations in the handling strategy library based on the initial root cause features; After the initial handling operation is executed, a pre-set observation window is opened, and power grid business indicator data are collected within the observation window and stored in a temporary storage area. At the end of the observation window, the power grid business indicator data is compared with the preset recovery target threshold. If the power grid business indicator data does not reach the preset recovery target threshold, the initial handling operation is determined to be invalid and the reverse tracing logic is triggered. In response to the reverse tracing logic, read the manual intervention records in the unified operation and maintenance platform, and extract the actual effective operation data that restored the power grid business to normal from the manual intervention records; The initial root cause characteristics, initial treatment operations, and actual effective operation data are associated and packaged to generate a knowledge conflict record set. The causal association weight values ​​in the causal weight matrix are dynamically adjusted based on the knowledge conflict record set, and the recommended priority sequence in the disposal strategy library is updated based on actual effective operational data. Obtain real-time alarm data for the next time period, and apply the revised causal weight matrix and the updated handling strategy library to issue early warnings.

[0007] Furthermore, the steps of acquiring the alarm data set output by the power grid operation system and diagnosing and extracting initial root cause features from the alarm data set include: Extract alarm feature values ​​from each dimension of the alarm dataset and construct an alarm feature vector; Read the causal weight matrix stored in the database. The row dimension of the causal weight matrix represents the alarm feature category, the column dimension represents the root cause node category, and the matrix elements represent the causal association strength between the corresponding alarm feature category and the root cause node category. The alarm feature vector is input into the feature mapping logic, and the dot product of the alarm feature vector and the causal weight matrix is ​​calculated through the feature mapping logic to obtain the initial root cause distribution vector. Extract the root cause node category corresponding to the element with the largest value in the initial root cause distribution vector, and mark the root cause node category as the initial root cause feature; In the step of constructing alarm feature vectors, a time sliding window is used to extract the original alarm quantity within a continuous timestamp according to a preset time step, and the extracted original alarm quantity is normalized and mapped to eliminate the difference in dimensions between different alarm sources and generate a standardized input data structure.

[0008] Furthermore, the steps to obtain the initial root cause distribution vector by calculating the dot product of the alarm feature vector and the causal weight matrix through feature mapping logic include: The initial root cause distribution vector is obtained by calculating the dot product between the alarm feature vector and the causal weight matrix through feature mapping logic. The calculation formula is as follows: ; In the formula, Represents the initial root cause distribution vector. This represents the alarm feature vector. Represents the causal weight matrix. This represents the preset confidence adjustment coefficient for the current state, with a value greater than zero and less than one. This represents the historical root cause distribution vector of the previous time period. This represents the preset historical state forgetting coefficient, with a value range greater than zero and less than one. The feature mapping logic retains the accumulated state information in the previous operation of the system by introducing the historical root cause distribution vector and the historical state forgetting coefficient, and integrates them to generate a low-level diagnostic data structure with time-series smoothing properties.

[0009] Furthermore, if the power grid business indicator data does not reach the preset recovery target threshold, the steps to determine that the initial handling operation is invalid and trigger the reverse tracing logic include: Separate the multi-dimensional indicators in the temporary storage area to obtain transaction success rate data sequences and response time data sequences; Obtain the target baseline values ​​for transaction success rate and response time calculated within the historical normal operation cycle as preset recovery target thresholds; Calculate the first absolute difference between the mean of the trade success rate data series and the target baseline value for trade success rate; Calculate the second absolute difference between the mean of the response time data series and the target baseline value of the response time; Determine whether the first absolute difference is greater than the first permissible error limit, and determine whether the second absolute difference is greater than the second permissible error limit; When the first absolute difference is greater than the first allowable error limit, or the second absolute difference is greater than the second allowable error limit, a target deviation indicator is generated; The target deviation indicator is used as the input control signal to send a reverse tracing start command to the power grid operation and maintenance control service system, thereby activating the reverse tracing logic.

[0010] Furthermore, the steps for extracting actual, effective operational data from manual intervention records to restore normal power grid operations include: Send a data retrieval request with the fault occurrence timestamp to the log interface of the unified operation and maintenance platform; Receive the work order processing text stream returned by the log interface after the timestamp of the fault occurrence; Perform word segmentation and slicing on the work order processing text stream to obtain a sequence of word groups containing multiple sets of operational terms; By using a pre-established standard dictionary of operation and maintenance procedures, entity recognition and matching are performed on the word sequence, descriptive words that are detached from the semantics of operation are stripped away, and standardized operation instructions that represent intervention actions are extracted. Extract the system status return code after the standardized operation command is executed. If the system status return code indicates that the service has returned to normal, then confirm the corresponding standardized operation command and output it as the actual valid operation data, thereby capturing the underlying intervention control command that indicates the service recovery.

[0011] Furthermore, the steps for generating a knowledge conflict record set include: Establish a standardized data structure containing four data field areas; The acquired alarm data set is written into the environmental feature field area of ​​the data structure as the baseline data for the scenario context in which the conflict occurred; The initial root cause characteristics and initial treatment operations are written into the negative sample field area of ​​the data structure and marked as failure decision combination data. Write the actual effective operation data into the positive sample field area of ​​the data structure and mark it as the correction target data; Calculate the time difference between the execution time of the initial handling operation and the execution time of the actual effective operation data, and write the time difference into the recovery delay field area of ​​the data structure; Encapsulate the data in the environmental feature field area, negative sample field area, positive sample field area, and recovery delay field area, and output a knowledge conflict record set.

[0012] Furthermore, the steps for dynamically adjusting the causal association weight values ​​in the causal weight matrix based on the knowledge conflict record set include: Analyze the knowledge conflict record set to read the specific alarm feature categories that caused the misdiagnosis and the categories of the root cause nodes of the misjudged errors; Based on specific alarm feature categories and error root cause node categories, locate the corresponding error causal association weight values ​​in the causal weight matrix; Apply a preset penalty attenuation factor to the weight value of the erroneous causal association to reduce the directional strength of a specific alarm feature category to the erroneous root cause node category; Read the actual valid operation data in the knowledge conflict record set, and deduce the true root cause node category in reverse according to the preset mapping rules; Locate the correct causal association weight values ​​between specific alarm feature categories and the true root cause node categories in the causal weight matrix; Apply a preset reward enhancement factor to the correct causal association weight values ​​to complete the numerical update and iteration of the model matrix.

[0013] This application provides a data analysis-based proactive early warning system for power grid information operation and maintenance, used to implement a data analysis-based proactive early warning method for power grid information operation and maintenance, including: Feature extraction module, window preset module, data comparison module, data recovery module, data packaging module, weight correction module, matrix application module; The feature extraction module is used to acquire the alarm data set output by the power grid operation system, diagnose and extract initial root cause features from the alarm data set, and match and execute initial handling operations in the handling strategy library based on the initial root cause features. The window preset module is used to open an observation window of preset duration after the initial processing operation is executed, and collect power grid business indicator data in the observation window and store it in a temporary storage area. The data comparison module is used to compare the power grid business indicator data with the preset recovery target threshold at the end of the observation window. If the power grid business indicator data does not reach the preset recovery target threshold, the initial handling operation is determined to be invalid and the reverse tracking logic is triggered. The data recovery module is used to respond to the reverse tracing logic, read the manual intervention records in the unified operation and maintenance platform, and extract the actual effective operation data that enables the power grid business to return to normal from the manual intervention records; The data packaging module is used to associate and package the initial root cause features, initial treatment operations, and actual effective operation data to generate a knowledge conflict record set. The weight correction module is used to dynamically correct the causal association weight values ​​in the causal weight matrix according to the knowledge conflict record set, and update the recommended priority sequence in the disposal strategy library according to the actual effective operation data. The matrix application module is used to obtain real-time alarm data for the next time period, and apply the corrected causal weight matrix and the updated handling strategy library to issue warnings.

[0014] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: By constructing a closed-loop process that includes feedback on handling effects and records of knowledge conflicts, the proactive early warning method possesses the ability to continuously learn from historical misjudgments. An observation window is set up after the initial handling operation is executed. By comparing the actual recovery of business indicators with preset targets, invalid handling operations are accurately identified, triggering reverse tracing logic. This reverse tracing logic extracts the actual effective operations that ultimately led to business recovery by reading manual intervention records from the unified operations and maintenance platform, solving the problem of the inability to self-correct after automated handling failures.

[0015] Furthermore, by associating and packaging initial root cause features, invalid initial treatment operations, and actual effective operation data to generate a knowledge conflict record set, this record set is used to dynamically correct the causal weight matrix used in the root cause analysis process and synchronously update the recommended priority sequence in the treatment strategy library. This solves the problems of root cause analysis model rigidity and treatment strategy inflexibility, enabling the diagnostic logic to adaptively adjust according to the actual treatment results.

[0016] Furthermore, when new real-time alarm data is received in subsequent time periods, the corrected causal weight matrix and the updated handling strategy library are directly applied for early warning analysis. This allows the entire early warning method to continuously accumulate experience, gradually improve the diagnostic accuracy in complex and ever-changing operating scenarios, reduce the probability of misjudgment caused by model mismatch, and thus provide more reliable decision-making basis for operation and maintenance personnel. Attached Figure Description

[0017] Figure 1 A flowchart of a proactive early warning method for power grid information operation and maintenance based on data analysis provided in this application embodiment; Figure 2 A schematic diagram of the structure of the proactive early warning system for power grid information operation and maintenance based on data analysis provided in the embodiments of this application. Detailed Implementation

[0018] This application provides a proactive early warning method and system for power grid information operation and maintenance based on data analysis. It solves the problems in the prior art where the early warning model cannot adapt to business changes and lacks the ability to learn and correct from error handling. By setting an observation window to identify invalid handling and reverse tracking effective manual operations to dynamically correct the causal weight matrix and handling strategy library, it realizes continuous learning and optimization based on actual handling feedback, effectively improves the accuracy of root cause diagnosis and avoids business interruption caused by error handling.

[0019] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0020] like Figure 1The diagram shown is a flowchart of a proactive early warning method for power grid information operation and maintenance based on data analysis provided in this application embodiment. This method is applied to a proactive early warning system for power grid information operation and maintenance based on data analysis. The method includes the following steps: obtaining a set of alarm data output by the power grid operation system; diagnosing and extracting initial root cause features from the alarm data set based on a causal weight matrix; and matching and executing initial handling operations in the handling strategy library according to the initial root cause features. After the initial handling operation is executed, a pre-set observation window is opened, and power grid business indicator data are collected within the observation window and stored in a temporary storage area. At the end of the observation window, the power grid business indicator data is compared with the preset recovery target threshold. If the power grid business indicator data does not reach the preset recovery target threshold, the initial handling operation is determined to be invalid and the reverse tracing logic is triggered. In response to the reverse tracing logic, read the manual intervention records in the unified operation and maintenance platform, and extract the actual effective operation data that restored the power grid business to normal from the manual intervention records; The initial root cause characteristics, initial treatment operations, and actual effective operation data are associated and packaged to generate a knowledge conflict record set. The causal association weight values ​​in the causal weight matrix are dynamically adjusted based on the knowledge conflict record set, and the recommended priority sequence in the disposal strategy library is updated based on actual effective operational data. Obtain real-time alarm data for the next time period, and apply the revised causal weight matrix and the updated handling strategy library to issue early warnings.

[0021] Furthermore, the steps of acquiring the alarm data set output by the power grid operation system and diagnosing and extracting initial root cause features from the alarm data set include: Extract alarm feature values ​​from each dimension of the alarm dataset and construct an alarm feature vector; Read the causal weight matrix stored in the database. The row dimension of the causal weight matrix represents the alarm feature category, the column dimension represents the root cause node category, and the matrix elements represent the causal association strength between the corresponding alarm feature category and the root cause node category. The alarm feature vector is input into the feature mapping logic, and the dot product of the alarm feature vector and the causal weight matrix is ​​calculated through the feature mapping logic to obtain the initial root cause distribution vector. Extract the root cause node category corresponding to the element with the largest value in the initial root cause distribution vector, and mark the root cause node category as the initial root cause feature; In the step of constructing alarm feature vectors, a time sliding window is used to extract the original alarm quantity within a continuous timestamp according to a preset time step, and the extracted original alarm quantity is normalized and mapped to eliminate the difference in dimensions between different alarm sources and generate a standardized input data structure.

[0022] For the process of diagnosing and extracting initial root cause features from the alarm data set based on the causal weight matrix, the alarm feature values ​​of each dimension in the alarm data set are first extracted to construct an alarm feature vector.

[0023] In practice, a time sliding window is used to extract the raw alarm quantity within a continuous timestamp according to a preset time step (e.g., sixty seconds), and the extracted raw alarm quantity is normalized and mapped to eliminate the difference in dimensions between different alarm sources and generate a standardized input data structure.

[0024] Subsequently, the causal weight matrix stored in the database is read. The row dimension of the causal weight matrix represents the alarm feature category, the column dimension represents the root cause node category, and the matrix elements represent the causal association strength between the corresponding alarm feature category and the root cause node category.

[0025] The alarm feature vector is input into the feature mapping logic, which calculates the dot product between the alarm feature vector and the causal weight matrix to obtain the initial root cause distribution vector. The initial root cause distribution vector is then traversed, and the root cause node category corresponding to the element with the largest value is extracted. This root cause node category is then marked as the initial root cause feature. This processing method directly utilizes the underlying algebraic mapping relationship, reducing computational overhead.

[0026] Furthermore, the steps to obtain the initial root cause distribution vector by calculating the dot product of the alarm feature vector and the causal weight matrix through feature mapping logic include: The initial root cause distribution vector is obtained by calculating the dot product between the alarm feature vector and the causal weight matrix through feature mapping logic. The calculation formula is as follows: ; In the formula, Represents the initial root cause distribution vector. This represents the alarm feature vector. Represents the causal weight matrix. This represents the preset confidence adjustment coefficient for the current state, with a value greater than zero and less than one. This represents the historical root cause distribution vector of the previous time period. This represents the preset historical state forgetting coefficient, with a value range greater than zero and less than one. The feature mapping logic retains the accumulated state information in the previous operation of the system by introducing the historical root cause distribution vector and the historical state forgetting coefficient, and integrates them to generate a low-level diagnostic data structure with time-series smoothing properties.

[0027] To avoid misjudgments caused by transient alarm storms, the initial root cause distribution vector is obtained by calculating the dot product of the alarm feature vector and the causal weight matrix through feature mapping logic, and the calculation formula is executed at the underlying level: ; This formula employs the principles of a mature time-series smoothing algorithm. During the calculation... This represents the alarm feature vector. This represents the causal weight matrix. Performing a matrix dot product operation between the two yields a direct mapping result for the current period. To smooth out sudden fluctuations, a... This represents the preset confidence adjustment coefficient for the current state, and This represents the preset historical state forgetting coefficient, combined with... This represents the historical root cause distribution vector of the previous time period.

[0028] By adjusting the values ​​of the current state confidence adjustment coefficient and the historical state forgetting coefficient (set to 0.7 and 0.3 respectively), the feature mapping logic retains the accumulated state information from previous processes by introducing the historical root cause distribution vector and the historical state forgetting coefficient, and integrates them to generate a low-level diagnostic data structure with temporal smoothing properties, ultimately outputting... This represents the initial root cause distribution vector.

[0029] Furthermore, if the power grid business indicator data does not reach the preset recovery target threshold, the steps to determine that the initial handling operation is invalid and trigger the reverse tracing logic include: Separate the multi-dimensional indicators in the temporary storage area to obtain transaction success rate data sequences and response time data sequences; Obtain the target baseline values ​​for transaction success rate and response time calculated within the historical normal operation cycle as preset recovery target thresholds; Calculate the first absolute difference between the mean of the trade success rate data series and the target baseline value for trade success rate; Calculate the second absolute difference between the mean of the response time data series and the target baseline value of the response time; Determine whether the first absolute difference is greater than the first permissible error limit, and determine whether the second absolute difference is greater than the second permissible error limit; When the first absolute difference is greater than the first allowable error limit, or the second absolute difference is greater than the second allowable error limit, a target deviation indicator is generated; The target deviation indicator is used as the input control signal to send a reverse tracing start command to the power grid operation and maintenance control service system, thereby activating the reverse tracing logic.

[0030] The process involves comparing power grid business indicator data with a preset recovery target threshold. If the power grid business indicator data does not reach the preset recovery target threshold, the initial handling operation is deemed invalid, and reverse tracing logic is triggered.

[0031] The multi-dimensional indicators in the temporary storage area are separated to obtain the transaction success rate data sequence and the response time data sequence. The target baseline values ​​for transaction success rate and response time calculated within the historical normal operation period are retrieved as preset recovery target thresholds.

[0032] The system calculates the first absolute difference between the mean of the transaction success rate data series and the target baseline value for the transaction success rate, and simultaneously calculates the second absolute difference between the mean of the response time data series and the target baseline value for the response time. It then determines whether the first absolute difference exceeds a first allowable error limit, and whether the second absolute difference exceeds a second allowable error limit. If either the first absolute difference exceeds the first allowable error limit, or the second absolute difference exceeds the second allowable error limit, the underlying logic generates a target deviation flag. This target deviation flag is then used as an input control signal to send a reverse tracing start command to the power grid operation and maintenance control service, activating the reverse tracing logic.

[0033] Furthermore, the steps for extracting actual, effective operational data from manual intervention records to restore normal power grid operations include: Send a data retrieval request with the fault occurrence timestamp to the log interface of the unified operation and maintenance platform; Receive the work order processing text stream returned by the log interface after the timestamp of the fault occurrence; Perform word segmentation and slicing on the work order processing text stream to obtain a sequence of word groups containing multiple sets of operational terms; By using a pre-established standard dictionary of operation and maintenance procedures, entity recognition and matching are performed on the word sequence, descriptive words that are detached from the semantics of operation are stripped away, and standardized operation instructions that represent intervention actions are extracted. Extract the system status return code after the standardized operation command is executed. If the system status return code indicates that the service has returned to normal, then confirm the corresponding standardized operation command and output it as the actual valid operation data, thereby capturing the underlying intervention control command that indicates the service recovery.

[0034] By automatically reading the manual intervention records in the unified operation and maintenance platform, the actual effective operation data that enabled the power grid business to return to normal is extracted from the manual intervention records.

[0035] During execution, the program sends a data retrieval request with a fault occurrence timestamp to the unified operations and maintenance platform's log interface, and receives the work order processing text stream returned by the log interface after the fault occurrence timestamp. It then calls the word segmentation component to perform word segmentation and slicing operations on the work order processing text stream, obtaining a sequence of word groups containing multiple sets of operational terms.

[0036] To filter out useless information, a pre-established standard dictionary of operation and maintenance procedures is used to perform entity recognition and matching on phrase sequences, stripping away descriptive words that are detached from the semantics of the operation, and extracting standardized operation instructions representing intervention actions. Furthermore, the status return code after the standardized operation instructions are executed is extracted. If the status return code indicates that the service has returned to normal (status code set to 200), the corresponding standardized operation instruction is confirmed and output as actual valid operation data. This conversion process can capture the underlying intervention control instructions representing service recovery and generate structured data.

[0037] Furthermore, entity recognition and matching of phrase sequences are performed using a pre-established standard dictionary of operation and maintenance procedures, specifically employing cosine similarity calculation, with the following formula: ; In the formula, Indicates the first word in the sequence of phrases The first word group vector and the standard dictionary of operation and maintenance. Cosine similarity between standard term vectors This represents the total number of dimensions of the vector. This indicates the current dimension index number being calculated. Indicates the first The word vector at the th word position in the th word position... The numerical value of dimension, Indicates the first The standard term vector in the th ... The numerical value of the dimension; When the calculated cosine similarity is greater than the preset similarity threshold, the match is considered successful and the corresponding standard term vector is converted into standardized operation instructions. The structured matching result is output by quantifying the closeness of the text semantics by calculating the spatial angle.

[0038] Furthermore, the steps for generating a knowledge conflict record set include: Establish a standardized data structure containing four data field areas; The acquired alarm data set is written into the environmental feature field area of ​​the data structure as the baseline data for the scenario context in which the conflict occurred; The initial root cause characteristics and initial treatment operations are written into the negative sample field area of ​​the data structure and marked as failure decision combination data. Write the actual effective operation data into the positive sample field area of ​​the data structure and mark it as the correction target data; Calculate the time difference between the execution time of the initial handling operation and the execution time of the actual effective operation data, and write the time difference into the recovery delay field area of ​​the data structure; Encapsulate the data in the environmental feature field area, negative sample field area, positive sample field area, and recovery delay field area, and output a knowledge conflict record set.

[0039] A knowledge conflict record set is generated by associating and packaging initial root cause characteristics, initial treatment operations, and actual effective operation data. Specifically, a standardized data structure containing four data field areas is established in memory resources.

[0040] The acquired alarm data set is written into the environmental feature field area of ​​the data structure as the baseline data of the scenario in which the conflict occurred, and is used to reproduce the fault state.

[0041] Next, the initial root cause characteristics and initial treatment operations are written into the negative sample field area of ​​the data structure and marked as failure decision combination data.

[0042] Then, the actual effective operation data is written into the positive sample field area of ​​the data structure and marked as the correction target data.

[0043] Furthermore, by reading the timestamp, the time difference between the initial handling operation execution time and the actual effective operation data execution time is calculated, and this time difference is written into the recovery delay field area of ​​the data structure. The program encapsulates the data in the environment feature field area, negative sample field area, positive sample field area, and recovery delay field area, and outputs a knowledge conflict record set.

[0044] Furthermore, the steps for dynamically adjusting the causal association weight values ​​in the causal weight matrix based on the knowledge conflict record set include: Analyze the knowledge conflict record set to read the specific alarm feature categories that caused the misdiagnosis and the categories of the root cause nodes of the misjudged errors; Based on specific alarm feature categories and error root cause node categories, locate the corresponding error causal association weight values ​​in the causal weight matrix; Apply a preset penalty attenuation factor to the weight value of the erroneous causal association to reduce the directional strength of a specific alarm feature category to the erroneous root cause node category; Read the actual valid operation data in the knowledge conflict record set, and deduce the true root cause node category in reverse according to the preset mapping rules; Locate the correct causal association weight values ​​between specific alarm feature categories and the true root cause node categories in the causal weight matrix; Apply a preset reward enhancement factor to the correct causal association weight values ​​to complete the numerical update and iteration of the model matrix.

[0045] The causal association weight values ​​in the causal weight matrix are dynamically adjusted based on the knowledge conflict record set. First, the analysis module parses the knowledge conflict record set, reads the specific alarm feature categories that caused the misdiagnosis, and the categories of the root cause nodes of the misjudged errors.

[0046] Based on specific alarm feature categories and error root cause node categories, the program locates the corresponding error causal association weight values ​​in the causal weight matrix within the matrix library. A preset penalty attenuation factor is applied to these error causal association weight values ​​to reduce the directional strength of the specific alarm feature category towards the error root cause node category. Subsequently, the program reads actual valid operational data from the knowledge conflict record set and, according to preset mapping rules, reverse-engineers the true root cause node category.

[0047] The correct causal association weight values ​​corresponding to specific alarm feature categories and the true root cause node categories in the causal weight matrix are located. Finally, a preset reward enhancement factor is applied to the correct causal association weight values ​​to complete the numerical update iteration of the model matrix.

[0048] Furthermore, the update logic for applying a preset penalty decay factor to the weight values ​​of incorrect causal associations and a preset reward enhancement factor to the weight values ​​of correct causal associations includes: ; ; In the formula, This represents the updated erroneous causal association weight value. This indicates the weight value of the erroneous causal association before the update. This represents the preset penalty decay factor. This represents the updated correct causal association weight value. This represents the correct causal association weight value before the update. This represents the preset reward enhancement factor; The numerical parameters in the target data structure are directly replaced and written using exponential decay-based multiplication rules and nonlinear addition rules. The updated causal correlation strength values ​​of each item in the matrix are strictly constrained within a preset legal data range, and the final set of converged matrix parameters is output.

[0049] The process of applying a preset penalty decay factor to the weight values ​​of incorrect causal associations and a preset reward enhancement factor to the weight values ​​of correct causal associations is executed using the following formula. as well as .

[0050] This logic employs mature parameter attenuation and compensation rules. During the calculation, This represents the weight value of the erroneous causal association before the update, applied... This represents the preset penalty decay factor (set to 0.2), which weakens erroneous association paths and achieves... This represents the updated weight value for erroneous causal relationships. Similarly, This represents the correct causal association weight value before the update, applied... This indicates the preset reward enhancement factor (set to 0.15), calculated as follows: This represents the updated correct causal association weight values. This scheme uses a multiplication rule based on exponential decay and a non-linear addition rule to directly replace and write the numerical parameters in the target data structure, strictly constraining the updated causal association strength values ​​of each item in the matrix within a preset valid data range, and outputting the final converged set of matrix parameters.

[0051] like Figure 2 The diagram shown is a structural schematic of a data analysis-based proactive early warning system for power grid information operation and maintenance provided in this application embodiment. The data analysis-based proactive early warning system for power grid information operation and maintenance provided in this application embodiment includes: a feature extraction module, a window preset module, a data comparison module, a data recovery module, a data packaging module, a weight correction module, and a matrix application module. The feature extraction module is used to acquire the alarm data set output by the power grid operation system, diagnose and extract initial root cause features from the alarm data set, and match and execute initial handling operations in the handling strategy library based on the initial root cause features. The window preset module is used to open an observation window of preset duration after the initial processing operation is executed, and collect power grid business indicator data in the observation window and store it in a temporary storage area. The data comparison module is used to compare the power grid business indicator data with the preset recovery target threshold at the end of the observation window. If the power grid business indicator data does not reach the preset recovery target threshold, the initial handling operation is determined to be invalid and the reverse tracking logic is triggered. The data recovery module is used to respond to the reverse tracing logic, read the manual intervention records in the unified operation and maintenance platform, and extract the actual effective operation data that enables the power grid business to return to normal from the manual intervention records; The data packaging module is used to associate and package the initial root cause features, initial treatment operations, and actual effective operation data to generate a knowledge conflict record set. The weight correction module is used to dynamically correct the causal association weight values ​​in the causal weight matrix according to the knowledge conflict record set, and update the recommended priority sequence in the disposal strategy library according to the actual effective operation data. The matrix application module is used to obtain real-time alarm data for the next time period, and apply the corrected causal weight matrix and the updated handling strategy library to issue warnings.

[0052] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0053] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0054] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0055] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0056] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0057] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A proactive early warning method for power grid information operation and maintenance based on data analysis, characterized in that, Includes the following steps: Obtain the alarm data set output by the power grid operation system, diagnose the alarm data set to extract initial root cause features, match the initial handling operation in the handling strategy library based on the initial root cause features and execute it; After the initial handling operation is executed, a pre-set observation window is opened, and power grid business indicator data are collected within the observation window and stored in a temporary storage area. At the end of the observation window, the power grid business indicator data is compared with the preset recovery target threshold. If the power grid business indicator data does not reach the preset recovery target threshold, the initial handling operation is determined to be invalid and the reverse tracing logic is triggered. In response to the reverse tracing logic, read the manual intervention records in the unified operation and maintenance platform, and extract the actual effective operation data that restored the power grid business to normal from the manual intervention records; The initial root cause characteristics, initial treatment operations, and actual effective operation data are associated and packaged to generate a knowledge conflict record set. The causal association weight values ​​in the causal weight matrix are dynamically adjusted based on the knowledge conflict record set, and the recommended priority sequence in the disposal strategy library is updated based on actual effective operational data. Obtain real-time alarm data for the next time period, and apply the revised causal weight matrix and the updated handling strategy library to issue early warnings.

2. The proactive early warning method for power grid information operation and maintenance based on data analysis as described in claim 1, characterized in that, The steps for obtaining the alarm data set output by the power grid operation system and extracting initial root cause features from the alarm data set include: Extract alarm feature values ​​from each dimension of the alarm dataset and construct an alarm feature vector; Read the causal weight matrix stored in the database. The row dimension of the causal weight matrix represents the alarm feature category, the column dimension represents the root cause node category, and the matrix elements represent the causal association strength between the corresponding alarm feature category and the root cause node category. The alarm feature vector is input into the feature mapping logic, and the dot product of the alarm feature vector and the causal weight matrix is ​​calculated through the feature mapping logic to obtain the initial root cause distribution vector. Extract the root cause node category corresponding to the element with the largest value in the initial root cause distribution vector, and mark the root cause node category as the initial root cause feature; In the step of constructing alarm feature vectors, a time sliding window is used to extract the original alarm quantity within a continuous timestamp according to a preset time step, and the extracted original alarm quantity is normalized and mapped to eliminate the difference in dimensions between different alarm sources and generate a standardized input data structure.

3. The proactive early warning method for power grid information operation and maintenance based on data analysis as described in claim 2, characterized in that, The steps to obtain the initial root cause distribution vector by calculating the dot product of the alarm feature vector and the causal weight matrix through feature mapping logic include: The initial root cause distribution vector is obtained by calculating the dot product between the alarm feature vector and the causal weight matrix through feature mapping logic. The calculation formula is as follows: ； In the formula, Represents the initial root cause distribution vector. This represents the alarm feature vector. Represents the causal weight matrix. This represents the preset confidence adjustment coefficient for the current state, with a value greater than zero and less than one. This represents the historical root cause distribution vector of the previous time period. This represents the preset historical state forgetting coefficient, with a value range greater than zero and less than one. The feature mapping logic retains the accumulated state information in the previous operation of the system by introducing the historical root cause distribution vector and the historical state forgetting coefficient, and integrates them to generate a low-level diagnostic data structure with time-series smoothing properties.

4. The proactive early warning method for power grid information operation and maintenance based on data analysis as described in claim 1, characterized in that, If the power grid business indicator data does not reach the preset recovery target threshold, the steps to determine that the initial handling operation is invalid and trigger the reverse tracing logic include: Separate the multi-dimensional indicators in the temporary storage area to obtain transaction success rate data sequences and response time data sequences; Obtain the target baseline values ​​for transaction success rate and response time calculated within the historical normal operation cycle as preset recovery target thresholds; Calculate the first absolute difference between the mean of the trade success rate data series and the target baseline value for trade success rate; Calculate the second absolute difference between the mean of the response time data series and the target baseline value of the response time; Determine whether the first absolute difference is greater than the first permissible error limit, and determine whether the second absolute difference is greater than the second permissible error limit; When the first absolute difference is greater than the first allowable error limit, or the second absolute difference is greater than the second allowable error limit, a target deviation indicator is generated; The target deviation indicator is used as the input control signal to send a reverse tracing start command to the power grid operation and maintenance control service system, thereby activating the reverse tracing logic.

5. The proactive early warning method for power grid information operation and maintenance based on data analysis as described in claim 1, characterized in that, The steps involved in extracting actual, effective operational data from manual intervention records to restore normal power grid operations include: Send a data retrieval request with the fault occurrence timestamp to the log interface of the unified operation and maintenance platform; Receive the work order processing text stream returned by the log interface after the timestamp of the fault occurrence; Perform word segmentation and slicing operations on the work order processing text stream to obtain a sequence of word groups containing multiple sets of operation terms; By using a pre-established standard dictionary of operation and maintenance, entity recognition and matching are performed on the word sequence, descriptive words that are detached from the semantics of operation are stripped away, and standardized operation instructions that represent intervention actions are extracted. Extract the system status return code after the standardized operation command is executed. If the system status return code indicates that the service has returned to normal, then confirm the corresponding standardized operation command and output it as the actual valid operation data, thereby capturing the underlying intervention control command that indicates the service recovery.

6. The proactive early warning method for power grid information operation and maintenance based on data analysis as described in claim 1, characterized in that, The steps to generate a knowledge conflict record set include: Establish a standardized data structure containing four data field areas; The acquired alarm data set is written into the environmental feature field area of ​​the data structure as the baseline data for the scenario context in which the conflict occurred; The initial root cause characteristics and initial treatment operations are written into the negative sample field area of ​​the data structure and marked as failure decision combination data. Write the actual effective operation data into the positive sample field area of ​​the data structure and mark it as the correction target data; Calculate the time difference between the execution time of the initial handling operation and the execution time of the actual effective operation data, and write the time difference into the recovery delay field area of ​​the data structure; Encapsulate the data in the environmental feature field area, negative sample field area, positive sample field area, and recovery delay field area, and output a knowledge conflict record set.

7. The proactive early warning method for power grid information operation and maintenance based on data analysis as described in claim 1, characterized in that, The steps for dynamically revising the causal association weight values ​​in the causal weight matrix based on the knowledge conflict record set include: Analyze the knowledge conflict record set to read the specific alarm feature categories that caused the misdiagnosis and the categories of the root cause nodes of the misjudged errors; Based on specific alarm feature categories and error root cause node categories, locate the corresponding error causal association weight values ​​in the causal weight matrix; A preset penalty attenuation factor is applied to the weight value of the erroneous causal association to reduce the directional strength of a specific alarm feature category to the erroneous root cause node category; Read the actual valid operation data in the knowledge conflict record set, and deduce the true root cause node category in reverse according to the preset mapping rules; Locate the correct causal association weight values ​​between specific alarm feature categories and the true root cause node categories in the causal weight matrix; Apply a preset reward enhancement factor to the correct causal association weight values ​​to complete the numerical update and iteration of the model matrix.

8. A data-analysis-based proactive early warning system for power grid information operation and maintenance, used to implement the data-analysis-based proactive early warning method for power grid information operation and maintenance as described in any one of claims 1-7, characterized in that, include: Feature extraction module, window preset module, data comparison module, data recovery module, data packaging module, weight correction module, matrix application module; The feature extraction module is used to acquire the alarm data set output by the power grid operation system, diagnose and extract initial root cause features from the alarm data set, and match and execute initial handling operations in the handling strategy library based on the initial root cause features. The window preset module is used to open an observation window of preset duration after the initial processing operation is executed, and collect power grid business indicator data in the observation window and store it in a temporary storage area. The data comparison module is used to compare the power grid business indicator data with the preset recovery target threshold at the end of the observation window. If the power grid business indicator data does not reach the preset recovery target threshold, the initial handling operation is determined to be invalid and the reverse tracking logic is triggered. The data recovery module is used to respond to the reverse tracing logic, read the manual intervention records in the unified operation and maintenance platform, and extract the actual effective operation data that enables the power grid business to return to normal from the manual intervention records; The data packaging module is used to associate and package the initial root cause features, initial treatment operations, and actual effective operation data to generate a knowledge conflict record set. The weight correction module is used to dynamically correct the causal association weight values ​​in the causal weight matrix according to the knowledge conflict record set, and update the recommended priority sequence in the disposal strategy library according to the actual effective operation data. The matrix application module is used to obtain real-time alarm data for the next time period, and apply the corrected causal weight matrix and the updated handling strategy library to issue warnings.

Images