Storage medium, industrial centralized control accident recalling method and system

Abstract

The invention belongs to the technical field of power grid operation and maintenance, and particularly relates to a storage medium and an industrial centralized control accident recalling method and system. The method comprises the following steps: acquiring real-time power grid operation data, and constructing a current actual operation power grid model according to the real-time power grid operation data; when an accident occurs, the power grid model and the accident section data which are actually operated at the accident occurrence moment are stored; when accident recall needs to be carried out, accident inversion is carried out based on the power grid model actually running at the accident occurrence moment and the accident section data, or simulation is carried out on equipment parameters and states in the model based on a preset power grid model according to the selected simulation event, so that accident occurrence is simulated; the simulation data before and after the simulation accident are recorded, accident inversion is performed by combining the simulation data before and after the simulation accident so as to analyze the accident cause, the accuracy and efficiency of accident analysis are improved, a real power grid system does not need to be operated, and normal operation of a power grid is not affected.

Description

A storage medium, an industrial centralized control system accident tracing method and system Technical Field

[0001] This invention belongs to the field of power grid operation and maintenance technology, specifically relating to a storage medium, an industrial centralized control accident tracing method and system. Background Technology

[0002] In power automation systems, the accident recall function retrieves data from a period prior to an accident to locate the cause, which is crucial for analyzing the accident's origin and ensuring the safe and stable operation of the power grid. For example, Chinese patent document CN104484836B discloses a power grid fault auxiliary decision-making visualization system and method. The acquisition module collects system parameters in real time and transmits them to the recording module for real-time storage. When a fault occurs, the alarm module sends alarm information to the human-machine subsystem. Upon receiving the alarm information, the human-machine subsystem sends system parameter information to the accident recall module, which then retrieves data from a period prior to the fault to locate the accident.

[0003] Existing accident reconstructive methods typically involve quickly reviewing cross-sectional data of the power grid at the time of the accident, and then using this data to perform accident inversion analysis based on the current power grid architecture. However, if the power grid has undergone upgrades or significant changes after the accident, resulting in alterations to the grid structure, inconsistencies may arise between the equipment models and circuit diagrams of the current master station system and those in the cross-sectional data at the time of the accident. This affects the accuracy of accident inversion and consequently, the accuracy of accident analysis. Furthermore, when dispatchers use the master station system to remotely control field switches to trigger accident reconstructive analysis, it can lead to power outages and disrupt normal power grid operation. Summary of the Invention

[0004] The purpose of this invention is to provide a storage medium, an industrial centralized control accident tracing method and system to solve the problems of mismatch between power grid architecture and accident section data and the need for power outages to facilitate accident tracing, which affect the normal operation of the power grid.

[0005] To address the aforementioned technical problems, this invention provides an industrial centralized control system accident replay method, comprising: acquiring real-time power grid operation data; constructing a currently operating power grid model based on the real-time power grid operation data; when an accident occurs, saving the actual operating power grid model and accident section data at the time of the accident; when accident replay is required, performing accident inversion based on the actual operating power grid model and accident section data at the time of the accident to analyze the cause of the accident, or simulating the equipment parameters and states in the model based on a preset power grid model according to a selected simulation event to simulate the occurrence of an accident, recording the simulation data before and after the simulated accident, and performing accident inversion by combining the simulation data before and after the simulated accident to analyze the cause of the accident.

[0006] Furthermore, the method also includes saving the actual operating power grid model and cross-sectional data at the moment of the significant change in power grid state when a significant change in power grid state is detected, for the purpose of accident reconstructive analysis.

[0007] Furthermore, real-time power grid operation data is input into a real-time data stream anomaly detection model to detect whether an accident has occurred in the current power grid and whether the power grid state has undergone significant changes. The real-time data stream anomaly detection model is obtained by training a neural network model using power grid operation data during normal operation, when the power grid state has undergone significant changes, and when an accident occurs.

[0008] Furthermore, the preset power grid model can be the currently operating power grid model, the power grid model actually operating at the time of the accident, or the research-state power grid model. The research-state power grid model is obtained by adjusting the currently operating power grid model according to research needs.

[0009] Furthermore, the method also includes analyzing real-time power grid operation data using an accident prediction model to output accident early warning information and risk assessment reports; the accident prediction model is obtained by analyzing a dataset constructed using time series analysis algorithms and time series data from historical accident occurrences.

[0010] Furthermore, the method also includes processing the power grid operation data and dispatcher operation data at the time of the accident using an accident handling recommendation model after the accident occurs, and outputting an accident handling plan; the accident handling recommendation model is established based on the dispatcher's operation data during the accident handling process, historical accident handling cases, and recommendation algorithms.

[0011] Furthermore, the method also includes constructing a knowledge graph-based accident causal chain model based on accident time, accident type, and equipment name, and generating a visual report of the accident evolution path by combining equipment topology relationships and historical accidents. The visual report is presented in the form of charts, reports, or change curves.

[0012] Furthermore, the cross-sectional data saved at the time of the accident includes plant data associated with the accident and equipment data with high correlation to the accident. The equipment data with high correlation to the accident refers to the data of the top N equipment obtained by sorting the equipment data from highest to lowest accident correlation using a machine learning algorithm.

[0013] The beneficial effects of the above technical solution are as follows: This invention is an improved invention. By constructing a real-time model of the currently operating power grid, it saves the power grid model and accident section data at the corresponding moment when a power grid accident is detected. During accident replay, it directly calls the power grid model and accident section data corresponding to the moment the accident occurred to perform accident inversion, or it triggers the accident by simulating the equipment parameters and states in the preset model. By recording the simulation data before and after the simulated accident, and combining it with the section data saved at the moment the accident occurred, it performs accident inversion to analyze the cause of the accident. This ensures that the power grid model used in the accident inversion analysis is consistent with the section data at the moment the accident occurred, improving the accuracy and efficiency of accident analysis. Furthermore, by performing accident inversion using the directly saved power grid model and section data, or by performing accident replay through a simulation method that simulates the occurrence of an accident, it is not necessary to operate the real power grid system, avoiding power outages caused by initiating accident replay and not affecting the normal operation of the power grid.

[0014] To address the aforementioned technical problems, this invention also provides an industrial centralized control accident replay system, including a processor. The processor implements an industrial centralized control accident replay method, which includes: acquiring real-time power grid operation data; constructing a currently operating power grid model based on the real-time power grid operation data; when an accident occurs, saving the actual operating power grid model and accident section data at the time of the accident; when accident replay is required, performing accident inversion based on the actual operating power grid model and accident section data at the time of the accident to analyze the cause of the accident, or simulating the equipment parameters and states in the model based on a preset power grid model according to a selected simulated event to simulate the occurrence of an accident, recording the simulation data before and after the simulated accident, and performing accident inversion by combining the simulation data before and after the simulated accident to analyze the cause of the accident.

[0015] Furthermore, the method also includes saving the actual operating power grid model and cross-sectional data at the moment of the significant change in power grid state when a significant change in power grid state is detected, for the purpose of accident reconstructive analysis.

[0016] Furthermore, real-time power grid operation data is input into a real-time data stream anomaly detection model to detect whether an accident has occurred in the current power grid and whether the power grid state has undergone significant changes. The real-time data stream anomaly detection model is obtained by training a neural network model using power grid operation data during normal operation, when the power grid state has undergone significant changes, and when an accident occurs.

[0017] Furthermore, the preset power grid model can be the currently operating power grid model, the power grid model actually operating at the time of the accident, or the research-state power grid model. The research-state power grid model is obtained by adjusting the currently operating power grid model according to research needs.

[0018] Furthermore, the method also includes analyzing real-time power grid operation data using an accident prediction model to output accident early warning information and risk assessment reports; the accident prediction model is obtained by analyzing a dataset constructed using time series analysis algorithms and time series data from historical accident occurrences.

[0019] Furthermore, the method also includes processing the power grid operation data and dispatcher operation data at the time of the accident using an accident handling recommendation model after the accident occurs, and outputting an accident handling plan; the accident handling recommendation model is established based on the dispatcher's operation data during the accident handling process, historical accident handling cases, and recommendation algorithms.

[0020] Furthermore, the method also includes constructing a knowledge graph-based accident causal chain model based on accident time, accident type, and equipment name, and generating a visual report of the accident evolution path by combining equipment topology relationships and historical accidents. The visual report is presented in the form of charts, reports, or change curves.

[0021] Furthermore, the cross-sectional data saved at the time of the accident includes plant data associated with the accident and equipment data with high correlation to the accident. The equipment data with high correlation to the accident refers to the data of the top N equipment obtained by sorting the equipment data from highest to lowest accident correlation using a machine learning algorithm.

[0022] To address the aforementioned technical problems, the present invention also provides a storage medium storing computer-executable instructions. These instructions, when executed, implement an industrial centralized control system accident replay method. The method includes: acquiring real-time power grid operation data; constructing a currently operating power grid model based on the real-time power grid operation data; saving the actual operating power grid model and accident section data at the time of the accident when an accident occurs; and, when accident replay is required, performing accident inversion based on the actual operating power grid model and accident section data at the time of the accident to analyze the cause of the accident, or simulating the equipment parameters and states in the model based on a preset power grid model according to a selected simulated event to simulate an accident, recording the simulation data before and after the simulated accident, and combining the simulation data before and after the simulated accident to perform accident inversion to analyze the cause of the accident.

[0023] Furthermore, the method also includes saving the actual operating power grid model and cross-sectional data at the moment of the significant change in power grid state when a significant change in power grid state is detected, for the purpose of accident reconstructive analysis.

[0024] Furthermore, real-time power grid operation data is input into a real-time data stream anomaly detection model to detect whether an accident has occurred in the current power grid and whether the power grid state has undergone significant changes. The real-time data stream anomaly detection model is obtained by training a neural network model using power grid operation data during normal operation, when the power grid state has undergone significant changes, and when an accident occurs.

[0025] Furthermore, the preset power grid model can be the currently operating power grid model, the power grid model actually operating at the time of the accident, or the research-state power grid model. The research-state power grid model is obtained by adjusting the currently operating power grid model according to research needs.

[0026] Furthermore, the method also includes analyzing real-time power grid operation data using an accident prediction model to output accident early warning information and risk assessment reports; the accident prediction model is obtained by analyzing a dataset constructed using time series analysis algorithms and time series data from historical accident occurrences.

[0027] Furthermore, the method also includes processing the power grid operation data and dispatcher operation data at the time of the accident using an accident handling recommendation model after the accident occurs, and outputting an accident handling plan; the accident handling recommendation model is established based on the dispatcher's operation data during the accident handling process, historical accident handling cases, and recommendation algorithms.

[0028] Furthermore, the method also includes constructing a knowledge graph-based accident causal chain model based on accident time, accident type, and equipment name, and generating a visual report of the accident evolution path by combining equipment topology relationships and historical accidents. The visual report is presented in the form of charts, reports, or change curves.

[0029] Furthermore, the cross-sectional data saved at the time of the accident includes plant data associated with the accident and equipment data with high correlation to the accident. The equipment data with high correlation to the accident refers to the data of the top N equipment obtained by sorting the equipment data from highest to lowest accident correlation using a machine learning algorithm.

[0030] The beneficial effects of the above technical solution are as follows: This invention is an improved invention. By constructing a real-time model of the currently operating power grid, it saves the power grid model and accident section data at the corresponding moment when a power grid accident is detected. During accident replay, it directly calls the power grid model and accident section data corresponding to the moment the accident occurred to perform accident inversion, or it triggers an accident by simulating the operation of equipment parameters and states in a preset model. By recording the simulation data before and after the simulated accident, and combining it with the section data saved at the moment the accident occurred, it performs accident inversion to analyze the cause of the accident. This ensures that the power grid model used in the accident inversion analysis is consistent with the section data at the moment the accident occurred, improving the accuracy and efficiency of accident analysis. Furthermore, by performing accident inversion through directly saved power grid model and section data, or by performing accident replay through a simulation method that simulates the occurrence of an accident, it is not necessary to operate the real power grid system, avoiding power outages caused by initiating accident replay, and not affecting the normal operation of the power grid. Attached Figure Description

[0031] Figure 1 is a schematic diagram of the principle of the industrial centralized control accident tracing system according to an embodiment of the present invention; Figure 2 is an interaction diagram of each module of the industrial centralized control accident tracing system according to an embodiment of the present invention; Figure 3 is a data storage and processing flowchart according to an embodiment of the present invention; Figure 4 is a flowchart of accident cause diagnosis according to an embodiment of the present invention; Figure 5 is a flowchart of accident prediction according to an embodiment of the present invention; Figure 6 is a flowchart of accident handling recommendation according to an embodiment of the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of the present invention clearer, the specific embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0033] This invention improves the accuracy and efficiency of accident reconstructive analysis by constructing and simulating a multi-state power grid model to ensure that the power grid model used in accident inversion is consistent with the cross-sectional data at the time of the accident.

[0034] System Implementation An industrial centralized control accident tracing system according to the present invention includes a processor, as shown in Figure 1. The processor (server in Figure 1) includes a polymorphic model construction module, a simulation module, a data selection and storage module, a data analysis and display module, and an artificial intelligence function module. It also includes a user terminal and data acquisition equipment. The user terminal deploys an operation interface for receiving operation commands sent by the user, such as simulation operation commands and data query commands, and interacting with the processor. The data acquisition equipment includes a data collector and sensors for collecting power grid operation data and transmitting it to the processor. The interaction between the modules of the industrial centralized control accident tracing system is shown in Figure 2, and will be described in detail below.

[0035] The multi-state model construction module is used to build multi-state power grid models, including real-time power grid models and historical power grid models. The real-time power grid model is a model of the power grid currently in operation, built based on real-time power grid operation data, reflecting the current actual operating scenario of the system. The historical power grid model includes models of the power grid actually operating at the time of an accident (saved when an accident occurs), models of the power grid actually operating at the time of a significant change in power grid state (saved when a significant change in power grid state is detected), and models of the power grid actually operating at corresponding times at periodic nodes. The historical power grid model reflects the power grid system's scenarios at different historical moments. A real-time data stream anomaly detection model dynamically identifies real-time power grid operation data to detect the occurrence of specific events (such as a current power grid accident or a significant change in power grid state). When a specific event is detected, the system automatically saves the current power grid model and relevant cross-sectional data, creating a historical scenario. When accident re-analysis is needed, accident inversion is performed based on the power grid model actually operating at the time of the accident and the accident cross-sectional data to analyze the causes of the accident.

[0036] The real-time data stream anomaly detection model is trained on a neural network model using power grid operation data during normal operation, when there are significant changes in the power grid state, and when an accident occurs. The neural network model can employ LSTM, decision tree models, or GRU models, among others.

[0037] The multi-state power grid model also includes the research state power grid model. The research state power grid model is created by dispatchers or researchers based on the real-time power grid model after adjustments. Accident analysis and research can be carried out by adjusting parameters and performing hypothetical simulations on the research state power grid model.

[0038] The simulation module simulates the equipment parameters and states in a pre-set power grid model based on the selected simulation event to simulate an accident. It records simulation data before and after the accident, and combines this data with cross-sectional data saved at the time of the accident to perform accident inversion analysis to identify the cause. The software-simulated event triggers the accident replay function, providing dispatchers with a safe environment for analysis and practice. The pre-set power grid model can be the currently operating power grid model, the power grid model actually operating at the time of the accident, or the power grid model under study.

[0039] Specifically, the software interface includes simulation operation options. After a dispatcher initiates a simulation, the simulation module calls a preset power grid model from the polymorphic model construction module. The dispatcher can select simulation events (such as switch changes and measured value exceeding limits). When a simulation event is selected, the equipment parameters and states in the model change accordingly to simulate an accident. The system triggers the accident replay function according to preset logic, records the simulation data before and after the simulated accident, and performs accident inversion display, consistent with the actual accident handling process. For example, when accident replay is required, the simulation module simulates the equipment parameters and states in the model based on the actual operating power grid model at the time of the accident to simulate the accident, records the simulation data before and after the simulated accident, and combines it with the cross-sectional data saved at the time of the accident to perform accident inversion to analyze the cause of the accident. The simulation process does not affect the actual operation of the power grid and can simulate various complex accident scenarios, helping dispatchers become familiar with the accident handling process.

[0040] By constructing multi-state models and conducting simulation operations, the power grid model used in accident inversion is ensured to be consistent with the cross-sectional data at the time of the accident, thereby improving the accuracy of accident analysis. The simulation module provides dispatchers with a safe practice environment, which helps improve their operational skills and accident handling capabilities.

[0041] The data selection and storage module saves accident section data at the moment of the accident and section data at moments of significant changes in the power grid status. When saving accident section data, it employs an accident data selection method based on the power grid model, saving only data from plants and equipment associated with the accident. This avoids the database space waste and performance degradation caused by saving the entire network's data. Simultaneously, the collected data is appropriately compressed and archived for long-term storage and retrieval. This power grid model-based data selection method effectively reduces data storage volume and improves database performance and efficiency.

[0042] As a preferred implementation, data from equipment nodes with high relevance to the accident is prioritized for retention. As shown in Figure 3, when an accident occurs, the system quickly identifies associated power plants and equipment based on the power grid model. The system then uses a machine learning algorithm to sort the equipment data from highest to lowest relevance to the accident, obtaining the data of the top N devices, which are stored as the equipment data with high accident relevance. The machine learning algorithm can employ reinforcement learning (RL) algorithms. First, an auto-encoder is used to extract features from the high-dimensional data of the equipment. Then, a reinforcement learning algorithm is used to process and sort the feature data, thereby dynamically adjusting the data storage order and prioritizing the retention of equipment node data with high accident relevance. Preferably, only remote signaling and telemetry data from associated power plants and the status information of related equipment are collected and stored.

[0043] Efficient data compression algorithms are used to compress the data before storing it in the database, reducing storage space usage. Accident recap data is archived regularly, and historical data is stored on dedicated storage media for long-term preservation and retrieval.

[0044] The data analysis and visualization module provides powerful accident data analysis capabilities, offering rich data visualization features and supporting multi-dimensional queries, analysis, and visualization. Data is presented in the form of charts, reports, or trend curves, intuitively reflecting the development process and causes of accidents, helping dispatchers make quick and accurate decisions.

[0045] Specifically, dispatchers can use the system interface to input query conditions (such as accident time, accident type, and equipment name) to retrieve relevant accident recollection data. The data analysis and visualization module extracts data from the data selection and storage module based on the query conditions, constructs an accident causal chain model based on a knowledge graph, and combines equipment topology relationships with a historical accident database (such as total accident signal → protection tripping → protection malfunction). It then uses a graph neural network (GNN) to perform subgraph matching on real-time data, automatically generating a visual report of the accident evolution path. This data is displayed in the form of charts (such as line graphs and bar charts), reports (such as reports on changes in electrical quantities before and after the accident, reports on changes in equipment status, etc.), or curves, facilitating analysis by dispatchers. The data analysis and visualization module provides multi-dimensional data analysis and visualization, helping dispatchers quickly understand the accident situation and formulate appropriate accident handling plans.

[0046] The artificial intelligence module is used to achieve intelligent diagnosis of accident causes, accident prediction, and intelligent recommendation of accident handling solutions.

[0047] Intelligent Diagnosis: As shown in Figure 4, a large amount of historical accident data is collected. After cleaning, labeling, and preprocessing the data, a machine learning algorithm is trained to obtain a correlation model between accident characteristics and causes. When an accident occurs, the real-time collected accident data is input into the trained correlation model, which outputs possible accident causes and their probabilities.

[0048] Machine learning algorithms can employ decision tree models, neural network models, or other network models. By performing in-depth analysis of accident recollection data through correlation models, the system can quickly diagnose possible causes of new accidents, providing decision-making references for dispatchers.

[0049] Intelligent prediction: Based on time series analysis and deep learning technology, the system monitors and analyzes power grid operation data in real time, predicts the types and timing of possible accidents, issues early warning information in advance, and helps dispatchers take preventive measures.

[0050] Specifically, as shown in Figure 5, a dataset is constructed using time series analysis algorithms and historical accident time series data to obtain an accident prediction model. The model is optimized and adjusted by continuously updating real-time data. Then, the trained accident prediction model is used to analyze real-time power grid operation data. When a potential accident is predicted, accident warning information and risk assessment reports are output. Power grid operation data includes voltage, current, and power. Time series analysis algorithms can employ Support Vector Machine (SVM), LSTM, or Random Forest, among others. For example, SVM classifies or regresses data by finding an optimal hyperplane. In time series analysis, time series data can be transformed into feature vectors, which are then used for accident prediction.

[0051] Intelligent Recommendation: Based on the dispatcher's operating habits and historical experience, the recommendation algorithm provides the dispatcher with recommendations for accident handling solutions, helping them to handle accidents faster and more accurately.

[0052] Specifically, as shown in Figure 6, operational data of dispatchers during accident handling and historical accident handling cases are collected, and the data is analyzed and organized. Based on the operational data of dispatchers during accident handling, historical accident handling cases, and recommendation algorithms, an accident handling recommendation model is established. The recommendation algorithm can be a collaborative filtering algorithm or a content-based recommendation algorithm, etc. During the handling process after an accident occurs, the accident handling recommendation model is used to determine the current accident characteristics based on the power grid operation data at the time of the accident. After processing according to the current accident characteristics and the dispatchers' operational status, an accident handling plan is output.

[0053] The integration of artificial intelligence enables the accident replay system to have intelligent diagnosis, intelligent prediction, and intelligent recommendation capabilities, further improving the intelligence level and accident handling capabilities of the centralized control system and reducing the losses caused by accidents.

[0054] Method Implementation Method: An industrial centralized control accident recall method of the present invention includes the following steps: 1. Saving power grid model and cross-sectional data.

[0055] Acquire real-time power grid operation data and construct a current, actually operating power grid model based on this data. When an accident occurs, save the actual operating power grid model and accident section data at the time of the accident.

[0056] It can also save the actual operating power grid model and cross-sectional data at the time when a significant change in the power grid state is detected, as well as the actual operating power grid model and cross-sectional data at the corresponding time at periodic nodes.

[0057] The system utilizes a real-time data stream anomaly detection model to dynamically identify real-time power grid operation data, detecting the occurrence of specific events (such as current power grid accidents and significant changes in power grid status). When a specific event is detected, the current power grid model and relevant cross-sectional data are automatically saved. The real-time data stream anomaly detection model is trained on a neural network model using power grid operation data from normal operation, when there are significant changes in power grid status, and when accidents occur. The neural network model can employ LSTM, decision tree models, or GRU models, among others.

[0058] It can also create a research-state power grid model based on the real-time power grid model according to the research needs of dispatchers or researchers. Through parameter adjustment and assumption simulation of the research-state power grid model, accident simulation and accident analysis research can be carried out.

[0059] 2. Conduct a review of the accident.

[0060] When accident replay is required, accident inversion is performed based on the actual operating power grid model at the time of the accident and the accident section data to analyze the cause of the accident. Alternatively, based on the selected simulated event, a simulation is performed on the equipment parameters and states in the model using a preset power grid model to simulate the accident. The simulation data before and after the accident is recorded, and combined with the section data saved at the time of the accident, accident inversion is performed to analyze the cause of the accident. The preset power grid model can be the currently operating power grid model, the power grid model actually operating at the time of the accident, or the power grid model under study.

[0061] During the simulation, dispatchers can select the power grid model and simulated events to be simulated (such as switch changes and measured value exceeding limits). When a simulated event is selected, the equipment parameters and states in the model change accordingly to simulate an accident. The accident recall function is triggered according to the preset logic, and the simulation data before and after the simulated accident is recorded and stored. The accident is then displayed in reverse, consistent with the actual accident handling process.

[0062] For example, when it is necessary to conduct accident reconstruction, the equipment parameters and states in the model are simulated based on the actual operating power grid model at the time of the accident to simulate the occurrence of the accident. The simulation data before and after the accident is recorded, and the accident is inverted by combining the cross-sectional data saved at the time of the accident to analyze the cause of the accident.

[0063] When researchers conduct analysis based on their needs, they simulate the equipment parameters and states in the created research-state power grid model to simulate the occurrence of accidents, and record the simulation data before and after the simulated accident for the purpose of studying and analyzing the accident.

[0064] By constructing polymorphic models and conducting simulation operations, the power grid model used in accident inversion is ensured to be consistent with the cross-sectional data at the time of the accident, thereby improving the accuracy of accident analysis. The simulation module provides dispatchers with a safe practice environment, helping to improve their operational skills and accident handling capabilities. Furthermore, the simulation process does not affect the actual operation of the power grid and can simulate various complex accident scenarios, helping dispatchers become familiar with accident handling procedures.

[0065] When storing accident section data, a power grid model-based accident data selection method is adopted, storing only data related to the plants and equipment associated with the accident, including simulation data and section data stored during actual operation. This avoids the waste of database space and performance degradation caused by storing the entire network data. Simultaneously, the collected data is reasonably compressed and archived for long-term storage and retrieval. This power grid model-based data selection method effectively reduces data storage volume and improves database performance and efficiency.

[0066] In a preferred implementation, data from equipment nodes highly correlated with the accident is prioritized for retention. When an accident occurs, the system quickly identifies associated substations and equipment based on the power grid model. Using a machine learning algorithm, the system sorts the equipment data from highest to lowest accident relevance, storing the data of the top N devices as those with high accident relevance. The machine learning algorithm can employ reinforcement learning (RL). First, an auto-encoder is used to extract features from the high-dimensional data of the equipment. Then, a reinforcement learning algorithm is used to process and sort the feature data, dynamically adjusting the data storage order to prioritize the retention of equipment node data highly correlated with the accident. Preferably, only remote signaling and telemetry data from associated substations and the status information of related equipment are collected and stored.

[0067] Efficient data compression algorithms are used to compress the data before storing it in the database, reducing storage space usage. Accident recap data is archived regularly, and historical data is stored on dedicated storage media for long-term preservation and retrieval.

[0068] Based on the accident time, accident type, and equipment name, a knowledge graph-based accident causal chain model can be constructed. This model, combined with equipment topology relationships and historical accident data, generates a visualized report of the accident evolution path, presented in the form of charts, reports, or trend curves. Specifically, when querying relevant accident recollection data based on search criteria (such as accident time, accident type, and equipment name), the system extracts and saves the data according to the search criteria, constructs a knowledge graph-based accident causal chain model, and combines equipment topology relationships with a historical accident database (such as total accident signal → protection tripping → protection malfunction). A graph neural network (GNN) is used to perform subgraph matching on real-time data, automatically generating a visualized report of the accident evolution path. This data is then displayed in the form of charts (such as line charts and bar charts), reports (such as reports on changes in electrical quantities before and after the accident, reports on changes in equipment status), or trend curves. By providing multi-dimensional data analysis and visualization, this system helps dispatchers quickly understand the accident situation and formulate appropriate accident handling plans.

[0069] The method also includes intelligent diagnosis of accident causes, accident prediction, and intelligent recommendation of accident handling solutions.

[0070] Intelligent Diagnosis: A large amount of historical accident data is collected, cleaned, labeled, and preprocessed, then used to train machine learning algorithms to obtain a correlation model between accident characteristics and causes. When an accident occurs, real-time collected accident data is input into the trained correlation model, which outputs possible accident causes and their probabilities. Machine learning algorithms can employ decision tree models, neural network models, or other network models. Through in-depth analysis of accident recall data using the correlation model, the system can quickly diagnose possible accident causes when new accidents occur, providing decision-making references for dispatchers.

[0071] Intelligent prediction: Based on time series analysis and deep learning technology, the system monitors and analyzes power grid operation data in real time, predicts the types and timing of possible accidents, issues early warning information in advance, and helps dispatchers take preventive measures.

[0072] Specifically, a dataset is constructed using time series analysis algorithms and historical time series data from accident occurrences to obtain an accident prediction model. The model is optimized and adjusted by continuously updating real-time data. Then, the trained accident prediction model is used to analyze real-time power grid operation data. When a potential accident is predicted, accident warning information and risk assessment reports are output. Power grid operation data includes voltage, current, and power. Time series analysis algorithms can employ Support Vector Machines (SVM), LSTM, or Random Forests, among others. For example, SVM classifies or regresses data by finding an optimal hyperplane. In time series analysis, time series data can be transformed into feature vectors, which are then used for accident prediction.

[0073] Intelligent Recommendation: Based on the dispatcher's operating habits and historical experience, the recommendation algorithm provides the dispatcher with recommendations for accident handling solutions, helping them to handle accidents faster and more accurately.

[0074] Specifically, the system collects operational data from dispatchers during accident handling and historical accident handling cases, then analyzes and organizes the data. Based on this data, a recommendation model is established. The recommendation algorithm can employ collaborative filtering or content-based recommendation algorithms. During the handling of an accident, the accident handling recommendation model is used to determine the current accident characteristics from the power grid operation data at the time of the accident. After processing based on these characteristics and the dispatchers' actions, an accident handling plan is output. The integration of artificial intelligence enables the accident tracking system to possess intelligent diagnosis, intelligent prediction, and intelligent recommendation capabilities, further enhancing the intelligence level and accident handling capabilities of the centralized control system and reducing losses caused by accidents.

[0075] Storage Medium Implementation Method: A storage medium of the present invention stores computer-executable instructions, which, when executed, implement the steps in an industrial centralized control accident retrieval method, the method including: 1. Saving power grid model and cross-sectional data.

[0076] Acquire real-time power grid operation data and construct a current, actually operating power grid model based on this data. When an accident occurs, save the actual operating power grid model and accident section data at the time of the accident.

[0077] It can also save the actual operating power grid model and cross-sectional data at the time when a significant change in the power grid state is detected, as well as the actual operating power grid model and cross-sectional data at the corresponding time at periodic nodes.

[0078] When storing accident section data, a power grid model-based accident data selection method is adopted, storing only data related to the plants and equipment associated with the accident, including simulation data and section data stored during actual operation. This avoids the waste of database space and performance degradation caused by storing the entire network data. Simultaneously, the collected data is reasonably compressed and archived for long-term storage and retrieval. This power grid model-based data selection method effectively reduces data storage volume and improves database performance and efficiency.

[0079] The system utilizes a real-time data stream anomaly detection model to dynamically identify real-time power grid operation data, detecting the occurrence of specific events (such as current power grid accidents and significant changes in power grid status). When a specific event is detected, the current power grid model and relevant cross-sectional data are automatically saved. The real-time data stream anomaly detection model is trained on a neural network model using power grid operation data from normal operation, when there are significant changes in power grid status, and when accidents occur. The neural network model can employ LSTM, decision tree models, or GRU models, among others.

[0080] It can also create a research-state power grid model based on the real-time power grid model according to the research needs of dispatchers or researchers. Through parameter adjustment and assumption simulation of the research-state power grid model, accident simulation and accident analysis research can be carried out.

[0081] 2. Recollection of the accident.

[0082] When it is necessary to conduct accident replay, the accident inversion is performed based on the actual operating power grid model at the time of the accident and the accident section data to analyze the cause of the accident. Alternatively, based on the selected simulated event, the equipment parameters and states in the model are simulated according to the preset power grid model to simulate the occurrence of the accident. The simulation data before and after the simulated accident is recorded, and the accident inversion is performed in combination with the section data saved at the time of the accident to analyze the cause of the accident.

[0083] During the simulation, dispatchers can select the power grid model and simulated events to be simulated (such as switch changes and measured value exceeding limits). When a simulated event is selected, the equipment parameters and states in the model change accordingly to simulate an accident. The accident recall function is triggered according to the preset logic, and the simulation data before and after the simulated accident is recorded and stored. The accident is then displayed in reverse, consistent with the actual accident handling process.

[0084] It can construct an accident causal chain model based on knowledge graphs according to accident time, accident type and equipment name, and generate a visual report of accident evolution path by combining equipment topology relationship and historical accidents. The visual report is presented in the form of charts, reports or change curves.

[0085] The method also includes intelligent diagnosis of accident causes, accident prediction, and intelligent recommendation of accident handling solutions.

[0086] Intelligent Diagnosis: A large amount of historical accident data is collected, cleaned, labeled, and preprocessed, then used to train a machine learning algorithm to obtain a correlation model between accident characteristics and causes. When an accident occurs, the real-time collected accident data is input into the trained correlation model, which outputs possible accident causes and their probabilities.

[0087] Intelligent prediction: Based on time series analysis and deep learning technology, the system monitors and analyzes power grid operation data in real time, predicts the types and timing of possible accidents, issues early warning information in advance, and helps dispatchers take preventive measures.

[0088] Intelligent Recommendation: Based on the dispatcher's operating habits and historical experience, the recommendation algorithm provides the dispatcher with recommendations for accident handling solutions, helping them to handle accidents faster and more accurately.

[0089] This invention addresses the problems of mismatch between power grid models and accident section data, as well as the lack of simulation training scenarios, in existing accident recap functions by integrating multi-state power grid models, simulation, and artificial intelligence technologies into the accident recap function. At the same time, it provides dispatchers with a more intelligent and convenient training and learning environment, improves the accuracy and efficiency of accident analysis, enhances the intelligence level and accident handling capabilities of the centralized control system, reduces the losses caused by accidents, and meets the intelligence requirements of the centralized control system.

Claims

1. A method for tracing industrial centralized control system accidents, characterized in that, include: Acquire real-time power grid operation data and construct a current operating power grid model based on the real-time power grid operation data; when an accident occurs, save the actual operating power grid model and accident section data at the time of the accident; when accident replay is required, perform accident inversion based on the actual operating power grid model and accident section data at the time of the accident to analyze the cause of the accident, or simulate the equipment parameters and states in the model based on the selected simulated event and the preset power grid model to simulate the occurrence of the accident, record the simulation data before and after the simulated accident, and perform accident inversion by combining the simulation data before and after the simulated accident to analyze the cause of the accident.

2. The industrial centralized control system accident recall method according to claim 1, characterized in that, The method also includes saving the actual operating power grid model and cross-sectional data at the moment of the significant change in power grid status when a significant change in power grid status is detected, for the purpose of accident reconciliation.

3. The industrial centralized control system accident replay method according to claim 2, characterized in that, Real-time power grid operation data is input into a real-time data stream anomaly detection model to detect whether an accident has occurred in the current power grid and whether the power grid status has undergone significant changes. The real-time data stream anomaly detection model is obtained by training a neural network model using power grid operation data during normal operation, when the power grid status has undergone significant changes, and when an accident occurs.

4. The industrial centralized control system accident replay method according to claim 1, characterized in that, The preset power grid model can be the current operating power grid model, the power grid model at the time of the accident, or the research-state power grid model. The research-state power grid model is obtained by adjusting the current operating power grid model according to research needs.

5. The industrial centralized control system accident retrieval method according to any one of claims 1-3, characterized in that, The method also includes using an accident prediction model to analyze real-time power grid operation data and output accident early warning information and risk assessment reports; the accident prediction model is obtained by analyzing a dataset constructed using time series analysis algorithms and time series data from historical accident occurrences.

6. The industrial centralized control system accident retrieval method according to any one of claims 1-3, characterized in that, The method also includes processing the power grid operation data and dispatcher operation data at the time of the accident using an accident handling recommendation model after the accident occurs, and outputting an accident handling plan. The accident handling recommendation model is established based on the analysis of dispatchers' operational data during the accident handling process, historical accident handling cases, and recommendation algorithms.

7. The industrial centralized control system accident recall method according to any one of claims 1-3, characterized in that, The method also includes constructing a knowledge graph-based accident causal chain model based on accident time, accident type, and equipment name, and generating a visualization report of the accident evolution path by combining equipment topology relationships and historical accidents. The visualization report is presented in the form of charts, reports, or change curves.

8. The industrial centralized control system accident retrieval method according to claim 1, characterized in that, The cross-sectional data saved at the time of the accident includes plant data associated with the accident and equipment data with high correlation to the accident. The equipment data with high correlation to the accident refers to the data of the top N equipment obtained by sorting the equipment data from highest to lowest accident correlation using a machine learning algorithm.

9. An industrial centralized control accident tracing system, comprising a processor, characterized in that, The processor is used to implement the industrial centralized control accident recall method as described in any one of claims 1-8.

10. A storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are used, when executed, to implement the steps in the industrial centralized control accident recall method as described in any one of claims 1-8.

Images