Artificial intelligence-based monitoring eventization analysis model construction method and system

Abstract

This invention discloses a method and system for constructing an AI-based monitoring event analysis model. It acquires remote signaling alarm information, real-time telemetry data, remote control operation records, and tagging operation information from a dispatch data center; cleans the multi-source monitoring data based on a power grid topology model; clusters the remote signaling alarm information using the power grid topology model, including at least maintenance events and fault tripping events; performs time-series modeling of the multi-source monitoring data using a BiLSTM bidirectional long short-term memory network to capture the bidirectional dependencies of monitoring signals and extract key features; calculates equipment failure probabilities by combining power grid topology and real-time data to establish a monitoring event analysis model; inputs the initial monitoring data and clustering results into the monitoring event analysis model, and outputs event classification results and failure probabilities. This method can accurately and automatically classify real-time alarm information into fault tripping and maintenance / debugging events, improving the accuracy of the analysis.

Description

Technical Field

[0001] This invention relates to the field of power system monitoring technology, and in particular to a method and system for constructing monitoring event-based analysis models based on artificial intelligence. Background Technology

[0002] With the expansion of the power grid and the increasing demand for intelligent operation and maintenance, traditional manual monitoring methods face problems such as low efficiency in processing massive alarm signals, high false positive rates, and significant risks of missed detections. Current technologies primarily rely on human experience for monitoring signal processing, lacking the ability to perform correlation analysis on multi-source data, making it difficult to achieve automatic clustering and accurate identification of fault events. Furthermore, remote signaling alarm signals suffer from false transmissions, missed transmissions, and delayed uploading, further reducing the accuracy of event analysis. Therefore, improving the intelligence level and management efficiency of power grid operation monitoring is a pressing technical problem that needs to be solved at this stage. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing a method and system for constructing a monitoring event-based analysis model based on artificial intelligence.

[0004] The technical solution of this invention is a method for constructing a monitoring event-based analysis model based on artificial intelligence. This method includes the following steps: The system obtains remote signaling alarm information, real-time telemetry data, remote control operation records and tagging operation information from the dispatch data center to obtain multi-source monitoring data. The multi-source monitoring data is then cleaned based on the power grid topology model to obtain initial monitoring data. The remote signaling alarm information is clustered using the power grid topology model to obtain the clustering results, which include at least maintenance events and fault tripping events. A time-series model of multi-source monitoring data is performed using a BiLSTM bidirectional long short-term memory network. The BiLSTM network is used to capture the bidirectional dependencies of monitoring signals and extract key features. The probability of equipment failure is calculated by combining the power grid topology and real-time data, and a monitoring event analysis model is established. The initial monitoring data and the clustering results are input into the monitoring event analysis model, which outputs the event classification results and the probability of failure.

[0005] Furthermore, in the above-mentioned method for constructing an AI-based monitoring event-driven analysis model, the initial monitoring data, obtained by cleaning the multi-source monitoring data based on the power grid topology model and expert rule base, includes: Using PSS / E power system analysis software, the power grid structure is abstracted into a graph theory model, where nodes represent buses and equipment, edges represent the connection relationships between equipment, and the DFS depth-first search algorithm is used to identify connected subgraphs in the power grid and establish a power grid topology model. Based on the power grid topology model, unrelated feature information in monitoring and alarm information is marked as invalid information, and circuit breaker displacement and heavy load impact information are marked as associated information; By using telemetry data, tag information, and circuit breaker location information, we can determine whether the equipment status is consistent with the tag information and obtain multi-source monitoring data.

[0006] Furthermore, the above-mentioned method for constructing an AI-based monitoring event analysis model includes: The remote signaling alarm information is clustered according to the power grid topology model to obtain clustering results, which include at least the maintenance event analysis results and the fault tripping event analysis results. By combining equipment maintenance time, disconnector position information and tag status, the maintenance and commissioning status of switches, lines, busbars and main transformer equipment can be identified to obtain maintenance event analysis results. Based on protection actions, switch opening and closing signals, and power grid operation mode, the equipment fault type is determined, and the fault events of the same line, bus, and main transformer are clustered by decision tree to obtain the tripping event analysis results.

[0007] Furthermore, in the aforementioned method for constructing an AI-based monitoring event-based analysis model, the clustering of remote signaling alarm information using a power grid topology model to obtain clustering results includes at least maintenance events and fault tripping events, and also includes: A decision tree is established based on the tripping event time, power grid topology, and real-time operation mode; To determine whether switches that trip due to faults within a certain time range belong to the same line, are connected to the same busbar unit, and are powered by the same main transformer, cluster various equipment fault events.

[0008] Furthermore, in the aforementioned method for constructing an AI-based monitoring event analysis model, the step of using a BiLSTM bidirectional long short-term memory network to perform time-series modeling of multi-source monitoring data, capturing the bidirectional dependencies of monitoring signals using BiLSTM, and extracting key features; combining the power grid topology and real-time data to calculate equipment failure probabilities, and establishing a monitoring event analysis model, including: Design a BiLSTM bidirectional long short-term memory network structure, including an input layer, a BiLSTM layer, a fully connected layer, and an output layer; The input layer receives preprocessed time series data, the BiLSTM layer is used to capture bidirectional dependencies in the time series, the fully connected layer maps the output of the BiLSTM to the risk prediction space, and the output layer provides the event analysis results.

[0009] Furthermore, in the aforementioned method for constructing an AI-based monitoring event analysis model, the step of using a BiLSTM bidirectional long short-term memory network to perform time-series modeling of multi-source monitoring data, capturing the bidirectional dependencies of monitoring signals using BiLSTM, and extracting key features; combining the power grid topology and real-time data to calculate equipment failure probabilities, and establishing a monitoring event analysis model, including: A forget gate decision function is established using BiLSTM to save information in the monitoring alarm information sequence; In the mapping process between the input and output sequences of BiLSTM, contextual information is used to input the input data in two directions and compute the hidden representations in both directions to model long-distance dependencies in the text.

[0010] Furthermore, in the above-mentioned method for constructing an AI-based monitoring event analysis model, the step of inputting the initial monitoring data and the clustering results into the monitoring event analysis model and outputting event classification results and fault probabilities includes: The initial monitoring data and the clustering results are input into the monitoring event analysis model. Through forward and reverse time series modeling, the long-term dependency features of the signal are extracted, and the event classification results and failure probability are output. The classification results include at least maintenance events and tripping events, and the weight parameters of the monitoring event analysis model are updated using manually labeled corrected data.

[0011] Furthermore, in the AI-based monitoring event-based analysis model construction system, the monitoring event-based analysis model construction system includes the following modules: The multi-source data acquisition module is used to acquire remote signaling alarm information, real-time telemetry data, remote control operation records and tagging operation information from the dispatch data center to obtain multi-source monitoring data. Based on the power grid topology model, the multi-source monitoring data is cleaned to obtain initial monitoring data. The information clustering analysis module is used to cluster remote signaling alarm information using the power grid topology model to obtain clustering results, which include at least maintenance events and fault tripping events. The analysis model building module is used to perform time-series modeling of multi-source monitoring data through a BiLSTM bidirectional long short-term memory network, capture the bidirectional dependency relationship of monitoring signals through BiLSTM, extract key features, and calculate the equipment failure probability by combining the power grid topology and real-time data to build a monitoring event analysis model. The event analysis and evaluation module is used to input the initial monitoring data and the clustering results into the monitoring event analysis model, and output the event classification results and failure probability.

[0012] Furthermore, in the AI-based monitoring event-driven analysis model building system, the analysis model building module includes the following sub-modules: The storage submodule is used to establish a forget gate decision function through BiLSTM to store information in the monitoring alarm information sequence; The modeling submodule is used to utilize contextual information during the mapping process between the BiLSTM input and output sequences. It inputs the data in two directions and simultaneously computes the hidden representations in both directions to model long-distance dependencies in the text.

[0013] Furthermore, in the AI-based monitoring event analysis model construction system, the event analysis and evaluation module includes the following sub-modules: The extraction submodule is used to input the initial monitoring data and the clustering results into the monitoring event analysis model, extract the long-term dependency features of the signal through forward and reverse time series modeling, and output the event classification results and fault probability. The update submodule is used to classify results that include at least maintenance events and tripping events, and to update the weight parameters of the monitoring event analysis model using manually labeled correction data.

[0014] Its beneficial effects lie in the fact that, through research on key technologies for power grid monitoring event analysis, it presents a process for event-based analysis of power grid monitoring information, mines and analyzes monitoring alarm data, establishes an event-based analysis model for power grid monitoring information, and develops a method for extracting correlation features related to fault tripping and maintenance / debugging events. Furthermore, it uses the BiLSTM method combined with real-time alarm information to continuously optimize and update the monitoring event-based analysis model. Power grid monitoring events identified using this model can accurately and automatically classify real-time alarm information into fault tripping and maintenance / debugging events, significantly reducing the workload of manual analysis and improving the accuracy of the analysis. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0016] Figure 1 This is a schematic diagram of the first embodiment of the monitoring event analysis model construction method based on artificial intelligence in this invention. Figure 2 This is a schematic diagram of the first embodiment of the monitoring event analysis model construction system based on artificial intelligence in this invention. Figure 3 This is a schematic diagram of an embodiment of the monitoring event identification and analysis method based on artificial intelligence in this invention. Figure 4 This is a schematic diagram illustrating an embodiment of the device fault process analysis method based on artificial intelligence for constructing a monitoring event-based analysis model in this invention. Figure 5 This is a schematic diagram of an embodiment of the device failure event clustering method for constructing a monitoring event-based analysis model based on artificial intelligence in this invention. Figure 6 This is a schematic diagram of an embodiment of the power grid event identification method based on the artificial intelligence-based monitoring event analysis model construction method in this invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0019] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 As shown, the method for constructing a monitoring event-based analysis model based on artificial intelligence includes the following steps: Step 101: Obtain remote signaling alarm information, real-time telemetry data, remote control operation records, and tagging operation information from the dispatch data center to obtain multi-source monitoring data; clean the multi-source monitoring data based on the power grid topology model to obtain initial monitoring data.

[0020] Specifically, in this embodiment, monitoring data identification and cleaning based on multi-source data is carried out. Monitoring alarms and measurement information contain a certain amount of erroneous data, redundant data and test data, which have a great impact on the event-based analysis and research of monitoring data. Multi-source data such as power grid model and equipment status can be combined to identify and clean remote signaling and telemetry data.

[0021] (1) Identification of invalid data: Monitoring alarm information with the following associated characteristics is marked as invalid information and invalid information is not included in event analysis. Alarm suppression and blocking intervals and equipment information; frequently occurring actions and reset information. (2) Identification of accompanying information: Alarm information that does not affect the actual operation of the equipment and is caused by circuit breaker displacement, heavy load impact, etc., in the same interval and equipment non-accident (alarm classification) information, is marked as accompanying information. (3) Identification of circuit breaker or disconnector position mismatch: When the circuit breaker is in the open position and the circuit breaker current exceeds the limit by 2%, it is judged that the circuit breaker position is mismatched; when the line disconnector, main transformer disconnector, and each bus disconnector are in the open position and the circuit breaker current exceeds the limit by 2%, it is judged that the circuit breaker position is mismatched. (4) Identification of equipment status and signage inconsistency: When the real-time status of the equipment shows that the circuit breaker and one side disconnector are both in the closed position and a "power outage" or "maintenance" sign is set, it is judged that the equipment status and signage are inconsistent.

[0022] Specifically, in this embodiment, the power grid structure is abstracted into a graph theory model using PSS / E power system analysis software. Nodes represent buses and equipment, and edges represent the connections between equipment. A depth-first search (DFS) algorithm is used to identify connected subgraphs in the power grid and establish a power grid topology model. Based on the power grid topology model, unrelated feature information in monitoring alarm information is marked as invalid information, and circuit breaker displacement and heavy load impact information are marked as associated information. Through telemetry data, tag information, and circuit breaker location information, it is determined whether the equipment status is consistent with the tag information, thus obtaining multi-source monitoring data.

[0023] Step 102: Use the power grid topology model to cluster the remote signaling alarm information to obtain the clustering results, which should include at least maintenance events and fault tripping events. Specifically, in this embodiment, the remote signaling alarm information is clustered according to the power grid topology model to obtain clustering results, which include at least maintenance event analysis results and fault tripping event analysis results; combined with equipment maintenance time, disconnector change information and tag status, the maintenance and commissioning status of switches, lines, busbars and main transformer equipment is identified to obtain maintenance event analysis results; Based on protection actions, switch opening and closing signals, and power grid operation modes, the type of equipment fault is determined, and fault events of the same line, bus, and main transformer are clustered using a decision tree to obtain tripping event analysis results. A decision tree is established based on the tripping event occurrence time, power grid topology, and real-time operation mode. Based on the power grid topology and real-time operation mode, it is determined whether switches that tripped within a certain time range belong to the same line, are connected to the same bus unit, and are powered by the same main transformer, thus clustering various equipment fault events.

[0024] Maintenance event analysis based on equipment status analyzes the power grid operation mode, identifies relevant alarm information during the maintenance and commissioning of equipment such as switches, lines, busbars, and main transformers, and mainly analyzes the equipment operation status to determine whether various types of equipment are under maintenance.

[0025] (1) Switch maintenance event analysis: mainly based on power grid topology analysis, extracting switch and switch and grounding switch change information from remote signaling alarm information to determine whether the switch is under maintenance. (2) Line maintenance event analysis: mainly based on power grid topology analysis, extracting grounding switch (A and B station line side grounding switch) change information from remote signaling alarm information to determine whether the line is under maintenance. (3) Busbar maintenance event analysis: mainly based on power grid topology analysis, extracting busbar PT grounding switch change information from remote signaling alarm information to determine whether the busbar is under maintenance. (4) Main transformer maintenance event analysis: mainly based on power grid topology analysis, extracting grounding switch change information near the windings on the high, medium, and low voltage sides of the main transformer from remote signaling alarm information to determine whether the main transformer is under maintenance.

[0026] Tripping event analysis based on remote signaling alarms establishes a mapping between primary main equipment and secondary protection equipment according to power grid topology and point table information. Real-time alarm information is analyzed, and after removing equipment maintenance and commissioning related information, tripping-related alarms are identified based on the tripping event rule base, including switch opening and closing, protection actions, interval or station-wide fault information. The analysis focuses on tripping characteristic information to determine whether a switch has failed, what type of failure it has, and whether any information has been missed. Simultaneously, based on the power grid topology and operating mode, it analyzes whether multiple switch faults can be aggregated into line faults, bus faults, or main transformer faults.

[0027] Protection action analysis considers protection configurations for different voltage levels and equipment. ① 220kV and above switches, lines, busbars, main transformers, and other equipment generally have dual main protection configurations. ② Overhead lines are generally equipped with reclosing devices, while cables, main transformers, busbars, and other equipment generally do not. ③ If there is no reclosing information for a 10kV line, the reclosing status can be determined by the last state of the switch within the fault time range. If the switch state is closed and accompanied by the "spring not charged" message, the reclosing is considered successful; otherwise, the switch has tripped. ④ The power grid under the jurisdiction of the municipal company contains more automatic transfer switches, such as the coordination between bridge switches and line switches in the internal bridge connection, and the coordination between low-voltage side switches and sectionalizing switches in the main transformer.

[0028] Analysis of Missing Information: ① If "Protection Exit" information is missing, "Interval Fault Total" information will be used as a substitute. ② If "Switch Opening" or "Switch Closing" information is missing, real-time telemetry and remote control operation data will be used to assist in analyzing the switch status. If switch opening information is missing but closing information is available, opening information will be automatically supplemented. If switch opening information is missing but closing information is not available, telemetry values ​​will be used to assist in analyzing the switch opening / closing status. If reclosing is successful but "opening-closing" information is missing, the switch status will be determined to be "closed" through telemetry and real-time remote signaling status of the equipment, using information such as "Reclosing Output" + "Spring Not Charged" or "Control Circuit Disconnection". If "opening-closing" information is missing in a permanent fault (opening-closing-opening), it is equivalent to reclosing failure. ③ If "Reclosing Output" information is missing but "closing" information is available, and closing caused by fault testing or other reasons is ruled out from time or remote control operation, then the "Reclosing Output" information will be treated as missing information.

[0029] Fault phase analysis: ① When the faulty equipment is a line with a voltage level of 220kV or 500kV, fault phase analysis is required. ② Different weights are assigned to phases A, B, and C of the equipment. For phases AB, the weight is the sum of the weights for phases A and B; the same applies to phases BC and CA. ③ If only a single-phase switch trips, the phase classification is single-phase; if there is a phase-to-phase fault, the specific phase-to-phase fault is determined by adding the weights; if it is a three-phase fault, it is classified as a three-phase fault; otherwise, the switch tripping is displayed.

[0030] Fault process analysis primarily involves analyzing remote signaling alarm information after filtering out maintenance-related signals. Starting with protection output information, it extracts tripping information from relevant main protection equipment to determine switch tripping events. Based on confirmed tripping events, it extracts characteristic information such as protection reclosing, switch closing, interval fault signals, and total station fault data to analyze whether switch reclosing was successful and whether any information was missed.

[0031] Fault clustering analysis establishes a decision tree based on the tripping event time, power grid topology, and real-time operation mode. Based on the power grid topology and real-time operation mode, it clusters various equipment fault events within a certain time range by determining whether the switches that tripped due to faults belong to the same line, are connected to the same bus unit, or are powered by the same main transformer.

[0032] Specifically, based on the power grid topology-based monitoring event analysis model, this model integrates business data and aggregates basic information in an event-driven manner. It organically integrates monitoring alarms with specific logic and human experience to establish a model library, and comprehensively analyzes relevant data from various systems. Based on signal relationships, the power grid topology model, and equipment operating characteristics, it automatically establishes "atom-to-molecule-to-cell" relationships between the original discrete and large-scale monitoring alarm information. This enables the clustering and display of equipment monitoring information centered on power grid events, allowing operators to focus on effective alarms that truly reflect potential hazards in power grid equipment. According to experience in daily power grid monitoring and handling, power outage events can be divided into maintenance and commissioning events and fault tripping events. Alarm information occurring from the start to the end of equipment maintenance is classified as maintenance events; alarm information occurring from equipment fault tripping to power restoration is classified as fault tripping events. Equipment tripping includes switch tripping, line tripping, bus tripping, and main transformer tripping.

[0033] Step 103: Perform time-series modeling of multi-source monitoring data using a BiLSTM bidirectional long short-term memory network, capture the bidirectional dependency relationship of monitoring signals using BiLSTM, extract key features, and calculate equipment failure probability by combining power grid topology and real-time data to establish a monitoring event analysis model. Specifically, in this embodiment, a BiLSTM bidirectional long short-term memory network structure is designed, including an input layer, a BiLSTM layer, a fully connected layer, and an output layer; The input layer receives preprocessed time-series data, the BiLSTM layer captures bidirectional dependencies in the time series, the fully connected layer maps the output of the BiLSTM to the risk prediction space, and the output layer provides the event analysis results. A forget gate function is established using BiLSTM to preserve information in the monitoring and alarm information sequence. During the mapping process between the BiLSTM input and output sequences, contextual information is utilized to input the data in two directions, simultaneously calculating the hidden representations in both directions to model long-distance dependencies in the text.

[0034] A monitoring event analysis model based on bidirectional long short-term memory (LSTM), combined with power grid topology, equipment status, and remote signaling alarms, can accurately analyze most power grid tripping and maintenance events. However, remote signaling alarms suffer from false alarms, missed alarms, and delayed transmission, leading to decreased accuracy and untimely analysis, which hinders timely handling of power grid faults and anomalies. Therefore, manual intervention is required for incorrectly analyzed power grid events. A bidirectional LTM analysis method is introduced to dynamically learn from the results of manual intervention, continuously adjusting the parameters of the monitoring event analysis model. Through fault probability calculation, the model is continuously optimized to gradually improve the accuracy of power grid event analysis.

[0035] Bidirectional Long Short-Term Memory (BiLSTM) is an improved Long Short-Term Memory network that enhances the model's expressive power by simultaneously considering both forward and backward information in the sequence data. BiLSTM consists of two independent LSTMs: one processes the forward time series, and the other processes the backward time series. This structure allows BiLSTM to capture bidirectional dependencies in the time series, thus providing a more comprehensive understanding of the sequence data's characteristics. BiLSTM works by processing the sequence from beginning to end for a given time series, while the backward LSTM processes it from end to beginning. The outputs of the two LSTMs are concatenated or combined at each step to form the final output. This bidirectional processing mechanism enables BiLSTM to consider both past and future information, allowing for more accurate identification of risk patterns and trends when processing power grid risk data.

[0036] In power grid risk analysis, BiLSTM is primarily applied to three aspects: risk identification, assessment, and prediction. For risk identification, BiLSTM can learn from power grid operation data under normal conditions to identify abnormal states that significantly deviate from normal patterns, thus enabling early warning of risks. In risk assessment, BiLSTM can leverage its powerful sequence modeling capabilities to evaluate the severity and scope of impact of different risk factors.

[0037] In power monitoring signal analysis, BiLSTM is primarily applied to three aspects: signal feature extraction, pattern recognition, and anomaly detection. For signal feature extraction, BiLSTM learns from power monitoring signals under normal conditions to extract key signal features, such as grid operation mode, power flow changes, and equipment anomalies. In pattern recognition, BiLSTM leverages its powerful sequence modeling capabilities to identify different signal patterns, such as normal operation signals and fault signals. For anomaly detection, BiLSTM learns from historical monitoring data to capture abnormal patterns in signals; for example, by analyzing time-series data such as load changes and abnormal switch opening and closing, it can detect potential equipment faults.

[0038] Training a monitoring event analysis model, specifically building a BiLSTM-based monitoring event analysis model, mainly involves three steps: data preprocessing, model building, and training optimization. First, the power grid operation data needs to be preprocessed, including data cleaning, feature selection, and standardization. This step is crucial for improving model performance, as high-quality input data is the foundation for accurate predictions. In the model building phase, a suitable BiLSTM network structure needs to be designed. This typically includes an input layer, a BiLSTM layer, a fully connected layer, and an output layer. The input layer receives the preprocessed time-series data, the BiLSTM layer captures bidirectional dependencies in the time series, the fully connected layer maps the BiLSTM output to the risk prediction space, and the output layer provides the event analysis results. BiLSTM uses a forget gate function to preserve necessary information from the monitoring alarm information sequence, providing a foundation for subsequent event identification and calculation. Taking monitoring alarm information as an example, the object to be analyzed is a sequence of monitoring alarm information over a period of time; the context of the alarm information needs to be considered during the learning process. BiLSTM can utilize contextual information during the mapping process between input and output sequences, inputting input data in two directions and simultaneously computing hidden representations in both directions, effectively modeling long-distance dependencies in text.

[0039] Given an input sequence of alarm information, a bidirectional LSTM is used to process the alarm information. The alarm information sequence is then incorporated into a forget gate, an input gate, and an output gate. The information of each node is parsed and combined with the preceding and following context to obtain the information of each node. The results of the event analysis are labeled to obtain the output sequence. Each alarm message in the output can be labeled as an element in the set {B, M, E, S}, indicating that this alarm is information about the start, middle, and end of an event.

[0040] (1) Gate of Oblivion The formula for the forgetting gate is: in, This represents the current memory content, used here to store the last, unfinished alarms. xt represents the alarm input at time t. This describes the process of processing two parameters based on model information, power grid knowledge, and equipment status. During monitoring event analysis, key information in the monitoring event rule base needs to be stored for a relatively long time to limit the objects to be matched in subsequent alarm information. This key information is extracted by matching the combined results of ht-1 and xt with the information in the model rule base. Furthermore, the activation function σ is used to determine which information is retained and which is discarded in long-term memory.

[0041] (2) Input gate The input gate processes the input information. First, it passes through a tanh layer to obtain a representation of the current information (denoted as ). For example, when receiving an alarm message like "protect the exit," the Sigmoid layer (it) determines whether the information is key information in the monitoring event rule base. Additionally, the ft layer calculates the rule information of long-term memory stored in the cell to decide which information to retain.

[0042] The formula for calculating the input gate is: Where Ct-1 represents the information that needed to be retained long-term in the previous step, and Ct represents the information that needs to be retained long-term in the current step. If the current parsed information is protection exit information, it indicates that the type of subsequent information may be a tripping event.

[0043] (3) Output gate First, based on the rules of short-term memory and the current input, the identified information is output; second, the rules to be memorized in the short term are calculated.

[0044] The output gate formula is: in, .

[0045] By establishing an analytical model for information such as switch opening and closing, protection output, reclosing output, and total fault, and by passing the analytical results of relevant information as parameters into the "forgot gate" for processing, alarm information that needs to cross long distances can be saved.

[0046] Step 104: Input the initial monitoring data and clustering results into the monitoring event analysis model, and output the event classification results and failure probability.

[0047] Specifically, in this embodiment, the initial monitoring data and clustering results are input into the monitoring event analysis model. Through forward and reverse time series modeling, the long-term dependency features of the signal are extracted, and the event classification results and fault probability are output. The classification results include at least maintenance events and tripping events, and the weight parameters of the monitoring event analysis model are updated using manually labeled corrected data.

[0048] Its beneficial effects lie in the fact that, through research on key technologies for power grid monitoring event analysis, it presents a process for event-based analysis of power grid monitoring information, mines and analyzes monitoring alarm data, establishes an event-based analysis model for power grid monitoring information, and develops a method for extracting correlation features related to fault tripping and maintenance / debugging events. Furthermore, it uses the BiLSTM method combined with real-time alarm information to continuously optimize and update the monitoring event-based analysis model. Power grid monitoring events identified using this model can accurately and automatically classify real-time alarm information into fault tripping and maintenance / debugging events, significantly reducing the workload of manual analysis and improving the accuracy of the analysis.

[0049] The above describes embodiments of the artificial intelligence-based monitoring event analysis model construction method of the present invention. Please refer to [link / reference]. Figure 2 In the AI-based monitoring event-driven analysis model building system, the system includes the following modules: The multi-source data acquisition module is used to acquire remote signaling alarm information, real-time telemetry data, remote control operation records and tagging operation information from the dispatch data center to obtain multi-source monitoring data. The monitoring data cleaning module is used to clean multi-source monitoring data based on the power grid topology model to obtain initial monitoring data; The information clustering analysis module is used to cluster remote signaling alarm information using the power grid topology model to obtain clustering results, which include at least maintenance events and fault tripping events. The analysis model building module is used to perform time-series modeling of multi-source monitoring data through a BiLSTM bidirectional long short-term memory network, capture the bidirectional dependency relationship of monitoring signals through BiLSTM, extract key features, and calculate the equipment failure probability by combining the power grid topology and real-time data to build a monitoring event analysis model. The event analysis and evaluation module is used to input initial monitoring data and clustering results into the monitoring event analysis model, and output event classification results and failure probability.

[0050] Please see Figure 3 A schematic diagram of an embodiment of equipment tripping event identification and analysis in the construction method of monitoring event-based analysis model based on artificial intelligence.

[0051] Please see Figure 4 A schematic diagram illustrating an embodiment of equipment failure process analysis in an AI-based monitoring event-driven analysis model construction method.

[0052] Please see Figure 5 A schematic diagram of an embodiment of equipment failure event clustering in the construction method of monitoring event-based analysis model based on artificial intelligence.

[0053] Please see Figure 6A schematic diagram of an embodiment of the power grid event identification method in the construction method of monitoring event-based analysis model based on artificial intelligence.

[0054] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for constructing an AI-based monitoring event analysis model, characterized by, The method for constructing the monitoring event-based analysis model includes the following steps: The system obtains remote signaling alarm information, real-time telemetry data, remote control operation records and tagging operation information from the dispatch data center to obtain multi-source monitoring data. The multi-source monitoring data is then cleaned based on the power grid topology model to obtain initial monitoring data. The remote signaling alarm information is clustered using the power grid topology model to obtain the clustering results, which include at least maintenance events and fault tripping events. A time-series model of multi-source monitoring data is performed using a BiLSTM bidirectional long short-term memory network. The BiLSTM network is used to capture the bidirectional dependencies of monitoring signals and extract key features. The probability of equipment failure is calculated by combining the power grid topology and real-time data, and a monitoring event analysis model is established. The initial monitoring data and the clustering results are input into the monitoring event analysis model, which outputs the event classification results and the probability of failure. 2.The artificial intelligence-based monitoring eventization analysis model construction method of claim 1, wherein The process involves acquiring remote signaling alarm information, real-time telemetry data, remote control operation records, and tagging operation information from the dispatch data center to obtain multi-source monitoring data. This multi-source monitoring data is then cleaned based on a power grid topology model to obtain initial monitoring data, including: Using PSS / E power system analysis software, the power grid structure is abstracted into a graph theory model, where nodes represent buses and equipment, edges represent the connection relationships between equipment, and the DFS depth-first search algorithm is used to identify connected subgraphs in the power grid and establish a power grid topology model. Based on the power grid topology model, unrelated feature information in monitoring and alarm information is marked as invalid information, and circuit breaker displacement and heavy load impact information are marked as associated information; By using telemetry data, tag information, and circuit breaker location information, we can determine whether the equipment status is consistent with the tag information and obtain multi-source monitoring data. 3.The artificial intelligence-based monitoring eventization analysis model construction method of claim 1, wherein The method of clustering remote signaling alarm information using a power grid topology model to obtain clustering results includes at least maintenance events and fault tripping events, including: The remote signaling alarm information is clustered according to the power grid topology model to obtain clustering results, which include at least the maintenance event analysis results and the fault tripping event analysis results. By combining equipment maintenance time, disconnector position information and tag status, the maintenance and commissioning status of switches, lines, busbars and main transformer equipment can be identified to obtain maintenance event analysis results. Based on protection actions, switch opening and closing signals, and power grid operation mode, the equipment fault type is determined, and the fault events of the same line, bus, and main transformer are clustered by decision tree to obtain the tripping event analysis results. 4.The artificial intelligence-based monitoring eventization analysis model construction method of claim 1, wherein The clustering of remote signaling alarm information using a power grid topology model to obtain clustering results includes at least maintenance events and fault tripping events, and also includes: A decision tree is established based on the tripping event time, power grid topology, and real-time operation mode; To determine whether switches that trip due to faults within a certain time range belong to the same line, are connected to the same busbar unit, and are powered by the same main transformer, cluster various equipment fault events. 5.The artificial intelligence-based monitoring eventization analysis model construction method of claim 1, wherein The process involves using a BiLSTM bidirectional long short-term memory network to perform time-series modeling of multi-source monitoring data, capturing the bidirectional dependencies of monitoring signals through BiLSTM, and extracting key features. By combining power grid topology and real-time data to calculate equipment failure probabilities, a monitoring event analysis model is established, including: Design a BiLSTM bidirectional long short-term memory network structure, including an input layer, a BiLSTM layer, a fully connected layer, and an output layer; The input layer receives preprocessed time series data, the BiLSTM layer is used to capture bidirectional dependencies in the time series, the fully connected layer maps the output of the BiLSTM to the risk prediction space, and the output layer provides the event analysis results. 6.The artificial intelligence-based monitoring eventization analysis model construction method of claim 1, wherein The process involves using a BiLSTM bidirectional long short-term memory network to perform time-series modeling of multi-source monitoring data, capturing the bidirectional dependencies of monitoring signals through BiLSTM, and extracting key features. By combining power grid topology and real-time data to calculate equipment failure probabilities, a monitoring event analysis model is established, including: A forget gate decision function is established using BiLSTM to save information in the monitoring alarm information sequence; In the mapping process between the input and output sequences of BiLSTM, contextual information is used to input the input data in two directions and compute the hidden representations in both directions to model long-distance dependencies in the text. 7.The artificial intelligence-based monitoring eventization analysis model construction method of claim 1, wherein The step of inputting the initial monitoring data and the clustering results into the monitoring event analysis model and outputting event classification results and failure probabilities includes: The initial monitoring data and the clustering results are input into the monitoring event analysis model. Through forward and reverse time series modeling, the long-term dependency features of the signal are extracted, and the event classification results and failure probability are output. The classification results include at least maintenance events and tripping events, and the weight parameters of the monitoring event analysis model are updated using manually labeled corrected data.

8. An artificial intelligence-based monitoring eventization analysis model construction system, characterized by, The monitoring event-based analysis model construction system includes the following modules: The multi-source data acquisition module is used to acquire remote signaling alarm information, real-time telemetry data, remote control operation records and tagging operation information from the dispatch data center to obtain multi-source monitoring data. Based on the power grid topology model, the multi-source monitoring data is cleaned to obtain initial monitoring data. The information clustering analysis module is used to cluster remote signaling alarm information using the power grid topology model to obtain clustering results, which include at least maintenance events and fault tripping events. The analysis model building module is used to perform time series modeling of multi-source monitoring data through BiLSTM bidirectional long short-term memory network, capture the bidirectional dependency relationship of monitoring signals through BiLSTM, and extract key features. By combining power grid topology and real-time data to calculate equipment failure probability, a monitoring event analysis model is established. The event analysis and evaluation module is used to input the initial monitoring data and the clustering results into the monitoring event analysis model, and output the event classification results and failure probability. 9.The artificial intelligence-based monitoring eventization analysis model building system of claim 8, wherein The analysis model building module includes the following sub-modules: The storage submodule is used to establish a forget gate decision function through BiLSTM to store information in the monitoring alarm information sequence; The modeling submodule is used to utilize contextual information during the mapping process between the BiLSTM input and output sequences. It inputs the data in two directions and simultaneously computes the hidden representations in both directions to model long-distance dependencies in the text. 10.The artificial intelligence-based monitoring eventization analysis model construction system of claim 8, wherein The event analysis and evaluation module includes the following sub-modules: The extraction submodule is used to input the initial monitoring data and the clustering results into the monitoring event analysis model, extract the long-term dependency features of the signal through forward and reverse time series modeling, and output the event classification results and fault probability. The update submodule is used to classify results that include at least maintenance events and tripping events, and to update the weight parameters of the monitoring event analysis model using manually labeled correction data.

Images