A power grid operation situation analysis system combined with a knowledge graph
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- 内蒙古电力(集团)有限责任公司电力调度控制分公司
- Filing Date
- 2025-06-24
- Publication Date
- 2026-06-02
Smart Images
Figure CN120781047B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of smart grid technology, and more specifically to a power grid operation status analysis system that incorporates knowledge graphs. Background Technology
[0002] The power grid operation status is a comprehensive reflection of the real-time operating status of the power system, encompassing the dynamic interaction of power supply and demand balance, equipment status, the stability of new energy access, and external influencing factors. Its core lies in maintaining a real-time balance between power supply and demand, ensuring that power generation capacity meets load demand, and avoiding power shortages and surpluses. The operating status of power grid equipment, including the load, failure rate, and aging degree of transmission lines and transformers, directly affects the stability and reliability of the system. The rapid development of new energy sources significantly impacts the power grid operation status. The volatility of wind and solar power increases the pressure on frequency and voltage regulation, but intelligent control and dispatch systems can optimize their access and improve stability. Meanwhile, external factors may also cause disturbances to the power grid, requiring corresponding response capabilities. In terms of technical support, knowledge graph technology integrates multi-dimensional data to construct hierarchical models, helping to intuitively understand the power grid status. Intelligent monitoring systems collect and analyze data in real time, predict risks and issue early warnings, and combined with electronic dashboards, achieve management visualization and transparency, improving decision-making efficiency. The stability of the power grid operation status requires comprehensive consideration of supply and demand, equipment, new energy sources, and external factors, and optimization of management through technical support to ensure safe and efficient operation and support economic and social development.
[0003] To address the issues of low processing efficiency and reliance on manual experience leading to inefficiencies and misjudgments in power grid operation status analysis, existing technologies employ manual screening and static rule engines. However, these methods still suffer from limitations such as the inability to effectively correlate unstructured data and the inability of static rules to dynamically adapt to changes in power grid topology. This results in issues like missed critical alarms, misjudgments of multiple fault correlations, and failure of cross-site fault propagation path analysis. Therefore, a power grid operation status analysis system incorporating knowledge graphs is proposed to solve these problems. Summary of the Invention
[0004] The purpose of this invention is to provide a power grid operation status analysis system that incorporates knowledge graphs to solve the problems mentioned in the background section.
[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a power grid operation status analysis system combined with knowledge graph, including a power grid data acquisition module, a power grid feature extraction module, a monitoring information event analysis module, a power grid graph construction module, and a power grid event intelligent judgment module;
[0006] The power grid data acquisition module acquires and preprocesses power grid data, which includes remote signaling data, telemetry data, OMS system data, and D5000 system data.
[0007] The power grid feature extraction module extracts remote signaling features, maintenance event features, and telemetry features from the preprocessed power grid data.
[0008] The monitoring information event analysis module combines remote signaling features, maintenance event features, and telemetry features to construct a maintenance event analysis model, a power grid fault analysis model, and a power grid fault clustering model, and outputs equipment maintenance and debugging information tags, fault time tags, a list of associated equipment, and a list of associated faults.
[0009] The power grid graph construction module constructs a power grid knowledge graph based on equipment maintenance and debugging information tags, fault time tags, a list of associated equipment, and a list of associated faults. It outputs fault propagation paths and voltage over-limit probability thresholds, and provides tiered early warnings.
[0010] The intelligent power grid event analysis module analyzes power grid data in real time, generates handling plans, and pushes abnormal data and power grid faults.
[0011] A further improvement to the technical solution of the present invention is that the power grid data acquisition and preprocessing process in the power grid data acquisition module includes:
[0012] The remote signaling data of power grid equipment with voltage levels from 10kV to 500kV is received through the real-time service interface of the scheduling data center. The remote signaling data includes alarm information such as accidents, anomalies, changes and over-limits. The specific fields are equipment primary key, occurrence time, voltage level, alarm flag bit and switch open / closed status.
[0013] By scheduling the minute-by-minute measurement service interface of the data center, telemetry data of power grid equipment is periodically obtained. The telemetry data includes measurement parameters such as active power, reactive power, voltage amplitude, current value, and oil temperature.
[0014] OMS system data includes maintenance request data and trip record data, while D5000 system data includes monitoring operation data and tag data;
[0015] By connecting the data center to the OMS system service interface, maintenance application data and tripping record data are obtained in real time. The maintenance application data includes the power outage equipment, maintenance period and approval time, and the tripping record data includes the fault type, tripping equipment association information and tripping switch.
[0016] By connecting the data center to the D5000 system service interface, monitoring operation data and tag data can be obtained in real time. The monitoring operation data includes operation time, operated device and operation result, and the tag data includes tag type, device ID and operation time.
[0017] By verifying the logical rules of unreasonable parallel connection of main transformers and mismatch of switch positions, redundant and erroneous signals in remote signaling data are eliminated, and standardized equipment status data is generated.
[0018] Outliers in telemetry data are identified based on the 3σ principle, and missing telemetry data are filled in by linear interpolation of adjacent time points to generate continuous telemetry time series data.
[0019] By employing bidirectional long short-term memory network word segmentation and conditional random field entity recognition technology, unstructured maintenance application data and trip record data are converted into structured data associated with equipment, and the trip record parsing results are obtained.
[0020] The validity of operation data and tag data is verified by matching operation results with real-time status. The correlation between tag type and equipment maintenance status is analyzed by combining support vector machine model to generate standardized operation logs.
[0021] A further improvement to the technical solution of this invention lies in the fact that the process of extracting remote signaling features, maintenance event features, and telemetry features in the power grid feature extraction module includes:
[0022] Remote signaling features include alarm type codes and switch open / closed status vectors. Alarm information is mapped to numerical labels. Through conditional judgment logic, the corresponding encoding vector is generated based on the alarm flag bit to obtain the alarm type code. The binary status value of the switch open / closed status field in the remote signaling data is extracted to generate the real-time switch open / closed status vector of the device.
[0023] The maintenance event characteristics include maintenance window period flags and equipment maintenance flags. Based on maintenance application data and remote signaling data from the OMS system, it is determined whether the current time is within the approved maintenance application period. Combined with the switch opening status, a Boolean maintenance window period flag is generated, where 0 indicates a non-maintenance period and 1 indicates a maintenance period. Based on the tag data and maintenance window period flags from the D5000 system, a support vector machine model is used to analyze the correlation between tag type and equipment status, and output the equipment maintenance flag. The tag types include maintenance tags and test tags.
[0024] Telemetry features include active power fluctuation rate and voltage over-limit frequency. Based on preprocessed active power time series data, a time window is set, the average active power within the window is extracted, and the average absolute deviation of active power from its mean at each time point within the window is calculated to obtain the active power fluctuation rate. Based on preprocessed voltage amplitude time series data, a rated voltage is set, and the over-limit range is defined with the rated voltage as the benchmark. The number of times the voltage exceeds 1.1 times the rated voltage and falls below 0.9 times the rated voltage per unit time is counted to obtain the voltage over-limit frequency.
[0025] A further improvement to the technical solution of this invention lies in the following: the process of constructing a maintenance event analysis model and outputting equipment maintenance and debugging information tags in the monitoring information event analysis module includes:
[0026] Using the switch open / closed state vector and maintenance window period flag as input features, a decision tree model is adopted to construct a maintenance event analysis model based on whether the current time falls within the approved maintenance application period and whether the switch open / closed state is open.
[0027] The root node of the decision tree prioritizes splitting the maintenance window period flag to filter out equipment that is in the maintenance period. The child nodes split the switch opening and closing status. If both the maintenance application period and the switch opening condition are met, the power grid equipment is determined to be in the maintenance and commissioning state. 0 represents non-maintenance and commissioning state and 1 represents maintenance and commissioning state. Finally, the leaf nodes output the equipment maintenance and commissioning information tag.
[0028] A further improvement to the technical solution of this invention lies in the following: the process of constructing a power grid fault analysis model and outputting fault time tags and a list of associated devices in the monitoring information event analysis module includes:
[0029] Using alarm type code, active power fluctuation rate and trip record parsing results as input features, a fault mode rule base is established based on historical fault cases. The input features are compared with the conditions in the fault mode rule base one by one. If the alarm type code, active power fluctuation rate and trip device association information meet the rule conditions, the corresponding fault label is triggered.
[0030] The alarm type code, active power fluctuation rate, and tripped device primary key are converted into input feature vectors. The support vector classification algorithm is used to find the optimal classification hyperplane based on the principle of maximizing the margin. The fault category label is divided according to the projection position of the input feature vector on the hyperplane.
[0031] By combining the topology of the power grid model, the path of fault impact is determined, and the primary key of the equipment and the topology-related equipment in the trip record are extracted to form a list of related equipment;
[0032] Based on the alarm occurrence time and the fault mode rule matching results, mark the fault start timestamp, output the fault timestamp and a list of associated devices including the fault source device and the topology-related devices within a single fault event.
[0033] A further improvement to the technical solution of this invention lies in the following: the process of constructing a power grid fault clustering model and outputting a list of associated faults in the monitoring information event analysis module includes:
[0034] Using the fault start timestamp, the primary key of the tripped device, and the fault category label as input features, the time range of fault association is defined, and two fault events with a timestamp interval not exceeding the length of the time window are regarded as associated events within the same time window.
[0035] Check if the primary keys of the tripped devices are the same. If they are the same, they are classified into the same fault cluster. If they are different, they are not classified into the same fault cluster. According to the power grid topology model, if the faulty devices belong to the same bay and the same substation, they are classified into the same cluster. Through the topological connection relationship, cross-station equipment faults are included in the same fault cluster.
[0036] A decision tree clustering algorithm is adopted. The root node of the decision tree uses the time window as the first splitting condition to filter out fault events within the same time window. The child nodes split according to the primary key of the tripped device to distinguish faults from different source devices. The algorithm is refined layer by layer according to the topology level to generate fault subclusters and construct a power grid fault clustering model. Each leaf node represents a fault cluster and outputs an associated fault list and fault cluster labels. The associated fault list is a list of multiple fault events that are associated within the same time window, the same device, and the same topology.
[0037] A further improvement to the technical solution of this invention lies in the following: the process of constructing a power grid knowledge graph and outputting fault propagation paths and voltage over-limit probability thresholds in the power grid graph construction module includes:
[0038] A power grid knowledge graph is constructed using equipment maintenance and commissioning information tags, fault time tags, a list of associated equipment, a list of associated faults, and the frequency of voltage over-limit as inputs.
[0039] The power grid knowledge graph uses the tripped device primary key as the core to construct device node attributes. The device node attributes include device maintenance and commissioning information tags and voltage over-limit frequency. Based on the event sequence in the associated fault list, the fault propagation relationship between devices is established, and the initial expert rule weights are set according to historical fault data.
[0040] If the equipment is under maintenance, its associated path weight is multiplied by the attenuation coefficient, and a voltage over-limit frequency threshold is set. If the voltage over-limit frequency exceeds its corresponding threshold, the weight of the associated path increases, and the weight of the propagation path generated by the cross-site equipment due to the grid topology connection is increased by 20%.
[0041] Multiple fault events in the associated fault list are sorted by timestamp to generate fault propagation paths. The path trigger probability is obtained based on the sum of the weights of each node in the path. An expert experience coefficient is introduced, and the voltage limit exceedance probability threshold is dynamically adjusted according to the frequency of voltage exceedance and the path trigger probability. If the real-time voltage exceeds the voltage limit exceedance probability threshold, a graded warning is issued.
[0042] A further improvement to the technical solution of this invention lies in that: the process of hierarchical early warning in the power grid map construction module includes:
[0043] The power grid mapping module receives fault propagation paths, voltage over-limit probability thresholds, and real-time voltage data. Based on the path triggering probability, it classifies alarms into emergency alarms, important alarms, and general alarms. It matches different alarm levels with corresponding path triggering probability threshold ranges. Based on the frequency of voltage over-limits and path weights, it dynamically adjusts the voltage over-limit probability threshold. If the real-time voltage exceeds the voltage over-limit probability threshold, the corresponding alarm level is triggered.
[0044] Cross-site equipment failures affect substations through the power grid topology, increasing the risk of cascading failures. An emergency alarm is determined if the following conditions are met simultaneously: the path trigger probability falls within the emergency alarm threshold range, the real-time voltage exceeds the voltage over-limit probability threshold, and the fault propagation path includes cross-site equipment. An important alarm is determined if one of the following conditions is met: the path trigger probability is within the emergency alarm and important alarm threshold range and exceeds the voltage over-limit probability threshold, or a single device is overloaded and the real-time voltage exceeds the voltage over-limit probability threshold. A general alarm is determined if the path trigger probability falls within the general alarm threshold range and the real-time voltage does not exceed the voltage over-limit probability threshold.
[0045] A further improvement to the technical solution of this invention lies in the following: the process of analyzing power grid data in real time and generating a response plan in the intelligent power grid event analysis module includes:
[0046] Based on the fault propagation path and power grid topology model, and utilizing the equipment relationships and expert rule weights in the power grid knowledge graph, the loss load is automatically calculated. The load shedding scheme is generated, and its calculation process is as follows:
[0047] ;
[0048] in, This represents the active power at the i-th load point. This represents the priority weight of the i-th load point. This represents the total number of load points.
[0049] By combining a wiring diagram simulation exercise, the risk points of the load shedding scheme are analyzed, the strategy is dynamically optimized, and by integrating expert rules and machine learning, auxiliary decision-making that balances safety and efficiency is output to support dispatchers in responding quickly.
[0050] A further improvement to the technical solution of this invention lies in the fact that the process of pushing abnormal data and power grid faults in the intelligent power grid event analysis module includes:
[0051] Based on the alarm level and voltage over-limit probability threshold, hierarchical alarms are dynamically triggered. Abnormal data is classified according to voltage level and over-limit frequency. High-priority anomalies are prioritized and pushed to user terminals within the corresponding dispatch jurisdiction and fault impact range. Fault information is assigned to a region according to dispatch permissions and pushed to mobile terminals and duty terminals. The fault information includes fault location, associated fault list and load shedding scheme. Low-latency transmission is achieved through message queues to ensure real-time alarm delivery.
[0052] Due to the adoption of the above technical solution, the technical progress achieved by this invention compared to the prior art is as follows:
[0053] This invention provides a power grid operation status analysis system that combines knowledge graphs. It uses natural language processing technology to parse unstructured maintenance applications and trip records, and dynamically constructs a knowledge graph by integrating multi-source data. This significantly improves the processing efficiency of massive monitoring signals, solves the problems of missed detections and misjudgments in manual screening, and realizes automated cleaning and correlation analysis of tens of thousands of alarm information per day, reducing the repetitive work of dispatchers.
[0054] This invention provides a power grid operation status analysis system that combines knowledge graphs. Based on a dual diagnostic mode of scheduling expert rule base and support vector machine algorithm, it dynamically adjusts the weight of fault propagation path, accurately identifies the correlation of faults in cross-station equipment, improves the accuracy of fault judgment, and effectively avoids the risk of misjudging new fault modes by traditional static rule base.
[0055] This invention provides a power grid operation status analysis system that combines knowledge graphs. By statistically analyzing the frequency of voltage over-limits and calculating the probability of path triggering in real time, it dynamically generates three-level alarm thresholds. Combined with a mobile terminal permission-based push mechanism, it enables emergency faults to reach the dispatch terminal within 5 seconds, shortening the emergency response time and ensuring the safety and stability of power grid operation. Attached Figure Description
[0056] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.
[0057] Figure 1 This is a block diagram of the present invention. Detailed Implementation
[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0059] Examples, such as Figure 1 As shown, the present invention provides a power grid operation status analysis system that combines knowledge graphs, including a power grid data acquisition module, a power grid feature extraction module, a monitoring information event analysis module, a power grid graph construction module, and a power grid event intelligent judgment module;
[0060] The power grid data acquisition module collects and preprocesses power grid data, including remote signaling data, telemetry data, OMS system data, and D5000 system data. Through the real-time service interface of the dispatch data center, it receives remote signaling data from power grid equipment across all voltage levels from 10kV to 500kV. This remote signaling data includes alarm information such as accidents, anomalies, changes, and over-limits, with specific fields including equipment primary key, occurrence time, voltage level, alarm flag, and switch open / closed status. Through the minute-by-minute measurement service interface of the dispatch data center, it periodically acquires telemetry data from power grid equipment. This telemetry data includes measurement parameters such as active power, reactive power, voltage amplitude, current value, and oil temperature. OMS system data includes maintenance application data and trip record data. D5000 system data includes monitoring operation data and tagging data. Through the dispatch data center's interface with the OMS system service, it acquires maintenance application data and trip record data in real time. The maintenance application data includes the equipment to be de-energized, the maintenance period, and the approval time. The trip record data package... The system includes fault type, tripped equipment association information, and tripped switches. Through the D5000 system service interface connected to the dispatch data center, it obtains real-time monitoring operation data and tag data. The monitoring operation data includes operation time, operated equipment, and operation result. The tag data includes tag type, equipment ID, and operation time. Redundant and erroneous signals in the remote signaling data are eliminated through logical rules verifying unreasonable parallel operation of main transformers and mismatched switch positions, generating standardized equipment status data. Outliers in the telemetry data are identified based on the 3σ principle. Missing telemetry data is filled by linear interpolation of adjacent time points, generating continuous telemetry time-series data. Unstructured maintenance application data and tripped record data are converted into structured data associated with equipment using bidirectional long short-term memory network word segmentation and conditional random field entity recognition technology. Trip record parsing results are obtained. The validity of the operation data and tag data is verified by matching the operation result with the real-time status. The correlation between tag type and equipment maintenance status is analyzed using a support vector machine model, generating standardized operation logs.
[0061] The power grid feature extraction module extracts remote signaling features, maintenance event features, and telemetry features from the preprocessed power grid data. Remote signaling features include alarm type codes and switch open / closed status vectors. Alarm information is mapped to numerical labels. Through conditional judgment logic, a corresponding encoding vector is generated based on the alarm flag bits to obtain the alarm type code. The binary status value of the switch open / closed status field in the remote signaling data is extracted to generate a real-time switch open / closed status vector for the equipment. Maintenance event features include maintenance window period flags and equipment maintenance flags. Based on maintenance application data and remote signaling data from the OMS system, it determines whether the current time is within the approved maintenance application period. Combined with the switch open / closed status, a Boolean maintenance window period flag is generated, where 0 indicates a non-maintenance period and 1 indicates a maintenance period. Based on the tag data and maintenance window period markers of the D5000 system, a support vector machine model is used to analyze the correlation between tag type and equipment status, and output the equipment maintenance marker. The tag types include maintenance tags and test tags. Telemetry features include active power fluctuation rate and voltage over-limit frequency. Based on the preprocessed active power time series data, a time window is set, the average active power within the window is extracted, and the average absolute deviation of active power from its mean at each time point within the window is statistically analyzed to obtain the active power fluctuation rate. Based on the preprocessed voltage amplitude time series data, a rated voltage is set, and the over-limit range is defined with the rated voltage as the benchmark. The number of times the voltage exceeds 1.1 times the rated voltage and is lower than 0.9 times the rated voltage per unit time is statistically analyzed to obtain the voltage over-limit frequency.
[0062] The monitoring information event analysis module, combining remote signaling features, maintenance event features, and telemetry features, constructs a maintenance event analysis model, a power grid fault analysis model, and a power grid fault clustering model. It outputs equipment maintenance and debugging information tags, fault time tags, a list of associated equipment, and a list of associated faults. Using switch open / closed state vectors and maintenance window period flags as input features, a decision tree model is employed. Based on determining whether the current time falls within the approved maintenance application period and checking whether the switch open / closed state is open, a maintenance event analysis model is constructed. The root node of the decision tree prioritizes splitting the maintenance window period flag, filtering out equipment within the maintenance period. Child nodes split the switch open / closed state; if both conditions are met—being within the approved maintenance application period and the switch being open—then the appropriate switch is selected. If the conditions are met, the power grid equipment is determined to be in a maintenance and commissioning state, with 0 representing a non-maintenance and commissioning state and 1 representing a maintenance and commissioning state. The final leaf node outputs a maintenance and commissioning information label for the equipment. Using alarm type codes, active power fluctuation rates, and tripping record parsing results as input features, a fault mode rule base is established based on historical fault cases. The input features are compared one by one with the conditions in the fault mode rule base. If the alarm type code, active power fluctuation rate, and tripping equipment association information meet the rule conditions, the corresponding fault label is triggered. The alarm type code, active power fluctuation rate, and tripping equipment primary key are converted into input feature vectors. A support vector classification algorithm is used, based on the margin maximization principle, to find the optimal classification hyperplane. The classification is then performed according to the input feature vectors. The projection position of the quantity on the hyperplane is used to classify fault category labels. Combined with the topology relationship of the power grid model, the fault impact path is determined. The device primary key and topology-related devices in the trip record are extracted to form a list of related devices. Based on the alarm occurrence time and fault mode rule matching results, the fault start timestamp is marked. The fault timestamp and the list of related devices containing the fault source device and topology-related devices within a single fault event are output. Using the fault start timestamp, trip device primary key, and fault category label as input features, the time range of fault association is defined. Two fault events with a timestamp interval not exceeding the length of this time window are regarded as related events within the same time window. The primary keys of the trip devices are checked to see if they are the same. If they are the same, they are classified as faults of the same device. Clusters that are different are not classified into the same equipment fault cluster. According to the power grid topology model, if the faulty equipment belongs to the same bay and the same substation, it is classified into the same cluster. Through topological connection relationships, cross-station equipment faults are included in the same fault cluster. The decision tree clustering algorithm is adopted. The root node of the decision tree uses the time window as the first splitting condition to filter out fault events within the same time window. The child nodes split according to the primary key of the tripped equipment to distinguish faults from different source equipment. The fault sub-clusters are generated by refining the topology layer by layer to construct the power grid fault clustering model. Each leaf node represents a fault cluster and outputs an associated fault list and fault cluster labels. The associated fault list is a list of multiple fault events that include the same time window, the same equipment, and the same topology association.
[0063] The power grid knowledge graph construction module constructs a power grid knowledge graph based on equipment maintenance and commissioning information tags, fault time tags, a list of associated equipment, and a list of associated faults. It outputs fault propagation paths and voltage limit exceedance probability thresholds, and provides tiered early warnings. The module uses equipment maintenance and commissioning information tags, fault time tags, a list of associated equipment, a list of associated faults, and voltage limit exceedance frequency as inputs to construct the power grid knowledge graph. This knowledge graph uses the tripped equipment primary key as its core and constructs equipment node attributes, including equipment maintenance and commissioning information tags and voltage limit exceedance frequency. Based on the event sequence in the associated fault list, it establishes fault propagation relationships between equipment. It sets initial expert rule weights based on historical fault data. If equipment is under maintenance, its associated path weight is multiplied by an attenuation coefficient. A voltage limit exceedance frequency threshold is set. If the voltage limit exceedance frequency exceeds the corresponding threshold, the weight of the relevant path increases. Furthermore, the weight of propagation paths generated by cross-site equipment due to power grid topology connections is increased by 20%. Multiple fault events in the associated fault list are sorted by timestamp to generate fault propagation paths. The path trigger probability is obtained based on the sum of the weights of each node in the path. An expert experience coefficient is introduced, and the probability is determined based on the voltage limit exceedance frequency and... The path trigger probability is dynamically adjusted to determine the voltage over-limit probability threshold. If the real-time voltage exceeds the voltage over-limit probability threshold, a tiered warning is issued. The power grid topology construction module receives the fault propagation path, voltage over-limit probability threshold, and real-time voltage data. Based on the path trigger probability, it classifies alarms into emergency, important, and general levels, and matches corresponding path trigger probability threshold ranges to different alarm levels. The voltage over-limit probability threshold is dynamically adjusted based on the frequency of voltage over-limits and path weights. If the real-time voltage exceeds the voltage over-limit probability threshold, the corresponding alarm level is triggered. Cross-site equipment faults are detected through the power grid topology. If the path trigger probability is within the range of the emergency alarm threshold, the real-time voltage exceeds the voltage over-limit probability threshold, and the fault propagation path includes cross-site equipment, it is determined to be an emergency alarm. If the path trigger probability is within the range of emergency alarm and important alarm thresholds and exceeds the voltage over-limit probability threshold, or if a single device is overloaded and the real-time voltage exceeds the voltage over-limit probability threshold, it is determined to be an important alarm. If the path trigger probability is within the range of general alarm thresholds and the real-time voltage does not exceed the voltage over-limit probability threshold, it is determined to be a general alarm.
[0064] The intelligent power grid event analysis module analyzes power grid data in real time, generates response plans, and pushes abnormal data and power grid fault information. Based on the fault propagation path and power grid topology model, it automatically calculates the load loss by utilizing equipment relationships and expert rule weights in the power grid knowledge graph. The load shedding scheme is generated, and its calculation process is as follows:
[0065] ;
[0066] in, This represents the active power at the i-th load point. This represents the priority weight of the i-th load point. Based on the total number of load points and a simulation exercise using a wiring diagram, the risk points of the load shedding scheme are analyzed, and the strategy is dynamically optimized. Through the integration of expert rules and machine learning, auxiliary decision-making that balances safety and efficiency is output, supporting dispatchers to respond quickly. According to the alarm level and voltage over-limit probability threshold, hierarchical alarms are dynamically triggered. Abnormal data is classified according to voltage level and over-limit frequency, and high-priority anomalies are prioritized and pushed to user terminals within the corresponding dispatch jurisdiction and fault impact range. Fault information is assigned to a region according to dispatch authority and pushed to mobile terminals and duty terminals. The fault information includes fault location, associated fault list and load shedding scheme. Low-latency transmission is achieved through message queues, enabling alarms to reach users in real time.
[0067] First, the power grid data acquisition module accesses multi-source data in real time. This module obtains remote signaling and telemetry data for the entire voltage level from 10kV to 500kV from the dispatch data center, and simultaneously connects with the OMS system for maintenance requests and trip records, as well as the D5000 system for monitoring operations and tagging data. It uses logical rules to eliminate redundant remote signaling signals based on unreasonable transformer paralleling and mismatched switch positions. Based on the 3σ principle, it cleans up telemetry anomalies, fills in missing data, and generates a standardized time-series dataset. Next, the power grid feature extraction module performs structured processing on the preprocessed data, targeting the remote signaling data... According to the algorithm, alarm types are mapped to numerical labels, and binary vectors of switch open / close states are extracted. For unstructured maintenance requests from the OMS system, bidirectional long short-term memory network segmentation and conditional random field entity recognition technology are used to extract equipment entities and maintenance window indicators. For telemetry data, active power fluctuation rate and voltage over-limit frequency are calculated through a sliding window to form a dynamic feature matrix. Then, the monitoring information event analysis module constructs an intelligent analysis model, determines the equipment maintenance status through a decision tree model, outputs maintenance and debugging information labels, and combines support vector machine classification algorithms with historical fault patterns. The system classifies alarm types, active power fluctuations, and tripping records into fault categories, outputting fault timestamps and a list of associated devices. Using a decision tree clustering algorithm, fault events are clustered according to a 5-minute time window and topological hierarchy, generating a multi-layered fault list including cross-site associations. Subsequently, the power grid graph construction module dynamically generates a knowledge graph, using the tripped device primary key as the core node, integrating maintenance tags, fault tags, and clustering results to construct the power grid knowledge graph. The path weights are dynamically attenuated based on equipment maintenance status, and the cross-site fault propagation path weights are increased by combining voltage over-limit frequency, generating a weighted knowledge graph. The fault propagation path is analyzed, and the voltage over-limit threshold is dynamically adjusted based on the path trigger probability and expert coefficient to trigger a three-level alarm. Finally, the power grid event intelligent judgment module realizes real-time response and push. Based on the fault propagation path and over-limit threshold output by the knowledge graph, the load shedding plan is automatically generated, and the risk handling is verified through a wiring diagram simulation exercise. Abnormal data is pushed to the D5000 system according to voltage level and over-limit frequency, and fault information is pushed to the mobile terminal according to the scheduling authority. Emergency alarms are reached within 5 seconds, important alarms are pushed to the regional person in charge, and general alarms are recorded in the log.
[0068] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A power grid operation status analysis system combining knowledge graphs, characterized in that: It includes a power grid data acquisition module, a power grid feature extraction module, a monitoring information event analysis module, a power grid map construction module, and a power grid event intelligent judgment module; The power grid data acquisition module acquires and preprocesses power grid data, which includes remote signaling data, telemetry data, OMS system data, and D5000 system data. The power grid feature extraction module extracts remote signaling features, maintenance event features, and telemetry features from the preprocessed power grid data. The monitoring information event analysis module combines remote signaling features, maintenance event features, and telemetry features to construct a maintenance event analysis model, a power grid fault analysis model, and a power grid fault clustering model, and outputs equipment maintenance and debugging information tags, fault time tags, a list of associated equipment, and a list of associated faults. The power grid graph construction module constructs a power grid knowledge graph based on equipment maintenance and debugging information tags, fault time tags, a list of associated equipment, and a list of associated faults. It outputs fault propagation paths and voltage over-limit probability thresholds, and provides tiered early warnings. The intelligent power grid event analysis module analyzes power grid data in real time, generates handling plans, and pushes abnormal data and power grid faults. In the monitoring information event analysis module, the process of constructing a maintenance event analysis model and outputting equipment maintenance and debugging information tags includes: Using the switch open / closed state vector and maintenance window period flag as input features, a decision tree model is adopted to construct a maintenance event analysis model based on whether the current time falls within the approved maintenance application period and whether the switch open / closed state is open. The root node of the decision tree prioritizes splitting the maintenance window period flag to filter out equipment that is in the maintenance period. The child nodes split the switch opening and closing status. If both the maintenance application period and the switch opening condition are met, the power grid equipment is determined to be in the maintenance and commissioning state. 0 represents non-maintenance and commissioning state and 1 represents maintenance and commissioning state. Finally, the leaf nodes output the equipment maintenance and commissioning information tag. In the event-driven analysis module for monitoring information, the process of constructing a power grid fault analysis model and outputting fault time stamps and a list of associated devices includes: Using alarm type code, active power fluctuation rate and trip record parsing results as input features, a fault mode rule base is established based on historical fault cases. The input features are compared with the conditions in the fault mode rule base one by one. If the alarm type code, active power fluctuation rate and trip device association information meet the rule conditions, the corresponding fault label is triggered. The alarm type code, active power fluctuation rate, and tripped device primary key are converted into input feature vectors. The support vector classification algorithm is used to find the optimal classification hyperplane based on the principle of maximizing the margin. The fault category label is divided according to the projection position of the input feature vector on the hyperplane. By combining the topology of the power grid model, the path of fault impact is determined, and the primary key of the equipment and the topology-related equipment in the trip record are extracted to form a list of related equipment; Based on the alarm occurrence time and the fault mode rule matching results, mark the fault start timestamp, and output the fault timestamp and a list of associated devices including the fault source device and the topology-related devices within a single fault event; The process of constructing a power grid fault clustering model and outputting a list of associated faults in the monitoring information event analysis module includes: Using the fault start timestamp, the primary key of the tripped device, and the fault category label as input features, the time range of fault association is defined, and two fault events with a timestamp interval not exceeding the length of the time window are regarded as associated events within the same time window. Check if the primary keys of the tripped devices are the same. If they are the same, they are classified into the same fault cluster. If they are different, they are not classified into the same fault cluster. According to the power grid topology model, if the faulty devices belong to the same bay and the same substation, they are classified into the same cluster. Through the topological connection relationship, cross-station equipment faults are included in the same fault cluster. A decision tree clustering algorithm is adopted. The root node of the decision tree uses the time window as the first splitting condition to filter out fault events within the same time window. The child nodes split according to the primary key of the tripping device to distinguish faults from different source devices. The algorithm is refined layer by layer according to the topology level to generate fault subclusters and construct a power grid fault clustering model. Each leaf node represents a fault cluster and outputs an associated fault list and fault cluster labels. The associated fault list is a list of multiple fault events that are associated within the same time window, the same device, and the same topology. The process of constructing a power grid knowledge graph and outputting fault propagation paths and voltage over-limit probability thresholds in the power grid graph construction module includes: A power grid knowledge graph is constructed using equipment maintenance and commissioning information tags, fault time tags, a list of associated equipment, a list of associated faults, and the frequency of voltage over-limit as inputs. The power grid knowledge graph uses the tripped device primary key as the core to construct device node attributes. The device node attributes include device maintenance and commissioning information tags and voltage over-limit frequency. Based on the event sequence in the associated fault list, the fault propagation relationship between devices is established, and the initial expert rule weights are set according to historical fault data. If the equipment is under maintenance, its associated path weight is multiplied by the attenuation coefficient, and a voltage over-limit frequency threshold is set. If the voltage over-limit frequency exceeds its corresponding threshold, the weight of the associated path increases, and the weight of the propagation path generated by the cross-site equipment due to the grid topology connection is increased by 20%. Multiple fault events in the associated fault list are sorted by timestamp to generate fault propagation paths. The path trigger probability is obtained based on the sum of the weights of each node in the path. An expert experience coefficient is introduced, and the voltage limit exceedance probability threshold is dynamically adjusted according to the frequency of voltage exceedance and the path trigger probability. If the real-time voltage exceeds the voltage limit exceedance probability threshold, a graded warning is issued.
2. The power grid operation status analysis system combining knowledge graphs according to claim 1, characterized in that: The power grid data acquisition module includes the following processes for acquiring and preprocessing power grid data: The remote signaling data of power grid equipment with voltage levels from 10kV to 500kV is received through the real-time service interface of the scheduling data center. The remote signaling data includes alarm information such as accidents, anomalies, changes and over-limits. The specific fields are equipment primary key, occurrence time, voltage level, alarm flag bit and switch open / closed status. By scheduling the minute-by-minute measurement service interface of the data center, telemetry data of power grid equipment is periodically obtained. The telemetry data includes measurement parameters such as active power, reactive power, voltage amplitude, current value, and oil temperature. OMS system data includes maintenance request data and trip record data, while D5000 system data includes monitoring operation data and tag data; By connecting the data center to the OMS system service interface, maintenance application data and tripping record data are obtained in real time. The maintenance application data includes the power outage equipment, maintenance period and approval time, and the tripping record data includes the fault type, tripping equipment association information and tripping switch. By connecting the data center to the D5000 system service interface, monitoring operation data and tag data can be obtained in real time. The monitoring operation data includes operation time, operated device and operation result, and the tag data includes tag type, device ID and operation time. By verifying the logical rules of unreasonable parallel connection of main transformers and mismatch of switch positions, redundant and erroneous signals in remote signaling data are eliminated, and standardized equipment status data is generated. Outliers in telemetry data are identified based on the 3σ principle, and missing telemetry data are filled in by linear interpolation of adjacent time points to generate continuous telemetry time series data. By employing bidirectional long short-term memory network word segmentation and conditional random field entity recognition technology, unstructured maintenance application data and trip record data are converted into structured data associated with equipment, and the trip record parsing results are obtained. The validity of operation data and tag data is verified by matching operation results with real-time status. The correlation between tag type and equipment maintenance status is analyzed by combining support vector machine model to generate standardized operation logs.
3. The power grid operation status analysis system combining knowledge graphs according to claim 2, characterized in that: The process of extracting remote signaling features, maintenance event features, and telemetry features in the power grid feature extraction module includes: Remote signaling features include alarm type codes and switch open / closed status vectors. Alarm information is mapped to numerical labels. Through conditional judgment logic, the corresponding encoding vector is generated based on the alarm flag bit to obtain the alarm type code. The binary status value of the switch open / closed status field in the remote signaling data is extracted to generate the real-time switch open / closed status vector of the device. The maintenance event characteristics include maintenance window period flags and equipment maintenance flags. Based on maintenance application data and remote signaling data from the OMS system, it is determined whether the current time is within the approved maintenance application period. Combined with the switch opening status, a Boolean maintenance window period flag is generated, where 0 indicates a non-maintenance period and 1 indicates a maintenance period. Based on the tag data and maintenance window period flags from the D5000 system, a support vector machine model is used to analyze the correlation between tag type and equipment status, and output the equipment maintenance flag. The tag types include maintenance tags and test tags. Telemetry features include active power fluctuation rate and voltage over-limit frequency. Based on preprocessed active power time series data, a time window is set, the average active power within the window is extracted, and the average absolute deviation of active power from its mean at each time point within the window is calculated to obtain the active power fluctuation rate. Based on preprocessed voltage amplitude time series data, a rated voltage is set, and the over-limit range is defined with the rated voltage as the benchmark. The number of times the voltage exceeds 1.1 times the rated voltage and falls below 0.9 times the rated voltage per unit time is counted to obtain the voltage over-limit frequency.
4. The power grid operation status analysis system combining knowledge graphs according to claim 3, characterized in that: The process of hierarchical early warning in the power grid mapping module includes: The power grid mapping module receives fault propagation paths, voltage over-limit probability thresholds, and real-time voltage data. Based on the path triggering probability, it classifies alarms into emergency alarms, important alarms, and general alarms. It matches different alarm levels with corresponding path triggering probability threshold ranges. Based on the frequency of voltage over-limits and path weights, it dynamically adjusts the voltage over-limit probability threshold. If the real-time voltage exceeds the voltage over-limit probability threshold, the corresponding alarm level is triggered. An alarm is considered an emergency alarm if it simultaneously meets the following conditions: the path trigger probability falls within the emergency alarm threshold range, the real-time voltage exceeds the voltage over-limit probability threshold, and the fault propagation path includes cross-site equipment. An alarm is considered an important alarm if it meets one of the following conditions: the path trigger probability is within the emergency alarm and important alarm threshold range and exceeds the voltage over-limit probability threshold, or a single device is overloaded and the real-time voltage exceeds the voltage over-limit probability threshold. An alarm is considered a general alarm if it meets the following conditions: the path trigger probability falls within the general alarm threshold range and the real-time voltage does not exceed the voltage over-limit probability threshold.
5. The power grid operation status analysis system combining knowledge graphs according to claim 4, characterized in that: The process of analyzing power grid data in real time and generating a response plan in the aforementioned intelligent power grid event assessment module includes: Based on the fault propagation path and power grid topology model, and utilizing the equipment relationships and expert rule weights in the power grid knowledge graph, the loss load is automatically calculated. Generate load shedding scheme; By combining a wiring diagram simulation exercise, the risk points of the load shedding scheme are analyzed, the strategy is dynamically optimized, and by integrating expert rules and machine learning, auxiliary decision-making that balances safety and efficiency is output to support dispatchers in responding quickly.
6. The power grid operation status analysis system combining knowledge graphs according to claim 5, characterized in that: The process of pushing abnormal data and power grid faults in the intelligent power grid event analysis module includes: Based on the alarm level and voltage over-limit probability threshold, hierarchical alarms are dynamically triggered. Abnormal data is classified according to voltage level and over-limit frequency. High-priority abnormalities are pushed to the D5000 system first. Fault information is assigned to a region according to scheduling authority and pushed to mobile terminals and duty terminals. The fault information includes fault location, associated fault list and load shedding scheme. Low-latency transmission is achieved through message queues to ensure alarms reach users in real time.