A power grid dispatch multi-source data fusion and panoramic visualization management method

Abstract

The present application relates to the field of power grid dispatching, and more particularly to a power grid dispatching multi-source data fusion and panoramic visualization management method. The method first collects real-time measurement, equipment account and meteorological environment and other multi-source heterogeneous data, and performs time and space unified reference transformation to obtain aligned unified data view. Subsequently, based on the view, the topology association and business logic relationship of the power grid entity are constructed using knowledge graph technology to form a power grid knowledge graph. By mapping the unified data view to the knowledge graph in real time and performing graph calculation reasoning, abnormal events and potential risks of power grid operation are identified, and event and risk labels are generated. Finally, according to the label, a three-dimensional panoramic visualization engine is driven to dynamically render and highlight abnormal areas and risk levels, and a power grid dispatching panoramic visualization view is generated. The present application can accurately identify abnormalities and risks, provide intuitive and comprehensive visualization views for dispatchers, and improve the scientificity, timeliness and accuracy of dispatching decisions.

Description

Technical Field

[0001] This invention relates to the field of power grid dispatching, and in particular to a method for multi-source data fusion and panoramic visualization governance of power grid dispatching. Background Technology

[0002] Power grid dispatching is a core component ensuring the safe, stable, and efficient operation of the power system. The scientific nature and timeliness of its dispatching decisions directly affect the reliability of power supply, the economic efficiency of power grid operation, and the security of electricity consumption for users. With the continuous advancement of smart grid construction, data sources in the field of power grid dispatching are exhibiting significant characteristics of diversification, heterogeneity, and real-time processing. Various types of data related to power grid dispatching are constantly emerging, mainly including real-time measurement data from dispatch automation systems, equipment ledger data from production management systems, and meteorological environmental data from meteorological service centers. This multi-source, heterogeneous data covers the real-time status of power grid operation, basic equipment information, and external environmental influencing factors, providing crucial data support for power grid dispatching decisions, operation monitoring, and risk early warning.

[0003] However, current power grid dispatch data governance and visualization applications still face core technological bottlenecks, severely restricting the improvement of power grid dispatch intelligence and failing to meet the needs of modern power grid refined dispatch. First, multi-source heterogeneous data originates from different system platforms, and the lack of unified standards for time references and spatial coordinates among data sources leads to serious time asynchrony and coordinate inconsistencies. For example, there are discrepancies between the timestamps of real-time measurement data and meteorological environmental data, and the coordinate formats of equipment location coordinates in equipment ledger data are inconsistent with those in the power grid geographic information system. This prevents effective data correlation and fusion, resulting in fragmented data that cannot comprehensively reflect the overall state of power grid operation. Second, existing data governance methods lack effective analysis of the topological relationships and business logic between power grid entities, only achieving simple data storage and display. They cannot uncover the underlying power grid operation patterns and cannot identify abnormal events and potential risks in power grid operation through data correlation analysis. Finally, existing visualization methods are mostly two-dimensional static displays, unable to deeply integrate multi-source fused data with the power grid entity topology and real-time operating status. Dispatchers cannot intuitively and comprehensively grasp the overall operating status of the power grid, and the identification of abnormal areas and risk levels is delayed, severely impacting the timeliness and accuracy of dispatch decisions. The aforementioned problems overlap, preventing the full realization of the value of multi-source data and hindering the transformation of power grid dispatch from "experience-driven" to "data-driven." They also prevent the provision of comprehensive, accurate, and efficient support for power grid dispatch decisions, thus necessitating a governance method that can effectively integrate multi-source data, unify its temporal and spatial dimensions, establish topological correlations, and provide panoramic visualization. Summary of the Invention

[0004] This invention provides a method for the fusion of multi-source data and panoramic visualization governance of power grid dispatching. The purpose is to solve the core technical problems in the existing power grid dispatching data governance, such as incomplete data utilization, inconsistent spatiotemporal benchmarks, lack of effective topological correlation, and insufficient visualization presentation, which lead to untimely and low-accuracy identification of abnormal events and potential risks, and the inability to dynamically reflect the panoramic operation status of the power grid.

[0005] To achieve the above objectives, the following technical solution is adopted.

[0006] A method for multi-source data fusion and panoramic visualization governance of power grid dispatching includes the following steps: Collect multi-source heterogeneous data related to power grid dispatching, including real-time measurement data, equipment ledger data and meteorological environmental data, to obtain the original multi-source dataset; The original multi-source dataset is subjected to a spatiotemporal unified reference transformation to eliminate the time asynchrony and coordinate inconsistency between data sources, resulting in a spatiotemporally aligned unified data view. Based on the unified data view, knowledge graph technology is used to construct the topological associations and business logic relationships between power grid entities, resulting in a power grid knowledge graph. The unified data view is mapped onto the power grid knowledge graph in real time, and graph computing reasoning is performed to identify abnormal power grid operation events and potential risks, and to obtain event and risk labels. Based on the event and risk labels, the 3D panoramic visualization engine is driven to dynamically render the power grid operation status map, and abnormal areas and risk levels are highlighted in the power grid operation status map to generate a panoramic visualization view of power grid dispatch.

[0007] Optionally, the step of collecting multi-source heterogeneous data related to power grid dispatch, including real-time measurement data, equipment ledger data, and meteorological environmental data, to obtain the original multi-source dataset specifically includes: The system collects voltage amplitude data, current amplitude data, active power data, reactive power data, and switch status data from each node of the power grid in real time from the dispatch automation system, forming a real-time measurement dataset. Export transformer parameter data, line parameter data, circuit breaker parameter data, disconnector switch parameter data, and generator set parameter data in batches from the production management system to form an equipment ledger dataset; The meteorological service center regularly acquires wind speed and direction data, temperature data, humidity data, light intensity data, and rainfall data for the target area to form a meteorological environment dataset. The real-time measurement dataset, the equipment ledger dataset, and the meteorological environment dataset are initially correlated according to the collection time and geographical location to generate an original multi-source dataset containing timestamps and spatial coordinate markers.

[0008] Optionally, the step of performing a spatiotemporal unified benchmark transformation on the original multi-source dataset to eliminate time asynchrony and coordinate inconsistency between data sources and obtain a spatiotemporally aligned unified data view specifically includes: Extract the timestamp information of each data record in the original multi-source dataset. Using the master clock of the scheduling automation system as the reference time source, the network time protocol synchronization algorithm is used to correct the deviation of the timestamps of other data sources. The corrected timestamps are then uniformly converted into Coordinated Universal Time format to obtain the time-aligned multi-source dataset. Geographic coordinate data of all substations, line towers and power plants are extracted from the power grid geographic information system map to establish a benchmark database of the spatial location of power grid entities; For each data point in the time-aligned multi-source dataset, match its corresponding geographic coordinates in the power grid entity spatial location reference library according to its associated device name or number, and append the geographic coordinates to the data record in a unified latitude and longitude coordinate format to obtain the spatially aligned multi-source dataset. The spatially aligned multi-source datasets are resampled and interpolated according to the temporal order and spatial grid to generate a unified spatiotemporally aligned data view that covers the entire scheduling region and has a uniform temporal and spatial resolution.

[0009] Optionally, the step of constructing the topological associations and business logic relationships between power grid entities based on a unified data view and using knowledge graph technology to obtain a power grid knowledge graph specifically includes: Extract all power grid equipment entities from the unified data view, including generator sets, transformers, busbars, circuit breakers, disconnectors, transmission lines, and loads, and assign a unique identifier to each entity. Based on the primary wiring diagram and secondary protection configuration diagram of the power grid, establish the electrical connection relationship between entities, create "connection" type edges between entities with direct electrical connections, and label the attributes of the edges, including voltage level and phase sequence; Based on the dispatching operation procedures and business rules, establish business logic relationships between entities, including the "power supply" relationship between generator sets and transmission lines, the "transformation" relationship between transformers and downstream busbars, and the "protection" relationship between circuit breakers and protection devices, and create corresponding types of business logic edges; The entities, as well as the "connection" type edges and business logic edges, are stored in a structured manner to form a power grid knowledge graph with entities as nodes and edges as relationships. Each node in the power grid knowledge graph is associated with corresponding real-time measurement data, equipment ledger data, and meteorological environment data.

[0010] Optionally, the step of mapping the unified data view onto the power grid knowledge graph in real time, performing graph computation and reasoning, identifying abnormal power grid operation events and potential risks, and obtaining event and risk labels specifically includes: The latest real-time measurement data in the unified data view is updated to the attributes of the corresponding nodes in the power grid knowledge graph in real time, so that the node attributes in the power grid knowledge graph are synchronized with the actual operating status of the power grid. Deploy predefined abnormal event detection rules on the power grid knowledge graph. The abnormal event detection rules include voltage over-limit rules, line overload rules, power reversal rules, and frequency fluctuation rules. When the rule conditions are triggered, mark the node or edge where the abnormal event occurs in the power grid knowledge graph and generate an abnormal event label. A graph neural network model is used to perform real-time reasoning on the power grid knowledge graph. The graph neural network model predicts the trend of the current load rate of each line or transformer in the near future by aggregating the feature information of nodes and neighboring nodes, identifying weak links with overload risk, and generating potential risk labels. Associate the abnormal event labels and potential risk labels with the corresponding nodes and edges in the power grid knowledge graph, and update the label attributes of the power grid knowledge graph.

[0011] Optionally, the step of driving the 3D panoramic visualization engine to dynamically render the power grid operation status map based on event and risk tags, and highlighting abnormal areas and risk levels in the power grid operation status map to generate a panoramic visualization view of power grid dispatch specifically includes: Read the real-time attribute data of all nodes and edges in the power grid knowledge graph, as well as the latest event and risk tags, and construct a panoramic data model of the power grid to be rendered; The power grid panoramic data model is input into the 3D panoramic visualization engine, which first loads the pre-generated 3D terrain scene and the 3D model library of power grid equipment. Based on real-time measurement data of nodes and edges, the color and brightness of the device surface in the 3D model are dynamically updated, with green indicating normal operation, yellow indicating warning, and red indicating abnormal operation. For nodes or edges labeled with abnormal events, a flashing red halo is generated around the corresponding 3D model, and the type and specific value of the abnormal event are displayed floating above it. For nodes or edges marked with potential risk labels, a gradient yellow halo is generated around the corresponding 3D model, and the risk level and predicted overload time are overlaid. All rendered 3D scenes and labeled information are combined to output an interactive panoramic visualization view of power grid dispatch, which provides functions such as view rotation, zoom, and information query.

[0012] Optionally, after the step of collecting voltage amplitude data, current amplitude data, active power data, reactive power data, and switch status data from each node of the power grid in real time from the dispatch automation system to form a real-time measurement dataset, the method further includes an online data quality assessment and cleaning step for the real-time measurement dataset, specifically including: A dynamic data quality assessment model based on Kalman filtering is established for each type of measurement data in the real-time measurement dataset. The real-time measurement data collected at the current moment is input into the dynamic data quality assessment model, and the residual between the real-time measurement data and the model prediction value is calculated to obtain the residual sequence. The residual sequence is compared with a preset residual threshold. When the residual value in the residual sequence exceeds the residual threshold three times consecutively, it is determined that the real-time measurement data at the current moment has a data quality problem, and a data quality anomaly mark is generated. For real-time measurement data with a data quality anomaly mark, a data repair program based on spatiotemporal correlation is started. Normal data sequences are extracted from the historical data of the same period of the node, and spatially related data sequences are extracted from the measurement data of the same moment of adjacent nodes. The historical data sequences and spatially related data sequences are input into the Bayesian tensor completion model. Through tensor decomposition and iterative filling algorithms, the repair value for the current abnormal data is calculated. The repaired values ​​replace the original abnormal data, and a repair marker is added to update the real-time measurement dataset.

[0013] Optionally, the step of establishing electrical connection relationships between entities based on the primary wiring diagram and secondary protection configuration diagram of the power grid, creating "connection" type edges between entities with direct electrical connections, and labeling the attributes of the edges including voltage level and phase sequence specifically includes: The connection relationships between all busbar nodes, circuit breaker nodes, disconnector nodes, transformer winding nodes, and line endpoints are parsed from the primary wiring diagram file of the power grid, and an initial electrical connection relationship table is generated. The relationship between the protection device and the circuit breaker, as well as the sampling connection relationship between the protection device and the current transformer and voltage transformer, are extracted from the power grid secondary protection configuration diagram file to generate a secondary equipment connection relationship table. By jointly analyzing the initial electrical connection table and the secondary equipment connection table, the protection logic unit consisting of circuit breakers, disconnect switches and protection devices is identified, and "protection coordination" type edges are created between entities within the protection logic unit. A graph isomorphism-based topology verification algorithm is used to compare the constructed electrical connection diagram with the templates in the power grid standard topology template library to detect missing or redundant connection edges. Based on the detection results, missing "connection" type edges are automatically added, and redundant "connection" type edges are deleted.

[0014] Optionally, the step of using a graph neural network model to perform real-time reasoning on the power grid knowledge graph, wherein the graph neural network model, by aggregating the feature information of nodes and neighboring nodes, predicts the changing trend of the current load rate of each line or each transformer in the near future, identifies weak links with overload risk, and generates potential risk labels, specifically includes: The real-time measurement data of all nodes in the power grid knowledge graph at time t are concatenated with the equipment ledger data to form the initial feature matrix of the nodes, and the rated capacity of all edges is concatenated with the current power flow value to form the initial feature matrix of the edges. The initial feature matrix of the node and the initial feature matrix of the edge are input into the pre-trained spatiotemporal graph convolutional network model. The spatiotemporal graph convolutional network model first aggregates the first-order neighbor node features and second-order neighbor node features of each node through graph convolutional layers to obtain spatial feature encoding. Spatial feature encoding and similar feature encoding sequences from multiple historical time steps are input into a gated recurrent unit layer. The time dependence of load changes is captured by the reset gate and update gate of the gated recurrent unit layer to obtain a spatiotemporal fusion feature vector. The spatiotemporal fusion feature vector is input into the fully connected output layer. Through the linear transformation and sigmoid activation function of the fully connected output layer, the load rate prediction value of each node at three time points in the next 15 minutes, 30 minutes and 60 minutes is output. The load rate prediction value of each node is compared with the rated capacity threshold of the corresponding device. If the load rate prediction value exceeds the first threshold in the next 15 minutes, it is marked as "emergency overload risk". If the load rate prediction value exceeds the second threshold in the next 30 minutes, it is marked as "short-term overload risk". If the load rate prediction value exceeds the third threshold in the next 60 minutes, it is marked as "medium-term overload risk". A potential risk label containing risk type, risk time and risk level is generated.

[0015] Optionally, the step of inputting the power grid panoramic data model into the 3D panoramic visualization engine, wherein the 3D panoramic visualization engine first loads a pre-generated 3D terrain scene and a 3D model library of power grid equipment, specifically includes: The geographic coordinate information in the power grid panoramic data model is converted into the world coordinate system position in the three-dimensional scene, and the best observation position and observation angle of the current viewpoint are determined according to the coordinate range. A progressive loading strategy based on viewpoint prediction is adopted. First, the 3D model of the device in the view frustum in the viewpoint direction is loaded. Then, the position of the next viewpoint is predicted according to the viewpoint movement trend, and the 3D terrain scene data of the area corresponding to the predicted position is loaded in advance. When loading the 3D model library of power grid equipment, for substation equipment with voltage levels higher than the preset level, a refined model containing detailed internal structure is used for loading, while for tower and line equipment with voltage levels lower than the preset level, a lightweight model containing texture and basic outline is used for loading. The loaded 3D terrain scene and the 3D model of the power grid equipment are precisely stitched and aligned according to spatial coordinates, and the seams of the models are smoothed to eliminate visual abrupt changes caused by differences in model accuracy.

[0016] Compared with the prior art, the present invention has the following beneficial effects: This application systematically collects multi-source heterogeneous data related to power grid dispatching, performs spatiotemporal unified benchmark transformation on the original multi-source datasets to eliminate time asynchrony and coordinate inconsistency issues between data sources, and constructs a spatiotemporally aligned unified data view. Based on this unified data view, knowledge graph technology is used to construct topological associations and business logic relationships between power grid entities. The unified data view is mapped to the power grid knowledge graph in real time and graph computation and reasoning are performed. Ultimately, this drives a 3D panoramic visualization engine to dynamically render a power grid operation status diagram. This effectively solves the core technical problems in existing power grid dispatching data governance, such as incomplete data utilization, inconsistent spatiotemporal benchmarks, lack of effective topological associations, and insufficient visualization presentation, leading to untimely and inaccurate identification of abnormal events and potential risks, and an inability to dynamically reflect the panoramic operation status of the power grid. It achieves efficient fusion and deep utilization of multi-source heterogeneous power grid data, accurately identifies abnormal power grid operation events and potential risks, provides dispatchers with a comprehensive and intuitive panoramic visualization view of the power grid, significantly improves the scientific, timely, and accurate nature of power grid dispatching decisions, and provides reliable data support and technical assurance for the safe, stable, and efficient operation of the power grid.

[0017] By performing online data quality assessment and cleaning of real-time measurement data, a dynamic data quality assessment model based on Kalman filtering was constructed. Combined with a Bayesian tensor completion model to repair abnormal data, the accuracy and reliability of real-time measurement data were effectively improved. This laid a high-quality data foundation for subsequent data fusion and inference analysis, avoiding interference from abnormal data in the judgment of power grid operation status and risk prediction. Through precise analysis of the primary wiring diagram and secondary protection configuration diagram of the power grid, electrical connection relationships and business logic relationships between entities were established. A graph isomorphic topology verification algorithm was used for verification, ensuring the integrity and accuracy of the power grid knowledge graph. This allows for a more accurate reflection of the relationships between power grid entities and improves the reliability of graph computation inference. Using a spatiotemporal graph convolutional network model for real-time inference of the power grid knowledge graph effectively captures the spatiotemporal characteristics of power grid load changes, accurately predicts load rate change trends in the near future, and identifies vulnerable links at risk of overload in advance, achieving proactive early warning of potential risks to the power grid. By employing a progressive loading strategy and layered model loading method in the 3D panoramic visualization engine, the rendering efficiency of the visualization scene is ensured while also taking into account the display accuracy of equipment at different voltage levels. Combined with the prominent marking of abnormal areas and risk levels, as well as the interactive view design, the ease of operation for dispatchers is greatly improved, the workload of dispatchers is reduced, and the level of intelligent power grid dispatching is further promoted, helping power grid dispatching achieve the goal of "panoramic perception, accurate decision-making, and efficient control". Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the steps of an embodiment of the power grid dispatching multi-source data fusion and panoramic visualization governance method of the present invention. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0020] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0021] like Figure 1As the power grid continues to expand, the various types of data generated during power grid dispatching exhibit characteristics of multi-source, heterogeneous, and real-time requirements. Problems such as time asynchrony and coordinate inconsistencies exist between different data sources, making effective data fusion difficult and resulting in low efficiency in presenting a panoramic view of the power grid's operational status and identifying risks. To address these technical issues, this implementation provides a method for multi-source data fusion and panoramic visualization governance of power grid dispatching. Through systematic multi-source data acquisition, spatiotemporal unification, knowledge graph construction, graph computation reasoning, and 3D panoramic visualization, it achieves efficient governance of power grid dispatching data and accurate presentation of operational status, thereby improving the intelligence level and safe operation assurance capabilities of power grid dispatching.

[0022] First, multi-source heterogeneous data related to power grid dispatch are collected to obtain the raw multi-source dataset. The comprehensiveness and accuracy of this multi-source heterogeneous data are the foundation for subsequent data fusion and governance. Therefore, the collection process must cover the core data dimensions of power grid operation, and simultaneously establish a preliminary correlation mechanism for the multi-source data to ensure data traceability and relevance. Specifically, the collection process is implemented through multi-interface adaptation and multi-protocol compatibility, employing differentiated collection strategies for different types of data to ensure the real-time performance, completeness, and reliability of data collection.

[0023] Real-time measurement data acquisition is achieved through a dedicated data acquisition interface of the dispatch automation system, using the IEC 61850 standard protocol to communicate with the system and ensure standardized and efficient data transmission. During acquisition, electrical quantity data from each node of the power grid are collected in real time according to a preset acquisition frequency. The acquisition frequency for voltage amplitude, current amplitude, active power, and reactive power data is set to milliseconds, while the acquisition frequency for switch status data is set to microseconds to capture instantaneous changes in the power grid's operating status. During acquisition, the timestamp and node identifier for each measurement data point are recorded synchronously. The timestamp is dual-marked using both the local clock and the dispatch automation system's master clock, and the node identifier uses a unique code for the power grid equipment, ensuring accurate association of data with specific power grid devices. After acquisition, the real-time measurement data undergoes preliminary format standardization processing, converting different types of measurement data into JSON format for easier subsequent data processing and storage, thus forming a real-time measurement dataset.

[0024] To further improve the quality of real-time measurement data, after constructing the real-time measurement dataset, it is necessary to conduct online data quality assessment and cleaning to avoid abnormal data affecting subsequent fusion and inference results. Specifically, firstly, a dynamic data quality assessment model based on Kalman filtering is established for each type of measurement data in the real-time measurement dataset. This model is trained using historical measurement data and can adaptively adjust prediction parameters according to dynamic changes in power grid operation. The real-time measurement data collected at the current moment is input into this dynamic data quality assessment model. Based on the historical patterns and current trends of power grid operation, the model outputs the predicted value of the measurement data at that moment. The residual between the real-time measurement data and the model's predicted value is calculated, resulting in a residual sequence. The residual sequence reflects the degree of deviation between the real-time measurement data and normal operating patterns. The residual sequence is compared with a preset residual threshold, which is set according to the characteristics of different types of measurement data and can be dynamically adjusted according to changes in the power grid operating status. When the residual value in the residual sequence exceeds the residual threshold three times consecutively, the real-time measurement data at the current moment is determined to have a data quality problem, and a data quality anomaly marker is generated. The anomaly marker includes the acquisition time, node identifier, data type, and degree of anomaly of the abnormal data. For real-time measurement data with data quality anomaly markers, a data repair program based on spatiotemporal correlation is initiated. First, normal data sequences are extracted from the historical data of the node, and normal measurement data under the same conditions in the same period within the past 3 months are selected to construct a historical data sample library. At the same time, spatially correlated data sequences are extracted from the measurement data of adjacent nodes at the same moment. The selection of adjacent nodes is based on electrical distance, selecting all nodes within an electrical distance of 5 kilometers from the abnormal data node. The historical data sequences and spatially correlated data sequences are input into a Bayesian tensor completion model. This model decomposes high-dimensional data into low-dimensional tensor factors through tensor decomposition, and combined with an iterative filling algorithm, gradually corrects the deviation of the abnormal data and calculates the repair value for the current abnormal data. After the repair is completed, the repair value replaces the original abnormal data, and a repair marker is added, specifying the repair time and basis, and updated to the real-time measurement dataset to ensure the accuracy of the real-time measurement data.

[0025] Equipment ledger data is collected through the batch export interface of the production management system, using the FTP protocol for data transmission to ensure data security and stability. During the collection process, full lifecycle parameter data for all power grid equipment is exported in batches. Transformer parameters include core parameters such as rated capacity, rated voltage, short-circuit voltage, no-load loss, and load loss; line parameters include conductor type, conductor length, resistance, reactance, and susceptance; circuit breaker parameters include rated current, rated breaking current, closing time, and opening time; disconnector parameters include rated current, dynamic stability current, and thermal stability current; and generator set parameters include rated power, rated voltage, rated speed, and regulation characteristics. During the export process, duplicate equipment ledger records are removed, and missing parameters are marked for manual supplementation later. After export, the equipment ledger data is converted to a relational data format, categorized and stored according to equipment type, forming an equipment ledger dataset. Each ledger record is associated with a unique equipment code, consistent with the node identifier in the real-time measurement data, laying the foundation for subsequent data association.

[0026] Meteorological environmental data is collected through a dedicated data interaction interface with the meteorological service center. The HTTP protocol is used to periodically acquire meteorological data from the target area. The acquisition frequency is set according to the power grid dispatch requirements, typically every 15 minutes, but increased to every 5 minutes during extreme weather events. The collected meteorological environmental data includes wind speed and direction, temperature, humidity, light intensity, and rainfall. Wind speed and direction data are in vector format, including wind speed magnitude and wind direction angle; temperature, humidity, and light intensity data are in numerical format, retaining one decimal place; rainfall data is recorded using both cumulative and instantaneous values ​​to ensure data integrity. During acquisition, the timestamps of the meteorological data and the geographical coordinates of the acquisition area are simultaneously acquired, with the geographical coordinates using latitude and longitude format to match the spatial coordinates of the power grid equipment. After acquisition, the meteorological environmental data undergoes validity verification, and data exceeding reasonable ranges are removed. For example, temperature data exceeding the range of -40℃ to 60℃ is considered invalid and discarded, thus forming the meteorological environmental dataset.

[0027] The real-time measurement dataset, equipment ledger dataset, and meteorological environment dataset are initially correlated according to collection time and geographical location to generate an original multi-source dataset containing timestamps and spatial coordinate labels. During the correlation process, timestamps are used as the first correlation dimension to match real-time measurement data, equipment ledger data, and meteorological environment data at the same timestamp. Geographical location is used as the second correlation dimension; based on the geographical coordinates of the equipment in the equipment ledger data and the geographical coordinates of the collection area in the meteorological environment data, the meteorological environment data is correlated to the corresponding power grid equipment, achieving a three-dimensional correlation of "equipment-measurement-meteorology". After the correlation is completed, a unique data identifier is added to each data record in the original multi-source dataset. The data identifier is composed of a timestamp, equipment code, and data type code, ensuring the uniqueness and traceability of the data and providing a data foundation for subsequent spatiotemporal unified benchmark transformation.

[0028] After collecting the original multi-source datasets, a spatiotemporal unified benchmark transformation is performed on them to eliminate time asynchrony and coordinate inconsistency between data sources, resulting in a spatiotemporally aligned unified data view. Spatiotemporal unification is a core prerequisite for multi-source data fusion. By unifying the time and spatial benchmarks, precise alignment of data from different data sources is achieved, providing standardized data support for subsequent knowledge graph construction and graph computation reasoning.

[0029] In the process of unifying the time base, the timestamp information of each data record in the original multi-source dataset is first extracted, including the acquisition timestamp of real-time measurement data, the update timestamp of equipment ledger data, and the acquisition timestamp of meteorological and environmental data. Due to clock discrepancies between different data sources, the timestamps are not synchronized. Therefore, the master clock of the dispatch automation system is used as the reference time source. This master clock uses a high-precision atomic clock with microsecond-level accuracy to ensure the accuracy of the time base. A network time protocol synchronization algorithm is used to correct the time discrepancies of the timestamps from other data sources. This algorithm calculates the time difference between each data source's clock and the master clock to generate a time discrepancy correction value. The timestamps of each data source are then adjusted according to the correction value to eliminate time discrepancies. After correction, the timestamps of all data records are uniformly converted to Coordinated Universal Time (UTC) format to ensure time format consistency. Through the above processing, a time-aligned multi-source dataset is obtained, with all data records maintaining a consistent time base, accurately reflecting the operating status of power grid equipment and surrounding meteorological and environmental conditions at the same time.

[0030] In the process of unifying spatial benchmarks, the first step is to extract the geographic coordinate data of all substations, line towers, and power plants from the power grid geographic information system map to construct a spatial location benchmark database for power grid entities. During the extraction process, the spatial data extraction interface of the geographic information system is used to obtain the precise geographic coordinates of power grid equipment. Substation coordinates are based on the center coordinates of their main control building, line tower coordinates on the center coordinates of their tower foundations, and power plant coordinates on the center coordinates of their main power plant building. The extracted geographic coordinate data adopts the WGS84 coordinate system to ensure the universality and accuracy of the coordinates. A unique spatial identifier is added to each power grid device, consistent with the device code, to achieve a precise association between spatial coordinates and equipment. The constructed spatial location benchmark database for power grid entities adopts a distributed storage structure, supporting rapid coordinate query and matching. The spatial coordinates of power grid equipment are updated regularly. When power grid equipment is relocated, added, or removed, the data in the benchmark database is updated promptly to ensure the timeliness of the benchmark database.

[0031] For each data point in the time-aligned multi-source dataset, its corresponding geographic coordinates are matched against the spatial location reference database of the power grid entity based on its associated device name or number. During the matching process, a combination of fuzzy and precise matching is used. First, precise matching is performed based on the device code; if a match is successful, the corresponding geographic coordinates are directly obtained. If a precise match fails, fuzzy matching is performed based on the device name, combined with the device's location information, to filter out the best-matching geographic coordinates and mark the match confidence level. When the match confidence level is below 90%, it is considered a failed match, and manual verification and supplementation are required. After matching, the geographic coordinates are appended to the data records in a unified latitude and longitude coordinate format, retaining 6 decimal places to ensure spatial coordinate accuracy. Through the above processing, a spatially aligned multi-source dataset is obtained, in which all data points contain accurate spatial coordinates, accurately reflecting the spatial distribution and operating status of power grid equipment.

[0032] To further improve data consistency and integrity, the spatially aligned multi-source datasets are resampled and interpolated according to temporal order and spatial grid. During temporal resampling, a unified time resolution is set based on the actual needs of power grid dispatching. Normally, the time resolution is set to 1 minute; when the power grid is in an abnormal operating state, the time resolution is adjusted to 10 seconds to ensure rapid changes in the power grid's operating status are captured. A linear interpolation algorithm is used to supplement missing data in the temporal dimension. For consecutive missing data, if the missing time is no more than 5 minutes, linear interpolation is performed using data from adjacent time points; if the missing time exceeds 5 minutes, it is marked as missing data and subsequently supplemented using other data sources. During spatial resampling, the entire dispatching area is divided into a uniform spatial grid. The grid size is set according to the scope of the dispatching area; normally, the grid side length is set to 1 kilometer to ensure accurate coverage of all power grid equipment. A nearest neighbor interpolation algorithm is used to supplement missing data in the spatial dimension. Based on the valid data points surrounding the missing data points, the interpolation result for the missing data is obtained through distance-weighted calculation. After resampling and interpolation, a unified spatiotemporally aligned data view is generated, covering the entire scheduling area with uniform temporal and spatial resolution. This data view integrates real-time measurement data, equipment ledger data, and meteorological environmental data, achieving spatiotemporal unification of multi-source data and providing standardized data input for subsequent knowledge graph construction.

[0033] Based on a unified data view aligned to spatiotemporal coordinates, a knowledge graph is constructed using knowledge graph technology to establish topological associations and business logic relationships between power grid entities, resulting in a power grid knowledge graph. This knowledge graph can structurally organize dispersed multi-source data, clearly presenting the relationships between power grid entities, providing logical support for identifying abnormal power grid operation events and potential risks, while simultaneously improving data interpretability and utilization efficiency. During the construction process, steps such as entity extraction, relationship construction, topology verification, and structured storage are employed to achieve accurate construction of the power grid knowledge graph. Furthermore, an innovative topology verification mechanism is introduced to enhance the accuracy and reliability of the knowledge graph.

[0034] First, all power grid equipment entities are extracted from a unified data view. A deep learning-based entity recognition algorithm, combined with a power grid domain dictionary, is used to automatically extract these entities, including generator sets, transformers, busbars, circuit breakers, disconnectors, transmission lines, and loads. During entity extraction, the text data in the unified data view is first segmented to remove stop words and extract keywords. Then, the keywords are categorized using an entity recognition model to identify different types of power grid equipment entities. Finally, each entity is assigned a unique identifier, composed of a combination of equipment type code and equipment number, ensuring uniqueness. Simultaneously, the entity's attribute information is associated with the unique identifier, including equipment parameter data, real-time measurement data, and meteorological environmental data, achieving comprehensive association of entity attributes. For ambiguous entities encountered during extraction, such as equipment entities with similar names, manual verification is performed using the equipment's spatial coordinates and parameter data to ensure accuracy.

[0035] Based on the primary wiring diagram and secondary protection configuration diagram of the power grid, electrical connection relationships between entities are established. "Connection" type edges are created between entities with direct electrical connections, and the edge attributes, including voltage level and phase sequence, are labeled. Specifically, firstly, the connection relationships between all busbar nodes, circuit breaker nodes, disconnector nodes, transformer winding nodes, and line endpoints are parsed from the primary wiring diagram file. A graphic recognition algorithm is used to parse the primary wiring diagram, extracting equipment nodes and connecting lines to generate an initial electrical connection relationship table. This table includes the identifiers of the two connected entities, the connection method, and the connection location. Then, the association relationships between protection devices and circuit breakers, as well as the sampling connection relationships between protection devices and current transformers and voltage transformers, are parsed from the secondary protection configuration diagram file. Similarly, a graphic recognition algorithm is used to parse this table, generating a secondary equipment connection relationship table. This table includes protection device identifiers, associated equipment identifiers, and connection types.

[0036] The initial electrical connection table and the secondary equipment connection table are jointly analyzed to identify protection logic units composed of circuit breakers, disconnectors, and protection devices. The identification of protection logic units is based on the protection range; circuit breakers, disconnectors, and protection devices that can achieve the same equipment protection function are grouped into one protection logic unit. "Protection coordination" type edges are created between entities within the protection logic unit, and the attributes of the edges include protection priority, action threshold, and action time, achieving accurate modeling of secondary protection relationships. To ensure the accuracy of electrical connection relationships, a graph isomorphism-based topology verification algorithm is used. The constructed electrical connection diagram is compared with templates in the power grid standard topology template library. The power grid standard topology template library contains power grid topology templates with different voltage levels and different wiring methods. The similarity between the constructed electrical connection diagram and the standard template is calculated using the graph isomorphism algorithm. When the similarity is less than 95%, a topology anomaly is determined, and missing or redundant connection edges are detected. Based on the detection results, missing "connection" type edges are automatically added and redundant "connection" type edges are deleted. At the same time, the modified electrical connection diagram is re-verified until the similarity reaches more than 95%, ensuring the accuracy and rationality of the electrical connection relationship.

[0037] Based on the dispatching operation procedures and business rules, establish business logic relationships between entities, including the "power supply" relationship between generator sets and transmission lines, the "voltage transformation" relationship between transformers and downstream busbars, and the "protection" relationship between circuit breakers and protection devices, creating corresponding types of business logic edges. The "power supply" relationship is established based on the output direction of the generator set. When the generator set's active power is output to the transmission line, a "power supply" edge is created between the generator set entity and the transmission line entity. The edge attributes include power supply capacity, power supply voltage, and power supply time. The "voltage transformation" relationship is established based on the voltage conversion function of the transformer. A "voltage transformation" edge is created between the transformer entity and the downstream busbar entity. The edge attributes include transformation ratio, rated capacity, and loss value. The "protection" relationship is established based on the protection range of the protection device. A "protection" edge is created between the protection device entity and the protected equipment entity. The edge attributes include protection type, operating current, and operating time. During the establishment of business logic relationships, the rationality and accuracy of the relationships are ensured by combining the specific requirements of the scheduling and operation procedures. At the same time, dynamic updates of business rules are supported, and business logic relationships can be quickly adjusted when the scheduling and operation procedures change.

[0038] Entities, "connection" type edges, and business logic edges are structured and stored to form a power grid knowledge graph with entities as nodes and edges as relationships. A graph database, Neo4j, is used for storage, offering efficient graph query and computation capabilities to meet the real-time update and query requirements of the power grid knowledge graph. Entities are treated as nodes in the graph database, with attributes including a unique identifier, equipment type, parameter data, real-time measurement data, and meteorological environmental data. "Connection" type edges and business logic edges are treated as edges, with attributes including edge type, association attributes, and update time. Indexes are also established for entity and edge attributes to improve data query efficiency; for example, equipment code, entity type, and edge type indexes are created to ensure quick retrieval of target entities and relationships. Furthermore, an update mechanism for the power grid knowledge graph is established. When data in the unified data view is updated, the node and edge attributes in the power grid knowledge graph are updated synchronously to ensure the power grid knowledge graph remains synchronized with the actual operating status of the power grid.

[0039] A unified data view is mapped in real time onto the power grid knowledge graph for graph computational reasoning, identifying abnormal events and potential risks in power grid operation and obtaining event and risk labels. Graph computational reasoning can fully utilize the entity relationships in the power grid knowledge graph, combined with real-time measurement data, to achieve rapid identification of abnormal events and accurate prediction of potential risks, providing a scientific basis for power grid dispatching decisions. During the reasoning process, through steps such as real-time data updates, abnormal event detection, potential risk prediction, and label updates, dynamic monitoring and risk early warning of the power grid's operating status are achieved. At the same time, an innovative graph neural network reasoning model is introduced to improve the accuracy and real-time performance of risk prediction.

[0040] First, the latest real-time measurement data from the unified data view is updated in real-time to the attributes of the corresponding nodes in the power grid knowledge graph. The update mechanism uses an incremental approach, updating only the real-time measurement data that has changed, avoiding resource waste and delays caused by full updates. The update frequency is consistent with the acquisition frequency of the real-time measurement data: key electrical quantities such as voltage and current amplitudes are updated at millisecond levels, and instantaneous change data such as switch status are updated at microsecond levels, ensuring that the node attributes in the power grid knowledge graph remain synchronized with the actual operating status of the power grid. During the update process, the data update time and the numerical changes before and after the update are recorded synchronously, providing a basis for subsequent anomaly tracing and analysis.

[0041] Predefined anomaly detection rules are deployed on the power grid knowledge graph. These rules are based on power grid dispatching and operation procedures and fault handling experience, covering various common anomalies in power grid operation, including voltage over-limit rules, line overload rules, power reversal rules, and frequency fluctuation rules. Voltage over-limit rules are set according to power grid operation standards for different voltage levels. A voltage over-limit rule is triggered when the voltage amplitude at a node exceeds ±10% of the rated voltage for 100 milliseconds. Line overload rules are set according to the rated capacity of the line. A line overload rule is triggered when the real-time active power of the line exceeds 110% of the rated capacity for 200 milliseconds. Power reversal rules are set according to the normal power flow direction of the line. A power reversal rule is triggered when the power flow direction of the line is opposite to the preset direction for 300 milliseconds. Frequency fluctuation rules are set according to the power grid frequency standard. A frequency fluctuation rule is triggered when the power grid frequency is below 49.5Hz or above 50.5Hz for 500 milliseconds. When a rule condition is triggered, the node or edge where the abnormal event occurred is marked in the power grid knowledge graph, and an abnormal event label is generated. The abnormal event label includes the abnormal event type, occurrence time, occurrence location, abnormal value, and severity. The severity is divided into three levels: general abnormality, moderate abnormality, and severe abnormality, which are determined based on the abnormal value and duration.

[0042] A graph neural network model is used to perform real-time reasoning on the power grid knowledge graph, predicting the current load rate trend of each line or transformer in the near future, identifying weak links with overload risk, and generating potential risk labels. The graph neural network model adopts a pre-trained spatiotemporal graph convolutional network model, which combines the spatial feature extraction capability of graph convolutional networks with the temporal feature capture capability of gated recurrent units, effectively mining the spatiotemporal correlation characteristics of power grid operation data and improving the accuracy of load rate prediction. The specific reasoning process is as follows: First, the real-time measurement data of all nodes in the power grid knowledge graph at time t is concatenated with the equipment ledger data to form the initial feature matrix of the nodes. The real-time measurement data includes electrical quantity data such as voltage, current, and power, and the equipment ledger data includes parameter data such as rated capacity, resistance, and reactance. The dimension of the initial feature matrix of the nodes is the number of nodes × the feature dimension. Then, the rated capacity of all edges is concatenated with the current power flow value to form the initial feature matrix of the edges. The dimension of the initial feature matrix of the edges is the number of edges × the feature dimension.

[0043] The initial feature matrices of nodes and edges are input into a pre-trained spatiotemporal graph convolutional network model. This model first aggregates the first-order and second-order neighbor node features of each node through graph convolutional layers. The graph convolutional layers employ an attention mechanism, assigning different weights to the features of different neighbor nodes. These weights are determined based on the electrical distance and correlation strength between nodes; the closer the electrical distance and the stronger the correlation, the greater the weight. This aggregation yields the spatial feature encoding for each node, which reflects the correlation between the node and its surrounding nodes, as well as differences in operational states. The spatial feature encoding, along with similar feature encoding sequences from multiple historical time steps, is then input into a gated recurrent unit layer. This gated recurrent unit layer controls the transmission of feature information through reset and update gates. The reset gate forgets irrelevant historical feature information, while the update gate updates the current feature information, thereby capturing the temporal dependence of load changes and obtaining a spatiotemporal fusion feature vector. This spatiotemporal fusion feature vector simultaneously reflects the spatial correlation and temporal trend of node operational states.

[0044] The spatiotemporal fusion feature vector is input to the fully connected output layer. Through linear transformation and the sigmoid activation function of the fully connected output layer, the load rate prediction values ​​for each node at three time points—15 minutes, 30 minutes, and 60 minutes—are output. The load rate prediction is calculated based on the node's real-time load and rated capacity, combined with trend information from the spatiotemporal fusion feature vector to achieve accurate prediction. The load rate prediction value for each node is compared with the rated capacity threshold of the corresponding equipment. The first, second, and third thresholds are set according to the equipment type and operating conditions and can be dynamically adjusted based on the grid operating status. If the load rate prediction value for the next 15 minutes exceeds the first threshold, it is marked as "emergency overload risk," requiring immediate dispatch measures. If the load rate prediction value for the next 30 minutes exceeds the second threshold, it is marked as "short-term overload risk," requiring short-term adjustments to the operating mode. If the load rate prediction value for the next 60 minutes exceeds the third threshold, it is marked as "medium-term overload risk," requiring the development of a medium- to long-term dispatch plan. Generate potential risk labels that include risk type, risk time, and risk level. Potential risk labels are associated with abnormal event labels. If a node has both abnormal events and potential risks, the abnormal event label is marked first, and a potential risk warning is attached.

[0045] Abnormal event tags and potential risk tags are associated with corresponding nodes and edges in the power grid knowledge graph, updating the tag attributes of the power grid knowledge graph. Simultaneously, the generation time and basis of the tags are recorded for subsequent querying and analysis. Tag updates are implemented in real-time; when a new abnormal event or potential risk is detected, the corresponding tag attributes are immediately updated; when the abnormal event is handled or the potential risk is eliminated, the corresponding tag is promptly deleted to ensure the timeliness and accuracy of the tags.

[0046] Based on event and risk tags, a 3D panoramic visualization engine dynamically renders a power grid operation status map, highlighting abnormal areas and risk levels to generate a panoramic visualization view of power grid dispatch. 3D panoramic visualization presents the spatial distribution, operation status, abnormal events, and potential risks of the power grid in an intuitive way, improving the decision-making efficiency of power grid dispatchers and enabling panoramic monitoring and precise control of the power grid's operation status. The rendering process is implemented step-by-step through data model construction, 3D scene loading, dynamic rendering, anomaly labeling, and scene compositing, introducing a creative progressive loading and refined rendering strategy to balance visualization effects and operational efficiency.

[0047] First, real-time attribute data of all nodes and edges in the power grid knowledge graph, along with the latest event and risk labels, are read to construct a panoramic data model of the power grid to be rendered. The panoramic data model includes the spatial coordinates of power grid equipment, 3D model parameters, real-time operating data, and event and risk labels. The spatial coordinates use the WGS84 coordinate system, consistent with the power grid entity spatial location benchmark library. The 3D model parameters are determined according to the equipment type; different types of equipment correspond to different 3D model parameters. Real-time operating data includes electrical quantities such as voltage, current, and power, which are updated to the model in real time. Event and risk labels contain specific information about anomalies and risks for subsequent annotation and display. During the construction process, the data undergoes format conversion and standardization to ensure it is compatible with the rendering requirements of the 3D panoramic visualization engine. Simultaneously, the data is compressed to reduce the pressure on data transmission and rendering.

[0048] The power grid panoramic data model is input into the 3D panoramic visualization engine. The engine first loads a pre-generated 3D terrain scene and a library of 3D power grid equipment models. During loading, a progressive loading strategy based on viewpoint prediction is adopted to improve loading efficiency and user experience. First, the geographic coordinate information in the power grid panoramic data model is converted into a world coordinate system position in the 3D scene. The optimal viewing position and viewing angle for the current viewpoint are determined based on the coordinate range. The optimal viewing position ensures a complete presentation of the power grid equipment in the entire dispatch area, and the optimal viewing angle clearly shows the spatial distribution and interrelationships of the power grid equipment. Then, a progressive loading strategy is adopted. First, the 3D models of the equipment within the view frustum in the viewpoint direction are loaded. The range of the view frustum is determined based on the current viewpoint's viewing angle and field of view, ensuring that the user can see the equipment within the current field of view first. Then, the next viewpoint position is predicted based on the viewpoint movement trend. During the prediction process, the 3D terrain scene data of the area corresponding to the predicted position is pre-loaded, taking into account the user's historical operating habits and key areas of power grid operation, reducing loading delays when the user moves the viewpoint.

[0049] When loading the 3D model library of power grid equipment, a differentiated loading strategy is adopted. Based on the voltage level and importance of the equipment, 3D models of varying precision are loaded to balance visualization effects and operational efficiency. For substation equipment with voltage levels higher than the preset level, a detailed model containing internal structural details is loaded. This detailed model clearly displays the specific structure and connection relationships of transformers, busbars, circuit breakers, and other equipment within the substation, and can even show the internal components and operating status of the equipment. For towers and line equipment with voltage levels lower than the preset level, a lightweight model containing textures and basic outlines is loaded. This lightweight model reduces the amount of data in the model while maintaining equipment recognizability, improving loading and rendering efficiency. During the loading process, texture mapping and lighting processing are performed on the 3D models. High-resolution equipment texture images are used for texture mapping to ensure the realism of the model; global illumination algorithms are used for lighting processing to simulate natural lighting effects and enhance the visual experience of the scene.

[0050] The loaded 3D terrain scene and the 3D model of the power grid equipment are precisely stitched and aligned according to spatial coordinates. A coordinate calibration algorithm is used during the stitching process to ensure accurate matching between the 3D model and the terrain scene, avoiding situations where the model floats or is embedded in the terrain. The seams between the models are smoothed using a texture blending algorithm to eliminate visual abrupt changes caused by differences in model precision, making the entire 3D scene more coherent and natural. Simultaneously, collision detection is performed on the 3D scene to prevent overlap or intersection between different equipment models, ensuring the scene's plausibility.

[0051] Based on real-time measurement data of nodes and edges, the color and brightness of the equipment surface in the 3D model are dynamically updated. Different colors distinguish the operating status of the equipment, facilitating quick identification by dispatchers. Green indicates normal operation; when the real-time measurement data is within the rated range, the equipment model surface is displayed in green with moderate brightness. Yellow indicates a warning state; when the real-time measurement data is close to but not exceeded, the equipment model surface is displayed in yellow with slightly increased brightness to alert dispatchers. Red indicates an abnormal state; when the real-time measurement data exceeds the rated threshold or an abnormal event tag is present, the equipment model surface is displayed in red with significantly increased brightness to highlight the abnormal equipment. The update frequency of color and brightness is consistent with the update frequency of real-time measurement data, ensuring real-time reflection of changes in the equipment's operating status.

[0052] For nodes or edges labeled with abnormal events, a flashing red halo is generated around the corresponding 3D model, with a flashing frequency set to twice per second to quickly attract the attention of dispatchers. Simultaneously, a floating display of the abnormal event type and specific value is shown above it. This floating information is displayed in a semi-transparent window, containing the abnormal event type, occurrence time, abnormal value, and severity, such as "Voltage Limit Exceeded: 10kV Node, Voltage 11.2kV, Severe Abnormality, Occurrence Time: 2026-03-05 10:00:00". The floating information moves with the viewpoint, ensuring dispatchers can clearly see the abnormal information from any perspective. When dispatchers click on the abnormal equipment model, more detailed abnormal information is displayed, including historical trends in the abnormal data and the operating status of related equipment, providing support for abnormal handling.

[0053] For nodes or edges marked with potential risk labels, a gradient yellow halo is generated around the corresponding 3D model. The halo's gradient speed is set to once per second, gradually fading from the inside out to highlight the potential risk area. Simultaneously, the risk level and predicted overload time are displayed in yellow text above the halo, such as "Emergency Overload Risk, Predicted Overload in 15 Minutes," allowing dispatchers to understand the severity and timing of the risk. When dispatchers click on a device model with a potential risk, detailed risk prediction data is displayed, including predicted load rates at different future time points and the criteria for determining the risk level, providing a reference for dispatchers to formulate response measures.

[0054] All rendered 3D scenes and labeled information are composited to output an interactive panoramic visualization view of power grid dispatching. During the compositing process, anti-aliasing is applied to the 3D scenes to improve their clarity; the labeled information is optimized for layout to avoid information overlap and ensure readability. The generated panoramic visualization view supports multiple interactive functions, including view rotation, zoom, and information query. The view rotation function allows dispatchers to control the rotation direction of the view using a mouse or keyboard, providing a comprehensive view of the spatial distribution and operating status of power grid equipment. The zoom function allows dispatchers to zoom in or out of the scene; zooming in allows viewing detailed equipment information, while zooming out provides a panoramic view of the entire dispatching area. The information query function allows dispatchers to click on equipment models to query equipment parameter data, real-time measurement data, event and risk tags, and other information. It also supports filtering queries by equipment type, operating status, anomaly type, and other criteria, improving data retrieval efficiency.

[0055] In addition, the panoramic visualization view also supports data export and alarm functions. Dispatchers can export the current power grid operation status diagram, abnormal event information and potential risk information as images or documents for subsequent analysis and reporting. When a serious abnormal event or emergency overload risk is detected, the system will automatically issue an audible and visual alarm to remind dispatchers to handle it in a timely manner and ensure the safe and stable operation of the power grid.

[0056] Through the specific implementation methods described above, efficient fusion and panoramic visualization governance of multi-source data for power grid dispatching were achieved, resolving issues such as heterogeneous multi-source data, spatiotemporal asynchrony, and delayed risk identification in existing technologies. Accurate data collection and quality cleaning ensured data accuracy and integrity; unified spatiotemporal benchmark transformation achieved precise alignment of different data sources; the construction of a power grid knowledge graph clearly presented the relationships between power grid entities; graph computation and reasoning enabled rapid identification of abnormal events and accurate prediction of potential risks; and three-dimensional panoramic visualization provided an intuitive presentation and precise control of the power grid's operational status. This method exhibits significant innovation and non-obviousness, significantly improving the intelligence level and safe operation assurance capabilities of power grid dispatching compared to existing technologies. It can meet the dispatching needs of large-scale and complex power grid operations and has broad application prospects.

[0057] As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments described above are merely illustrative in all respects and are not the only ones. All modifications within the scope of this invention or its equivalents are included in this invention.

Claims

1. A method for multi-source data fusion and panoramic visualization governance of power grid dispatching, characterized in that, Includes the following steps: Collect multi-source heterogeneous data related to power grid dispatching, including real-time measurement data, equipment ledger data and meteorological environmental data, to obtain the original multi-source dataset; The original multi-source dataset is subjected to a spatiotemporal unified reference transformation to eliminate the time asynchrony and coordinate inconsistency between data sources, resulting in a spatiotemporally aligned unified data view. Based on the unified data view, knowledge graph technology is used to construct the topological associations and business logic relationships between power grid entities, resulting in a power grid knowledge graph. The unified data view is mapped onto the power grid knowledge graph in real time, and graph computing reasoning is performed to identify abnormal power grid operation events and potential risks, and to obtain event and risk labels. Based on the event and risk labels, the 3D panoramic visualization engine is driven to dynamically render the power grid operation status map, and abnormal areas and risk levels are highlighted in the power grid operation status map to generate a panoramic visualization view of power grid dispatch.

2. The method for multi-source data fusion and panoramic visualization governance of power grid dispatching according to claim 1, characterized in that, The step of collecting multi-source heterogeneous data related to power grid dispatch, including real-time measurement data, equipment ledger data, and meteorological environmental data, to obtain the original multi-source dataset specifically includes: The system collects voltage amplitude data, current amplitude data, active power data, reactive power data, and switch status data from each node of the power grid in real time from the dispatch automation system, forming a real-time measurement dataset. Export transformer parameter data, line parameter data, circuit breaker parameter data, disconnector switch parameter data, and generator set parameter data in batches from the production management system to form an equipment ledger dataset; The meteorological service center regularly acquires wind speed and direction data, temperature data, humidity data, light intensity data, and rainfall data for the target area to form a meteorological environment dataset. The real-time measurement dataset, the equipment ledger dataset, and the meteorological environment dataset are initially correlated according to the collection time and geographical location to generate an original multi-source dataset containing timestamps and spatial coordinate markers.

3. The method for multi-source data fusion and panoramic visualization governance of power grid dispatching according to claim 1, characterized in that, The steps of performing a spatiotemporal unified reference transformation on the original multi-source dataset to eliminate time asynchrony and coordinate inconsistency between data sources and obtain a spatiotemporally aligned unified data view specifically include: Extract the timestamp information of each data record in the original multi-source dataset. Using the master clock of the scheduling automation system as the reference time source, the network time protocol synchronization algorithm is used to correct the deviation of the timestamps of other data sources. The corrected timestamps are then uniformly converted into Coordinated Universal Time format to obtain the time-aligned multi-source dataset. Geographic coordinate data of all substations, line towers and power plants are extracted from the power grid geographic information system map to establish a benchmark database of the spatial location of power grid entities; For each data point in the time-aligned multi-source dataset, match its corresponding geographic coordinates in the power grid entity spatial location reference library according to its associated device name or number, and append the geographic coordinates to the data record in a unified latitude and longitude coordinate format to obtain the spatially aligned multi-source dataset. The spatially aligned multi-source datasets are resampled and interpolated according to the temporal order and spatial grid to generate a unified spatiotemporally aligned data view that covers the entire scheduling region and has a uniform temporal and spatial resolution.

4. The method for multi-source data fusion and panoramic visualization governance of power grid dispatching according to claim 1, characterized in that, The steps of constructing the topological associations and business logic relationships between power grid entities based on a unified data view and employing knowledge graph technology to obtain a power grid knowledge graph specifically include: Extract all power grid equipment entities from the unified data view, including generator sets, transformers, busbars, circuit breakers, disconnectors, transmission lines, and loads, and assign a unique identifier to each entity. Based on the primary wiring diagram and secondary protection configuration diagram of the power grid, establish the electrical connection relationship between entities, create "connection" type edges between entities with direct electrical connections, and label the attributes of the edges, including voltage level and phase sequence; Based on the dispatching operation procedures and business rules, establish business logic relationships between entities, including the "power supply" relationship between generator sets and transmission lines, the "transformation" relationship between transformers and lower-level busbars, and the "protection" relationship between circuit breakers and protection devices, and create corresponding types of business logic edges; The entities, as well as the "connection" type edges and business logic edges, are stored in a structured manner to form a power grid knowledge graph with entities as nodes and edges as relationships. Each node in the power grid knowledge graph is associated with corresponding real-time measurement data, equipment ledger data, and meteorological environment data.

5. The method for multi-source data fusion and panoramic visualization governance of power grid dispatching according to claim 4, characterized in that, The steps of mapping a unified data view onto a power grid knowledge graph in real time, performing graph computation and reasoning, identifying abnormal power grid operation events and potential risks, and obtaining event and risk labels specifically include: The latest real-time measurement data in the unified data view is updated to the attributes of the corresponding nodes in the power grid knowledge graph in real time, so that the node attributes in the power grid knowledge graph are synchronized with the actual operating status of the power grid. Deploy predefined abnormal event detection rules on the power grid knowledge graph. The abnormal event detection rules include voltage over-limit rules, line overload rules, power reversal rules, and frequency fluctuation rules. When the rule conditions are triggered, mark the node or edge where the abnormal event occurs in the power grid knowledge graph and generate an abnormal event label. A graph neural network model is used to perform real-time reasoning on the power grid knowledge graph. The graph neural network model predicts the trend of the current load rate of each line or transformer in the near future by aggregating the feature information of nodes and neighboring nodes, identifying weak links with overload risk, and generating potential risk labels. Associate the abnormal event labels and potential risk labels with the corresponding nodes and edges in the power grid knowledge graph, and update the label attributes of the power grid knowledge graph.

6. The method for multi-source data fusion and panoramic visualization governance of power grid dispatching according to claim 5, characterized in that, The steps of driving a 3D panoramic visualization engine to dynamically render a power grid operation status map based on event and risk tags, highlighting abnormal areas and risk levels in the power grid operation status map, and generating a panoramic visualization view of power grid dispatch specifically include: Read the real-time attribute data of all nodes and edges in the power grid knowledge graph, as well as the latest event and risk tags, and construct a panoramic data model of the power grid to be rendered; The power grid panoramic data model is input into the 3D panoramic visualization engine, which first loads the pre-generated 3D terrain scene and the 3D model library of power grid equipment. Based on real-time measurement data of nodes and edges, the color and brightness of the device surface in the 3D model are dynamically updated, with green indicating normal operation, yellow indicating warning, and red indicating abnormal operation. For nodes or edges labeled with abnormal events, a flashing red halo is generated around the corresponding 3D model, and the type and specific value of the abnormal event are displayed floating above it. For nodes or edges marked with potential risk labels, a gradient yellow halo is generated around the corresponding 3D model, and the risk level and predicted overload time are overlaid. All rendered 3D scenes and labeled information are combined to output an interactive panoramic visualization view of power grid dispatch, which provides functions such as view rotation, zoom, and information query.

7. The method for multi-source data fusion and panoramic visualization governance of power grid dispatching according to claim 2, characterized in that, Following the step of collecting voltage amplitude data, current amplitude data, active power data, reactive power data, and switch status data from each node of the power grid in real time from the dispatch automation system to form a real-time measurement dataset, the method further includes an online data quality assessment and cleaning step for the real-time measurement dataset, specifically including: A dynamic data quality assessment model based on Kalman filtering is established for each type of measurement data in the real-time measurement dataset. The real-time measurement data collected at the current moment is input into the dynamic data quality assessment model, and the residual between the real-time measurement data and the model prediction value is calculated to obtain the residual sequence. The residual sequence is compared with a preset residual threshold. When the residual value in the residual sequence exceeds the residual threshold three times consecutively, it is determined that the real-time measurement data at the current moment has a data quality problem, and a data quality anomaly mark is generated. For real-time measurement data with a data quality anomaly mark, a data repair program based on spatiotemporal correlation is started. Normal data sequences are extracted from the historical data of the same period of the node, and spatially related data sequences are extracted from the measurement data of the same moment of adjacent nodes. The historical data sequences and spatially related data sequences are input into the Bayesian tensor completion model. Through tensor decomposition and iterative filling algorithms, the repair value for the current abnormal data is calculated. The repaired values ​​replace the original abnormal data, and a repair marker is added to update the real-time measurement dataset.

8. The method for multi-source data fusion and panoramic visualization governance of power grid dispatching according to claim 4, characterized in that, The steps of establishing electrical connection relationships between entities based on the primary wiring diagram and secondary protection configuration diagram of the power grid, creating "connection" type edges between entities with direct electrical connections, and labeling the edge attributes including voltage level and phase sequence, specifically include: The connection relationships between all busbar nodes, circuit breaker nodes, disconnector nodes, transformer winding nodes, and line endpoints are parsed from the primary wiring diagram file of the power grid, and an initial electrical connection relationship table is generated. The relationship between the protection device and the circuit breaker, as well as the sampling connection relationship between the protection device and the current transformer and voltage transformer, are extracted from the power grid secondary protection configuration diagram file to generate a secondary equipment connection relationship table. By jointly analyzing the initial electrical connection table and the secondary equipment connection table, the protection logic unit consisting of circuit breakers, disconnect switches and protection devices is identified, and "protection coordination" type edges are created between entities within the protection logic unit. A graph isomorphism-based topology verification algorithm is used to compare the constructed electrical connection diagram with the templates in the power grid standard topology template library to detect missing or redundant connection edges. Based on the detection results, missing "connection" type edges are automatically added, and redundant "connection" type edges are deleted.

9. The method for multi-source data fusion and panoramic visualization governance of power grid dispatching according to claim 5, characterized in that, The step of using a graph neural network model to perform real-time reasoning on the power grid knowledge graph, where the graph neural network model aggregates the feature information of nodes and their neighbors to predict the changing trend of the current load rate of each line or transformer in the near future, identifies weak links with overload risk, and generates potential risk labels, specifically includes: The real-time measurement data of all nodes in the power grid knowledge graph at time t are concatenated with the equipment ledger data to form the initial feature matrix of the nodes, and the rated capacity of all edges is concatenated with the current power flow value to form the initial feature matrix of the edges. The initial feature matrix of the node and the initial feature matrix of the edge are input into the pre-trained spatiotemporal graph convolutional network model. The spatiotemporal graph convolutional network model first aggregates the first-order neighbor node features and second-order neighbor node features of each node through graph convolutional layers to obtain spatial feature encoding. Spatial feature encoding and similar feature encoding sequences from multiple historical time steps are input into a gated recurrent unit layer. The time dependence of load changes is captured by the reset gate and update gate of the gated recurrent unit layer to obtain a spatiotemporal fusion feature vector. The spatiotemporal fusion feature vector is input into the fully connected output layer. Through the linear transformation and sigmoid activation function of the fully connected output layer, the load rate prediction value of each node at three time points in the next 15 minutes, 30 minutes and 60 minutes is output. The load rate prediction value of each node is compared with the rated capacity threshold of the corresponding device. If the load rate prediction value exceeds the first threshold in the next 15 minutes, it is marked as "emergency overload risk". If the load rate prediction value exceeds the second threshold in the next 30 minutes, it is marked as "short-term overload risk". If the load rate prediction value exceeds the third threshold in the next 60 minutes, it is marked as "medium-term overload risk". A potential risk label containing risk type, risk time and risk level is generated.

10. The method for multi-source data fusion and panoramic visualization governance of power grid dispatching according to claim 6, characterized in that, The step of inputting the power grid panoramic data model into the 3D panoramic visualization engine, wherein the 3D panoramic visualization engine first loads the pre-generated 3D terrain scene and the power grid equipment 3D model library, specifically includes: The geographic coordinate information in the power grid panoramic data model is converted into the world coordinate system position in the three-dimensional scene, and the best observation position and observation angle of the current viewpoint are determined according to the coordinate range. A progressive loading strategy based on viewpoint prediction is adopted. First, the 3D model of the device in the view frustum in the viewpoint direction is loaded. Then, the position of the next viewpoint is predicted according to the viewpoint movement trend, and the 3D terrain scene data of the area corresponding to the predicted position is loaded in advance. When loading the 3D model library of power grid equipment, for substation equipment with voltage levels higher than the preset level, a refined model containing detailed internal structure is used for loading, while for tower and line equipment with voltage levels lower than the preset level, a lightweight model containing texture and basic outline is used for loading. The loaded 3D terrain scene and the 3D model of the power grid equipment are precisely stitched and aligned according to spatial coordinates, and the seams of the models are smoothed to eliminate visual abrupt changes caused by differences in model accuracy.

Images