An Automatic Inspection Method and System for Power Grid Telemetry Information Based on a Large Model

Abstract

This invention application belongs to the field of power system automation technology, specifically relating to an automatic inspection method and system for power grid telemetry information based on a large model. Through five steps—multi-source heterogeneous data acquisition, power grid topology analysis and modeling, time-series data processing and prediction, large-model-based fusion reasoning, and intelligent workflow generation and management—it achieves integrated automatic inspection encompassing data acquisition, topology location, trend prediction, causal reasoning, and work order issuance. It is used for intelligent operation and maintenance, anomaly detection, and decision support in substation, distribution, and transmission systems, transforming passive alarms into proactive prevention, thereby improving the efficiency of power grid operation and maintenance decisions and the reliability of the system.

Description

Technical Field

[0001] This invention application belongs to the field of power system automation technology, specifically relating to an automatic inspection method and system for power grid telemetry information based on a large model. Background Technology

[0002] With the deepening of smart grid construction, the digitalization and informatization levels of substation, distribution, and transmission systems are continuously improving. Systems such as SCADA and DMS can collect massive amounts of telemetry data in real time, including voltage, current, power, and switch status. Meanwhile, topology models such as CIM / E and CIM / XML can comprehensively describe the electrical connections between power grid equipment. Against this backdrop, operation and maintenance personnel face the following prominent challenges in their daily work:

[0003] 1. Large and complex data volume: The amount of real-time telemetry data and topological information is enormous, making it difficult to complete multi-dimensional cross-analysis in a timely and accurate manner manually.

[0004] 2. Difficulty in anomaly identification: A single alarm signal often cannot reflect the overall operating status of the system. For example, a voltage anomaly may be caused by an upstream power supply, line fault, or measurement error. It is difficult to locate the root cause by relying solely on alarms.

[0005] 3. Alarm proliferation and difficulty in prioritizing: In power grid operation, multiple devices often alarm simultaneously in a short period of time. The lack of an effective mechanism to distinguish the primary and secondary relationships and the scope of impact of events can easily lead to important anomalies being covered up or delayed in handling.

[0006] 4. Lack of specificity in handling recommendations: Existing methods rely heavily on human experience and can only provide general suggestions such as "check the line" and "reset the switch", failing to provide precise guidance based on power grid topology, load distribution and operational constraints.

[0007] 5. Lack of prediction and prevention capabilities: Existing technologies mainly provide post-event warnings and in-event processing, which cannot identify potential risks in advance, nor can they analyze trends or preventive control capabilities.

[0008] Existing technologies generally rely on the basic alarm functions of SCADA systems and manual inspection experience, lacking in-depth fusion analysis of power grid topology and real-time time-series data, and even more so lacking the semantic understanding, causal reasoning, and decision support capabilities based on large-scale artificial intelligence models. Therefore, they are insufficient to meet the urgent needs of smart grid operation for proactive early warning, root cause diagnosis, risk classification, and intelligent response recommendations. Summary of the Invention

[0009] This invention aims to address the following technical problems existing in the prior art: the difficulty in deeply integrating and correlating power grid topology information with massive amounts of real-time telemetry data (i.e., "four-remote" data); the low accuracy and lack of interpretability in the identification and root cause localization of abnormal events; the reliance on human experience in alarm handling decisions, making it difficult to scientifically prioritize events; and the general lack of forward-looking early warning capabilities for operational trends, resulting in incomplete and ineffective work order information. This invention provides an automatic inspection method and system for power grid telemetry information based on a large model.

[0010] The technical solution for achieving the objective of this invention is as follows:

[0011] The first aspect of this invention is to provide an automatic inspection method for power grid telemetry information based on a large model, comprising the following steps:

[0012] Multi-source heterogeneous data acquisition steps: Acquire real-time telemetry data, switch events, alarm information, and contextual data including weather information and maintenance plans;

[0013] Power grid topology parsing and modeling steps: Parse the CIM / E format power grid topology file and convert it into a structured graph data model to reflect the electrical connection relationships between devices;

[0014] Time series data processing and prediction steps: The real-time telemetry data is cleaned, denoised, and feature extracted, and a time series prediction model is used to predict the future trends of key variables;

[0015] The fusion reasoning steps based on the large model are as follows: the power grid topology diagram, real-time time series data, prediction trend results and event context information are fused and provided as input to the large language model; the large language model performs semantic understanding, causal reasoning and anomaly detection, and outputs structured analysis results including root cause analysis, handling suggestions and impact scope;

[0016] Intelligent workflow generation and management steps: Based on the output of the large model, intelligent work orders containing event summaries, root cause analysis, handling suggestions, priority scores, and scope of impact are automatically generated, and a manual review and confirmation interface is provided. Confirmed work orders are converted into formal dispatch orders, operation tickets, or work maintenance tickets to achieve closed-loop management of decision-making and execution.

[0017] Furthermore, in the power grid topology parsing and modeling step, parsing the CIM / E format power grid topology file specifically includes: dividing the file content into blocks according to the file's structured tags; extracting metadata from the file; parsing data records line by line, with each line representing the attribute value of an object; and outputting the parsing results in a preset structured data format.

[0018] Furthermore, the construction of the power grid topology map includes: extracting node identifiers from the parsed structured data and defining them as electrical connection points, i.e., nodes, in the power grid topology map; defining conductive device entities containing at least two node identifiers as edges connecting nodes; automatically generating virtual nodes for devices containing only one node identifier to construct complete device end edges; and storing the electrical attributes of the devices and their relationships as attribute information of the nodes.

[0019] Furthermore, the time-series prediction model employs a Transformer or TCN architecture to predict key telemetry trends for the next 1-24 hours.

[0020] Furthermore, the input to the large model-based fusion inference step is a structured JSON object containing topological snapshots, telemetry summaries, event information, and contextual data; the output is a structured result containing root causes, treatment recommendations, scope of impact, and confidence levels.

[0021] Furthermore, in the intelligent workflow generation and management steps, the priority score is calculated by a weighted algorithm that comprehensively considers the importance of the equipment, the scope of impact, the severity of the anomaly, the predictability of the duration, and the time urgency.

[0022] A second aspect of the present invention is to provide an automatic inspection system for power grid telemetry information based on a large model, comprising the following modules:

[0023] The data acquisition module is used to collect real-time telemetry data, switch events, alarm information, and context data.

[0024] The CIM / E parsing module is used to parse CIM / E format power grid topology files and output structured data;

[0025] The topology building module is used to convert structured data into a power grid topology graph represented by nodes and edges;

[0026] The graph database module is used to store and manage the power grid topology graph and provides a graph algorithm interface;

[0027] The time-series data processing module is used to clean, denoise, and extract features from telemetry data.

[0028] The time series forecasting module is used to predict future trends based on processed time series data;

[0029] The large model inference module is used to integrate topology graphs, time series data, prediction results and contextual information into the large language model, and output root cause analysis, treatment suggestions and impact scope.

[0030] An intelligent work order generation module, which is used to automatically generate intelligent work orders containing event summaries, root cause analyses, disposal suggestions, priority scores, and impact scopes according to the output of a large model;

[0031] A work order review and conversion module, which is used to manually review intelligent work orders and convert them into formal tickets to achieve closed-loop management.

[0032] The implementation method of the data acquisition module is to collect real-time telemetry data, switch events, alarm information, and context data (including weather information and maintenance plans, etc.), and output the standardized data to the CIM / E parsing module and the time series data processing module to provide basic information for topology parsing and time series analysis.

[0033] Furthermore, the input of the large model inference module is a structured JSON object, including a topology snapshot, a telemetry summary, event information, and context data; the output is a structured analysis result for direct invocation by the intelligent work order generation module.

[0034] Furthermore, the intelligent work order generation module has a built-in priority scoring algorithm, which comprehensively calculates the priority score based on the importance of the equipment, the impact scope, the severity of the abnormality, the predicted persistence, and the time urgency.

[0035] Furthermore, the work order review and conversion module provides a human-computer interaction interface, supports dispatchers to correct and confirm intelligent work orders, and converts them into dispatch orders, operation tickets, or work maintenance tickets at one key. At the same time, it tracks the execution status to form a closed loop.

[0036] Furthermore, the CIM / E parsing module analyzes the classes, objects, and their attribute information in the CIM / E file, extracts the device parameters and connection relationships required for topology, and maps them into a preset structured data format. Specifically, it analyzes the input CIM / E format power grid topology file, divides it into blocks based on <class name::entity name> and < / class name::entity name>, extracts the header meta information and attribute values, and converts the result into a structured data in JSON or CSV format, and outputs it to the topology construction module and the graph database module.

[0037] Furthermore, the topology construction module constructs a complete power grid topology map by mapping the connection points (ConnectivityNode) in the CIM model to graph nodes, associating and mapping the conducting equipment (ConductingEquipment) through its terminals (Terminal), and supplementing virtual nodes for single-ended equipment to ensure topology integrity, and outputs it to the graph database module and the large model inference module.

[0038] Furthermore, the graph database module is implemented by storing and managing the power grid topology graph generated by the topology construction module, and providing graph algorithm calculation and query interfaces to provide node connectivity, shortest path and impact range analysis capabilities for the large model inference module and the intelligent work order generation module.

[0039] Furthermore, the implementation method of the time series data processing module is to clean, denoise and extract features from the real-time telemetry data output by the data acquisition module, and generate time series summary information as input to the time series prediction module and the large model inference module.

[0040] Furthermore, the implementation method of the time series prediction module is based on the features generated by the time series data processing module, using Transformer or TCN models to predict trend information for the next 1-24 hours, and transmitting the prediction results to the large model inference module for trend warning and priority scoring.

[0041] Furthermore, the implementation method of the large model inference module is to take the topology graph generated by the topology construction module, the summary of the time series data processing module, the trend results of the time series prediction module, and the event context information as input, perform semantic understanding and anomaly detection, and output root cause analysis, handling suggestions and impact range for use by the intelligent work order generation module.

[0042] Furthermore, the implementation method of the intelligent work order generation module is to receive the output information of the large model inference module, generate an intelligent work order containing an event summary, root cause analysis, handling suggestions, priority score, scope of impact and required resources, and send the work order to the work order review and conversion module.

[0043] Furthermore, the work order review and conversion module is implemented by manually reviewing and confirming intelligent work orders, and converting the confirmed work orders into formal documents (dispatch orders, operation tickets, or work / maintenance tickets). Simultaneously, it tracks the execution status and effects of work orders, forming a closed-loop management system. Furthermore, the operation result display module can summarize and display statistical data including all information of executed workflows, the number of anomalies, the number of generated strategies, and the number of reports, presenting the operation results of the power system's autonomous patrol in an intuitive format.

[0044] Advantages and beneficial effects of the present invention:

[0045] 1. This invention achieves automated deep analysis of massive and complex data by deeply integrating power grid topology diagrams, real-time time-series data and contextual information, and by utilizing the causal reasoning capabilities of large models. This solves the technical problem that traditional methods struggle to penetrate surface alarms and accurately locate the root cause of faults.

[0046] 2. This invention enables the analysis of the impact range of abnormal events on the topology graph, and combines multiple factors such as equipment importance and anomaly severity for quantitative scoring, providing a scientific and dynamic priority ranking mechanism that effectively solves the problem of important events being overwhelmed or delayed in handling when alarms are rampant.

[0047] 3. This invention enables a comprehensive understanding of the current power grid topology, load distribution, and operational constraints using a large model, and achieves precise and intelligent generation of disposal suggestions.

[0048] 4. By introducing a time-series prediction module to analyze the future trends of key telemetry data, this invention enables forward-looking early warning of potential operational risks, transforming operation and maintenance work from the traditional "passive response" to "proactive prevention," and providing decision support for avoiding faults and ensuring the safe and stable operation of the power grid.

[0049] 5. The functional modules of this invention are clear and the logic is closed-loop. It has been successfully applied in the field. Attached Figure Description

[0050] Figure 1 This is a structural block diagram of the present invention.

[0051] Figure 2 This is a flowchart of the process when a specific workflow is executed. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below through specific embodiments. The following embodiments are merely descriptive and not limiting, and should not be used to limit the scope of protection of this invention.

[0053] An automated inspection and early warning workflow for power grid telemetry information based on a large model aims to address the following technical problems in existing technologies: the difficulty in deeply integrating and correlating power grid topology information with massive amounts of real-time telemetry data (i.e., "four remote" data); the low accuracy and lack of interpretability in the identification and root cause location of abnormal events; the reliance on human experience in alarm handling decisions, making it difficult to scientifically prioritize events; and the general lack of forward-looking early warning capabilities for operational trends, resulting in incomplete and ineffective work order information.

[0054] The present invention proposes the following modules: The present invention provides an automatic inspection and early warning system for power grid telemetry information based on a large model, which includes the following modules: data acquisition module 1, CIM / E parsing module 2, topology construction module 3, graph database module 4, time series data processing module 5, time series prediction module 6, large model inference module 7, intelligent work order generation module 8, and work order review and conversion module 9.

[0055] In a specific embodiment of the present invention, the specific technical implementation methods of the various modules described in the Summary of the Invention section are described:

[0056] 1. Implementation of the CIM / E parsing module 2 and the topology construction module 3

[0057] CIM / E file parsing process: When the CIM / E parsing module 2 performs parsing, it first reads the power grid model file in CIM / E format and processes the file content in chunks based on the <class name::entity name> and < / class name::entity name> tags. Subsequently, the module extracts the meta-information in the file, such as the header information starting with @, the attribute type definition starting with %, and the dimension information starting with. Finally, the module parses the data lines starting with # line by line, where each line represents the attribute values of an object, and finally outputs the parsing result as a preset structured data format such as JSON or CSV.

[0058] Topological abstraction modeling: After receiving the above-mentioned structured data, the topology construction module 3 performs topological modeling. Specifically, the connection point (Connectivity Node) objects in the parsed CIM model are defined as electrical connection points (i.e., nodes) in the power grid topology diagram, and the conducting equipment (Conducting Equipment) entities (such as circuit breakers, lines, transformers, etc.) are associated with the corresponding connection points through their terminals (Terminal), thereby defining as the edges connecting the nodes. For devices that only contain single-end nodes (such as generators and loads), the module will automatically generate a virtual node to construct a complete device end edge. At the same time, the electrical attributes of the device and its affiliation relationship (such as the affiliated substation and voltage level) are stored as rich information attributes of the node.

[0059] 2. Implementation of the large model inference module 7 and the intelligent work order generation module 8

[0060] Input structure of the large model: In order to enable the large language model to accurately understand the instantaneous state of the power grid, the large model inference module 7 integrates the multi-dimensional information collected from other modules into a structured input object, such as a JSON object. This object clearly defines the topological snapshot, telemetry summary, historical fault records, and context information.

[0061] Topological parameters: Define the node set (including node unique identification code, type, name, voltage level) and the connection relationship set (including start node ID, target node ID, switch device type, real-time closing and opening status).

[0062] Telemetry parameters: Real-time record the per-unit value of voltage, effective value of current, active power of each node, and the precise sampling moment.

[0063] Event parameters: include the type of alarm currently triggered (such as protection action), the associated affected devices, the alarm severity level, and a detailed text description.

[0064] Contextual parameters: real-time weather conditions (such as high temperature, rainfall), overall system load level, and current equipment maintenance plan status.

[0065] The structure and generation of intelligent work orders: After analyzing the above inputs, the large model inference module 7 outputs structured analysis results. The intelligent work order generation module 8 receives these results and automatically generates an intelligent work order with complete information elements.

[0066] Diagnostic elements include the work order serial number, core event summary, root cause diagnosis conclusion (such as load surge or equipment failure), and the confidence score of that conclusion.

[0067] Impact factors: Clearly list the directly affected equipment, the indirectly affected topology, and the number of electricity customers expected to be affected.

[0068] Decision-making elements include automatically calculated priority scores (derived from a weighted algorithm), urgency levels, and targeted on-site handling recommendations (such as load transfer paths and key inspection points).

[0069] Closed-loop elements: List the suggested monitoring and measurement points, estimated repair time, required spare parts and personnel configuration requirements.

[0070] Priority scoring algorithm: When generating a smart work order, the "Priority_Score" field can be calculated using a weighted scoring algorithm. This algorithm is executed by the smart work order generation module 8 and comprehensively considers the following factors: equipment importance weight, impact range coefficient, anomaly severity, prediction persistence provided by the time series prediction module 6, and time urgency. Its calculation formula can be set as follows:

[0071] Priority_Score = Σ(Factor Weight × Factor Score) × 10

[0072] The calculation factors include: Equipment importance: weights are set based on the voltage level and load characteristics of the equipment in the power grid hierarchy; Impact range: the number of nodes affected by the power outage calculated by the graph database through a graph traversal algorithm; Anomaly severity: the proportion of real-time telemetry values ​​deviating from the rated values; Prediction persistence: the duration of future trends output by the time series prediction module.

[0073] 3. Implementation of Time Series Prediction Module 6

[0074] Feature engineering: Before making predictions, the module first performs feature engineering on the input time series data. The features constructed may include: historical telemetry value series, moving mean and moving standard deviation of the series, lag features (such as the value at the same time the previous day), and statistical features of topological neighbor nodes obtained from graph database module 4 (such as the average voltage of neighbor nodes).

[0075] Prediction model architecture: Different model architectures can be adopted to meet different prediction needs for different durations. For example, for short-term predictions of 1-4 hours, a Temporal Convolutional Network (TCN) or Transformer model can be used; for medium-term predictions of 4-24 hours, a Transformer model with a sequence-to-sequence (Seq2Seq) structure can be used.

[0076] 4. Implementation of Module 9 for Work Order Approval and Conversion

[0077] Intelligent Work Order Positioning and Process: The intelligent work orders generated by this invention serve as a basis for auxiliary decision-making and internal work order dispatch within the system. The work order review and conversion module 9 manages its lifecycle, and this process includes:

[0078] Automatic generation: The work order is automatically generated by the intelligent work order generation module 8 and then submitted to this module.

[0079] Manual review: The dispatcher or operation and maintenance manager reviews the work order content (especially root cause analysis and handling suggestions) on the human-computer interaction interface provided in this module.

[0080] Decision Confirmation: The reviewers make final confirmation of the handling plan and priorities.

[0081] Invoice Conversion: After confirmation, the work order review and conversion module 9 provides an interface that can convert the key information of the smart work order into a formal invoice in the enterprise's existing production management system with one click.

[0082] Execution tracking: The work order review and conversion module 9 can continuously track the execution status and effect of formal invoices, forming a closed loop.

[0083] Formal ticket types: The formal tickets may include, but are not limited to: dispatch orders for adjusting the power grid operation mode, operation tickets for equipment operation, and work maintenance tickets for equipment repair and maintenance.

[0084] End-to-end integrated implementation: intelligent diagnosis and closed-loop handling of 10kV distribution network line overload tripping.

[0085] The following example of a complete scenario illustrates how the modules of this invention work together.

[0086] Step 1: Event triggering and multi-source data collection.

[0087] Reference Figure 1 In this step, data acquisition module 1, through its interface with the SCADA system, monitors and captures in real time the tripping event of the "10kV commercial line K01 switch" and the associated "overcurrent stage II protection action" alarm information. Data acquisition module 1 then collects a snapshot of telemetry data for 5 minutes before and after the event, centered on the moment of the event. For example, it collects data showing that the current of switch K01 suddenly increased from 80A to 650A before the trip, and that the upstream bus voltage slightly dropped from 10.3kV to 10.25kV. Simultaneously, data acquisition module 1 also collects contextual information such as the current weather as "summer high temperature (38℃)".

[0088] Step 2: Topology localization and temporal preprocessing.

[0089] In this step, the system first locates the "K01 switch" and its electrical connections from the power grid topology model pre-constructed and stored in the graph database module 4 by the CIM / E parsing module 2 and the topology construction module 3. Subsequently, the time-series data processing module 5 cleans and summarizes the telemetry data collected in step 1. Simultaneously, the time-series forecasting module 6, based on historical load data and current high-temperature weather characteristics, uses a pre-trained Transformer model to predict that the load in the area will remain high for the next two hours.

[0090] Step 3: Large-scale deep reasoning and root cause diagnosis.

[0091] Reference Figure 1 In this step, the large model inference module 7 integrates all the information obtained in steps 1 and 2, including events, telemetry data, topological relationships, prediction conclusions, and context, into a structured input object, which is then provided to the large language model for inference. The input object contains specific business numerical values:

[0092] Telemetry change values: It was clearly recorded that the current of switch K01 suddenly increased from 80A to 650A at the moment of tripping, while the upstream bus voltage only fluctuated slightly from 10.3kV to 10.25kV.

[0093] Specific alarm: The structured description of "overcurrent stage II protection action" has been triggered.

[0094] Prediction conclusion: The time series forecast module gives the judgment that "the load in this area will continue to remain high in the next 2 hours".

[0095] Environmental values: The current ambient temperature is recorded as 38℃, which is a summer high temperature and high load scenario.

[0096] The large model reasoning module 7 performs comprehensive reasoning analysis based on the above inputs. For example, it can analyze and conclude that since the voltage drop of the upstream bus is extremely small (<0.5%), it does not meet the characteristics of a short-circuit fault. Combining the current growth process and the high temperature and high load context, it concludes that the root cause of the event is "downstream load overload" and gives a confidence level of 0.95.

[0097] Step 4: Intelligent work order generation and priority assessment.

[0098] In this step, the intelligent work order generation module 8 receives the structured inference results from the large model inference module 7 and automatically generates an intelligent work order. The priority scoring algorithm built into the intelligent work order generation module 8 calculates a high priority score, for example, 9.2, based on factors such as the scope of impact, event severity, and the predictive persistence provided by the time series prediction module 6. The final generated intelligent work order can be a JSON object containing multiple fields such as event summary, root cause analysis, and handling suggestions. An example is shown below:

[0099] Qualitative diagnostic conclusion: The dominant cause is determined to be "line overload caused by a sudden increase in the summer cooling load of the downstream large commercial center Z", and a high confidence level of 0.95 is given.

[0100] Priority quantification conclusion: Based on the scope of impact and predicted duration, a high priority score of 9.2 is calculated.

[0101] Operational instructions: The dispatcher is explicitly required to immediately verify the internal load of commercial center Z, and it is emphasized that power restoration of switch K01 is strictly prohibited before the downstream load is reduced.

[0102] Step 5: Manual review and closed-loop execution.

[0103] Reference Figure 1 In this step, the work order review and conversion module 9 pushes the intelligent work order generated in step 4 to the dispatcher's human-machine interface as a high-priority alarm. After the dispatcher reviews and confirms the work order, the key information of the intelligent work order is converted into a formal "operation ticket" through the interface provided by the work order review and conversion module 9 and then issued to the maintenance team. The work order review and conversion module 9 can also continuously track the execution status of the operation ticket and archive the final execution result, thereby realizing end-to-end closed-loop management from intelligent fault diagnosis to on-site handling.

[0104] Figure 2 The flowchart illustrates the complete execution process of the automatic inspection method for power grid telemetry information based on a large model. The specific steps are as follows:

[0105] S1: Collect multi-source heterogeneous data;

[0106] S2a: Parse CIM / E and construct a topology map; S2b: Process time series data and make predictions (executed in parallel).

[0107] S3: Integrating topology / time series / events / context;

[0108] S4: Large models perform semantic understanding and reasoning;

[0109] S5: Generate smart work orders for root cause / treatment suggestions;

[0110] S6: Submit to the manual review interface;

[0111] S7: Has the review been confirmed?

[0112] If the audit has been confirmed, proceed to S8: convert to a formal invoice and execute;

[0113] S9: Track execution status and effects, then end.

[0114] If the review is not confirmed, proceed to S10: Archive or mark for further analysis, and end.

[0115] This flowchart illustrates the complete automated inspection workflow from data collection and intelligent analysis to manual review and closed-loop execution, realizing integrated management of data collection, topology positioning, trend prediction, causal reasoning, and work order issuance.

[0116] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention.

Claims

1. An automatic inspection method for power grid telemetry information based on a large model, characterized in that, Includes the following steps: Multi-source heterogeneous data acquisition steps: Acquire real-time telemetry data, switch events, alarm information, and contextual data including weather information and maintenance plans; Power grid topology parsing and modeling steps: Parse the CIM / E format power grid topology file and convert it into a structured graph data model to reflect the electrical connection relationships between devices; Time series data processing and prediction steps: The real-time telemetry data is cleaned, denoised, and feature extracted, and a time series prediction model is used to predict the future trends of key variables; The fusion inference steps based on the large model are as follows: fusion of power grid topology diagram, real-time time series data, prediction trend results and event context information to construct a structured input for large model inference; The large model performs semantic understanding, causal reasoning, and anomaly detection, and outputs structured analysis results including root cause analysis, treatment suggestions, and scope of impact. Intelligent workflow generation and management steps: Based on the output of the large model, an intelligent work order containing event summary, root cause analysis, handling suggestions, priority score and scope of impact is automatically generated, and an interface for manual review and confirmation is provided to convert the confirmed work order into a formal dispatch order, operation ticket or work maintenance ticket. In the power grid topology parsing and modeling steps, parsing the CIM / E format power grid topology file specifically includes: dividing the file content into blocks according to the file's structured tags; extracting metadata from the file; parsing data records line by line, with each line representing the attribute value of an object; and outputting the parsing results in a preset structured data format. The construction of the power grid topology map includes: extracting node identifiers from the parsed structured data and defining them as electrical connection points, i.e., nodes, in the power grid topology map; defining conductive device entities containing at least two node identifiers as edges connecting nodes; automatically generating virtual nodes for devices containing only one node identifier to construct complete device end edges; and storing the electrical attributes of the devices and their relationships as attribute information of the nodes.

2. The automatic inspection method for power grid telemetry information based on a large model according to claim 1, characterized in that, The time-series prediction model uses a Transformer or TCN architecture to predict key telemetry trends for the next 1-24 hours.

3. The automatic inspection method for power grid telemetry information based on a large model according to claim 1, characterized in that, The input to the large model-based fusion inference step is a structured JSON object containing topological snapshots, telemetry summaries, event information, and contextual data; the output is a structured result containing root causes, treatment recommendations, scope of impact, and confidence levels.

4. The automatic inspection method for power grid telemetry information based on a large model according to claim 1, characterized in that, In the intelligent workflow generation and management steps, the priority score is calculated by a weighted algorithm that considers the importance of the equipment, the scope of impact, the severity of the anomaly, the predictability of the duration, and the time urgency.

5. An automatic inspection system for power grid telemetry information based on a large model, according to any one of claims 1-4, characterized in that, include: The data acquisition module (1) is used to collect real-time telemetry data, switch events, alarm information and context data; The CIM / E parsing module (2) is used to parse CIM / E format power grid topology files and output structured data; Topology building module (3) is used to convert structured data into a power grid topology graph represented by nodes and edges; The graph database module (4) is used to store and manage the power grid topology graph and provide a graph algorithm interface; The time series data processing module (5) is used to clean, denoise and extract features from telemetry data; The time series prediction module (6) is used to predict future trends based on the processed time series data; The large model reasoning module (7) is used to integrate the topology diagram, time series data, prediction results and context information into the large language model and output the root cause analysis, treatment suggestions and impact range. The intelligent work order generation module (8) is used to automatically generate intelligent work orders containing event summaries, root cause analysis, handling suggestions, priority scores and impact scope based on the output of the large model; The work order review and conversion module (9) is used to manually review smart work orders and convert them into formal invoices; The data acquisition module (1), CIM / E parsing module (2), topology construction module (3), graph database module (4), time series data processing module (5), time series prediction module (6), large model inference module (7), intelligent work order generation module (8), and work order review and conversion module (9) are sequentially connected or interact with each other.

6. The automatic inspection system for power grid telemetry information based on a large model according to claim 5, characterized in that, The input of the large model inference module (7) is a structured JSON object, which includes topological snapshots, telemetry summaries, event information and context data; the output is a structured analysis result, which can be directly called by the intelligent work order generation module (8).

7. The automatic inspection system for power grid telemetry information based on a large model according to claim 5, characterized in that, The intelligent work order generation module (8) has a built-in priority scoring algorithm that calculates the priority score by comprehensively considering the importance of the equipment, the scope of impact, the severity of the anomaly, the predictability of the duration, and the time urgency.

8. The automatic inspection system for power grid telemetry information based on a large model according to claim 5, characterized in that, The work order review and conversion module (9) provides a human-computer interaction interface, which supports the dispatcher to correct and confirm the intelligent work order, and convert it into a dispatch order, operation ticket or work maintenance ticket with one click, while tracking the execution status to form a closed loop.

Images