Agent-based automatic power grid defect identification system

Abstract

The invention relates to the technical field of power grid defect recognition, in particular to an Agent-based automatic power grid defect recognition system. The system comprises the following steps: acquiring real-time acquisition data, analyzing the real-time acquisition data to obtain a data analysis result, and performing tagging processing on the real-time acquisition data according to the data analysis result to obtain tag data; based on a D-S evidence theory, according to a preset historical defect graph, performing classification processing on the label data; and performing defect identification on the classified label data by utilizing a defect identification Agent, generating a defect report according to an identification result, and feeding back the defect report to an equipment maintenance system. And the classified label data directly triggers a defect identification algorithm of the corresponding Agent, so that a communication link of traditional cross-system data calling is shortened, and the defect upgrading risk caused by processing delay can be reduced. A structured report is automatically generated, a traceable decision basis is provided for an equipment maintenance system, and artificial transcription errors are reduced.

Description

Technical Field

[0001] This application relates to the technical field of power grid defect identification, and in particular, to an automated power grid defect identification system based on Agent. Background Art

[0002] As the construction of the smart grid enters the digital transformation stage, the power equipment monitoring system has formed a multi-dimensional perception network. The existing monitoring system generates PB-level structured monitoring data and unstructured operation and maintenance records every day, covering hundreds of equipment state parameters such as voltage fluctuations, insulation degradation, and mechanical displacement. The traditional defect identification method mainly relies on the combination of threshold alarm and manual inspection, which exposes significant limitations in the face of the complexity of the new power system.

[0003] With the improvement of the intelligence level of smart grid equipment, the defect handling architecture based on the multi-Agent system has become the mainstream technical direction in the industry. However, industry test data shows (see the IEEE PES 2024 white paper) that there are significant shortcomings in the key performance indicators of the existing Agent system: the collaborative error rate between defect identification and Agent matching is as high as 31%, the resource conflict rate during multi-Agent concurrent disposal reaches 27%, and the communication delay of cross-regional Agent collaboration exceeds 800 ms. These problems lead to about 15% of repeated disposal work orders generated every year, and 17% of secondary defects evolve into system failures due to delayed disposal. Therefore, how to accurately identify defects remains an urgent problem to be solved at present. Summary of the Invention

[0004] This application provides an automated power grid defect identification system based on Agent to solve the above problems.

[0005] In a first aspect, this application provides an automated power grid defect identification system based on Agent, and the system includes: Acquire real-time collected data, analyze the real-time collected data to obtain a data analysis result, and perform tagging processing on the real-time collected data according to the data analysis result to obtain tag data; Based on the D-S evidence theory, classify the tag data according to a preset historical defect map; Use a defect identification Agent to identify defects in the classified tag data, and generate a defect report according to the identification result and feedback it to the equipment maintenance system.

[0006] Through this solution, the dynamic analysis and tagging of real-time collected data are carried out to form structured feature vectors, providing standard data input for subsequent classification and enhancing the compatibility of heterogeneous data sources. Based on the fusion decision-making mechanism of D-S evidence theory and historical defect maps, the misjudgment probability caused by traditional single-threshold determination can be reduced, and the ability to associate and identify compound defects can be enhanced. By preprocessing data classification and the directional matching of dedicated defect recognition Agents, the probability of task overlap during multi-Agent concurrency is reduced, and the conflict risk caused by resource competition is lowered. The classified tag data directly triggers the defect recognition algorithm of the corresponding Agent, shortening the communication link of traditional cross-system data calls and facilitating the reduction of the defect escalation risk caused by disposal delays. A structured report containing classification tags, evidence weights, defect locations, etc. is automatically generated, providing a traceable decision-making basis for the equipment maintenance system and reducing manual transcription errors.

[0007] Optionally, the analysis of the real-time collected data to obtain the data analysis result includes: Parse the real-time collected data to determine the collected image; Analyze the collected image to determine a number of pixel features; Analyze the number of pixel features to determine the equipment contour and spatial distribution boundary; Determine the equipment contour and the spatial distribution boundary as the data analysis result.

[0008] Through this solution, image data is accurately separated from multi-source real-time collected data (such as video streams, sensor mixed data streams), and non-image interference information is excluded. Key pixel features in the image are extracted, and visual information is converted into numerical features and geometric parameters, making the image information quantifiable and eliminating the dependence on manual experience. The contour and boundary data are encapsulated into a standardized format (such as a coordinate sequence) and directly serve the defect classification and disposal process.

[0009] Optionally, the tagging of the real-time collected data according to the data analysis result to obtain tag data includes: Determine a number of independent equipment entities in the data according to the equipment contour; Obtain the power grid GIS database, and determine the physical connection relationship between independent equipment entities according to the power grid GIS database; Determine independent regions according to the spatial distribution boundary; Perform tagging on the real-time collected data of a number of independent equipment entities according to the independent regions and the physical connection relationship to obtain tag data.

[0010] Through this solution, by matching the device outline with the GIS library, it is ensured that there is a one-to-one correspondence between the data and the physical devices, eliminating incorrect device labels. The physical connection relationship labels enable the Agent to understand the dependency relationships between devices, reducing the collaborative error rate caused by topological misunderstandings. The independent area division restricts the global competition of multiple Agents, reducing the resource conflict rate. The structured label data directly serves the subsequent D-S evidence theory classification and Agent decision-making, shortening the defect identification and handling link.

[0011] Optionally, based on the D-S evidence theory, according to a preset historical defect map, the label data is classified, including: Analyze the label data to determine whether any real-time acquisition data belongs to multi-label data; If so, according to the D-S evidence theory and the preset historical defect map, conflict resolution is performed on the multi-label data to obtain normal label data; Analyze the normal label data to determine the label attributes, and classify the normal label data according to the label attributes.

[0012] Through this solution, by synthesizing the confidence levels of multi-labels based on the D-S evidence theory, redundant or contradictory labels are eliminated, preventing the Agent from misjudging the defect type due to data conflicts and reducing collaborative errors. Classifying the data according to label attributes (such as "insulation type", "mechanical type") makes the classification results more in line with the actual scenario. Automated conflict resolution and classification replace traditional manual review, shortening the handling response time.

[0013] Optionally, the real-time acquisition data of several independent device entities is labeled according to the independent area and the physical connection relationship to obtain label data, including: Based on the physical connection relationship, analyze the device distribution in the independent area to determine the associated devices belonging to the same electrical circuit in the independent area; And merge the environmental labels of the associated devices; After determining the associated devices, determine whether there are isolated devices according to the device spatial distribution in the independent area; If there are isolated devices, generate device-level labels separately; Obtain label data according to the merge result and the device-level labels.

[0014] Through the solution provided in this embodiment, the device sets of the same electrical circuit are identified through physical connection relationships, and the scattered device data is aggregated into circuit-level tags, avoiding the logical fragmentation caused by the isolated analysis of single-device data. The need to repeatedly label the same environmental parameters (such as circuit temperature and humidity) for multiple devices within the same circuit is reduced, compressing the magnitude of the original data. Independent devices that do not form a closed circuit are detected through the spatial distribution of devices, preventing their data from being misclassified or omitted. Exclusive device-level tags are generated for isolated devices to ensure that their independent operating states are completely recorded and traced. Through tag merging and classification, petabyte-scale original data is compressed into high-value tag data, reducing the data magnitude and computing load processed by the Agent.

[0015] Optionally, determining the physical connection relationship between independent device entities according to the power grid GIS database includes: Parsing the power grid GIS database to determine the device ID and device coordinate information of the independent devices; Based on the device ID, quickly matching through a spatial indexing algorithm to obtain the adjacent devices of the independent devices; According to the device coordinate information, determining the device spacing between adjacent devices and the independent devices; Comparing the device spacing with a preset interval threshold; If the device spacing is less than the preset interval threshold, it is determined that there is a physical connection relationship.

[0016] Through this solution, by parsing the power grid GIS database, the mapping relationship between device IDs and coordinate information is extracted, ensuring that the geographical location of each independent device is unique and traceable, providing basic data support for subsequent physical connection analysis. Through the spatial indexing algorithm, the list of neighboring devices of the target device is quickly screened out, reducing the time complexity of neighboring device queries and adapting to the efficient processing requirements of petabyte-scale power grid data. Calculating the actual spacing based on device coordinates and combining with the preset interval threshold to exclude invalid associations of distant devices, ensuring that the connection relationship conforms to actual engineering constraints. Clearly defining the electrical circuit attribution among device groups, providing a topological basis for subsequent defect identification and task assignment in the Agent system.

[0017] Optionally, before the step of determining that there is a physical connection relationship if the device spacing is less than the preset interval threshold, it further includes: Obtaining the device information of the adjacent devices; Determining the voltage level according to the device information; Comparing the voltage level with the voltage level of the independent device; If the device spacing is less than the preset interval threshold and the voltage levels are the same as that of the independent device, it is determined that there is a physical connection relationship.

[0018] Through this solution, voltage level matching is introduced to ensure that the physical connection relationship complies with the power grid safety specifications, avoiding the one-sidedness of relying solely on spatial distance. Double-condition filtering reduces the misjudgment of invalid connection relationships.

[0019] Optionally, the conflict resolution of the multi-label data according to the D-S evidence theory and the preset historical defect map to obtain normal label data includes: Determine the label confidence of the multi-label data according to the preset historical defect map; Analyze several labels of the multi-label data to determine whether there are any two mutually exclusive label attributes; If so, based on the D-S evidence theory, according to the label confidence, resolve the conflict between the two labels with mutually exclusive attributes to obtain normal label data.

[0020] Through this solution, by detecting and eliminating mutually exclusive labels, it is ensured that the label data of the same device does not conflict at the physical or logical level, avoiding misjudgment by the defect recognition Agent due to label contradictions, thereby reducing the collaborative error rate. Based on the label confidence of the historical defect map, it reduces the repeated handling work orders caused by low-quality labels, improves the accuracy of defect reports, and inhibits secondary defects from evolving into system failures. It alleviates the interference of label noise in PB-level data to the system and enhances the robustness of defect recognition under the new power system (high complexity, multi-parameter coupling).

[0021] Optionally, the conflict resolution of the two labels with mutually exclusive attributes according to the D-S evidence theory and the label confidence to obtain normal label data includes: Sort the label confidences of the two labels with mutually exclusive attributes, determine the label with low confidence according to the sorting result, and based on the D-S evidence theory, resolve the label with low confidence to obtain normal label data.

[0022] Through this solution, the conflict is resolved directly by confidence sorting, avoiding complex calculations and forcibly eliminating mutually exclusive labels to prevent the system from falling into logical chaos due to contradictory data.

[0023] Optionally, the determination of the label confidence of the multi-label data according to the preset historical defect map includes: Determine the data source according to the real-time collected data; Determine the confidence weight of the multi-label data according to the data source; Determine the label confidence of the multi-label data according to the preset historical defect map and the confidence weight.

[0024] Through this solution, based on the objective statistics of the preset historical defect map and real-time data sources, subjective interference from manual experience is avoided. The confidence calculation process is transparent (historical frequency × weight), facilitating maintenance personnel to trace the decision-making basis. Description of the Drawings

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0026] Figure 1 A schematic diagram of an application scenario provided by an embodiment of the present application; Figure 2 A flowchart of an automated power grid defect identification system based on an Agent provided by an embodiment of the present application. Detailed Embodiments

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts fall within the scope of protection of the present application.

[0028] In addition, the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this article generally represents an "or" relationship between the associated objects before and after, unless otherwise specified.

[0029] The following will further describe the embodiments of the present application in detail with reference to the drawings in the specification.

[0030] With the improvement of the intelligence level of smart grid devices, the defect handling architecture based on multi-Agent systems has become the mainstream technical direction in the industry. However, industry test data shows (see the IEEE PES 2024 white paper) that there are significant shortcomings in the existing Agent systems in terms of key performance indicators: the collaborative error rate between defect identification and Agent matching is as high as 31%, the resource conflict rate during multi-Agent concurrent handling reaches 27%, and the communication delay of cross-regional Agent collaboration exceeds 800 ms. These problems result in approximately 15% of repeated handling work orders being generated each year, and 17% of secondary defects evolving into system failures due to handling delays. Therefore, how to accurately identify defects remains an urgent problem to be solved currently.

[0031] Based on this, the present application provides an Agent-based automated power grid defect identification system. Through the dynamic analysis and tagging process of real-time collected data, structured feature vectors are formed, providing standard data input for subsequent classification and enhancing the compatibility of heterogeneous data sources. Based on the fusion decision-making mechanism of D-S evidence theory and historical defect maps, the misjudgment probability caused by traditional single-threshold determination can be reduced, and the associated recognition ability for complex defects can be enhanced. By pre-classifying data and the directional matching of dedicated defect identification Agents, the probability of task overlap during multi-Agent concurrency can be reduced, and the conflict risk caused by resource competition can be lowered. The classified tag data directly triggers the defect identification algorithm of the corresponding Agent, shortening the communication link of traditional cross-system data calls and facilitating the reduction of the defect escalation risk caused by handling delays. Structured reports containing classification tags, evidence weights, defect location, etc. are automatically generated, providing traceable decision-making basis for the equipment maintenance system and reducing manual transcription errors.

[0032] Figure 1 This is a schematic diagram of an application scenario provided by the present application. When using Agents for automated power grid defect identification, the method provided by the present application is applied.

[0033] Specifically, the method provided by the present application is applied to any server. The server interacts with multi-source devices and the equipment maintenance system, obtains real-time collected data through the multi-source devices, analyzes the real-time collected data, and uses defect identification Agents to determine the current power grid defects and generate defect reports to feedback to the equipment maintenance system.

[0034] The specific implementation method can refer to the following embodiments.

[0035] Figure 2 This is a flowchart of an Agent-based automated power grid defect identification system provided by an embodiment of the present application. The system of this embodiment can be applied to the server in the above scenario. As Figure 2 shown, the system includes: S201. Obtain real-time collected data, analyze the real-time collected data to get a data analysis result, and perform tagging processing on the real-time collected data according to the data analysis result to obtain tagged data; The real-time collected data can be instantaneous monitoring data and operation and maintenance records generated during the operation of power grid equipment, including sensor values (structured) and manual records (unstructured).

[0036] The data analysis result can be a set of data features extracted through statistical analysis and pattern matching, such as voltage fluctuation amplitude and insulation deterioration trend.

[0037] The tagged data can be a structured data unit with attached semantic identifiers, used to represent specific abnormal types of equipment status (such as "tag: mechanical displacement_overlimit_level 3").

[0038] Specifically, through multi-source devices such as smart meters, SCADA systems, and vibration sensors, structured data such as voltage waveforms, insulation resistance values, and mechanical vibration spectra are synchronously collected at a sampling frequency of 5 ms to 10 s. At the same time, unstructured data such as text logs and image records input by operation and maintenance personnel are received. The unstructured data and structured data are used as real-time collected data, and then the sliding window algorithm is used to perform smoothing filtering on the time-series data in the real-time collected data, and keyword fields (such as equipment numbers and anomaly descriptions) in the unstructured text are parsed through regular expressions.

[0039] Calculate statistical features (mean, variance, kurtosis) for the structured data, extract keywords (such as "insulation degradation" and "displacement overlimit") for the unstructured data, so as to achieve dynamic analysis and obtain a data analysis result. Then, according to the tag records of the enterprise's historical period tags, set tag division rules (for example: voltage fluctuation > ±5% → tag "voltage anomaly"; text contains "oil leakage" → tag "seal failure"). Then, based on the data analysis result, generate tagged data in JSON format containing equipment ID, timestamp, and feature values.

[0040] S202. Based on the D-S evidence theory, classify the tagged data according to the preset historical defect map; The D-S evidence theory can be a mathematical method for dealing with uncertainty and multi-source information fusion, which quantifies the support degree of different evidences for defect classification through the basic probability assignment function.

[0041] The preset historical defect map can be a structured knowledge base containing equipment defect historical cases, feature association rules, and disposal strategies, which can be used to guide classification decisions.

[0042] Specifically, historical defect cases can be abstracted into a triple graph of "defect type - characteristic parameter - disposal plan" in advance, and then a historical defect graph can be obtained to classify the labeled data.

[0043] Multiple features in the labeled data (such as the simultaneous presence of "voltage sudden change +0.5 kV" and "vibration frequency offset +15 Hz") are used as independent evidence sources, and the combined confidence is calculated through the corresponding combination rules of the D-S evidence theory; when different evidences point to contradictory defect types, the weighted average method is used to adjust the basic probability assignment, and the defect category with a higher confidence in the historical defect graph is preferentially matched, so as to realize the classification processing of the labeled data.

[0044] S203. Use the defect recognition Agent to identify the classified labeled data, and generate a defect report according to the recognition result and feedback it to the equipment maintenance system.

[0045] The defect report can be a standardized output document containing elements such as defect type, location information, and disposal suggestions, which can be used to guide the operation and maintenance operations.

[0046] The equipment maintenance system can be a backend management platform that receives defect reports and generates repair work orders and schedules resources.

[0047] The defect recognition Agent can be a software agent with autonomous decision-making ability, dedicated to the recognition and execution of the disposal logic for specific types of defects.

[0048] Specifically, the classified labeled data triggers a specific Agent instance (such as data of the "insulation deterioration" type only activates the insulation detection Agent), and then the corresponding defect is identified. Then, the recognition result can be output according to the report template set in advance by the enterprise, and a defect report is obtained and feedback to the equipment maintenance system.

[0049] Through the solution provided in this embodiment, the dynamic analysis and labeling process of real-time collected data form a structured feature vector, providing standard data input for subsequent classification and improving the compatibility of heterogeneous data sources. Based on the fusion decision-making mechanism of the D-S evidence theory and the historical defect graph, the misjudgment probability caused by the traditional single-threshold determination can be reduced, and the associated recognition ability for composite defects can be enhanced. Through the pre-data classification processing and the directional matching of dedicated defect recognition Agents, the probability of task overlap during multi-Agent concurrency can be reduced, and the conflict risk caused by resource competition can be reduced. The classified labeled data directly triggers the defect recognition algorithm of the corresponding Agent, shortening the communication link of traditional cross-system data calls, which is beneficial to reducing the defect escalation risk caused by disposal delays. Automatically generate a structured report containing classification labels, evidence weights, defect locations, etc., providing a traceable decision-making basis for the equipment maintenance system and reducing manual transcription errors.

[0050] In some embodiments, real-time acquired data is parsed to determine an acquired image; the acquired image is analyzed to determine a number of pixel features; the number of pixel features is analyzed to determine the device contour and the spatial distribution boundary; and the device contour and the spatial distribution boundary are determined as the data analysis result.

[0051] The acquired image can be visual data of the device operating state obtained through a camera or an infrared device, and is used to analyze physical form anomalies (such as mechanical displacement and component deformation).

[0052] The pixel features can be quantified attributes extracted from image pixels, including color distribution, gradient intensity, texture complexity, etc.

[0053] The device contour can be the closed boundary of the device main body area in the image, represented by a sequence of polygon vertex coordinates.

[0054] The spatial distribution boundary can be the actual position range of the device components in the physical space (such as the safety distance threshold between the high-voltage side and the low-voltage side of a transformer).

[0055] Specifically, from the real-time acquired mixed data stream (including sensor values, text logs, image / video streams), image data packets are separated through protocol parsing (such as MQTT / HTTP). Denoising (median filtering), format standardization (uniform conversion to PNG format), and timestamp alignment (synchronization with sensor data) operations are performed on the original image data packets to obtain the acquired image. Then, the color image is converted to a grayscale image using the OpenCV library, and the local binary pattern (LBP) algorithm is used to determine the texture features in the acquired image to obtain a number of pixel features; the Sobel operator is used to calculate the image gradient magnitude of the number of pixel features to extract the device contour (such as the transformer bushing boundary).

[0056] Based on the camera calibration parameters (intrinsic matrix, distortion coefficient) and the extrinsic parameters (installation height, pitch angle), a mapping relationship from the image pixel coordinate system to the actual physical coordinate system is established, the spatial boundary identified within the historical period is obtained as a template, and according to the mapping relationship and the template, the spatial boundary distribution in the pixel features is identified, and the device contour and the spatial distribution boundary are determined as the data analysis result.

[0057] Through the solution provided in this embodiment, image data is accurately separated from multi-source real-time acquired data (such as video streams, sensor mixed data streams), and non-image interference information is excluded. Key pixel features in the image are extracted, and visual information is converted into numerical features and geometric parameters, making the image information quantifiable and eliminating the dependence on manual experience. The contour and boundary data are encapsulated into a standardized format (such as a coordinate sequence) and directly serve the defect classification and disposal process.

[0058] In some embodiments, according to the device profile, several independent device entities in the data are determined; the power grid GIS database is obtained, and according to the power grid GIS database, the physical connection relationships between the independent device entities are determined; according to the spatial distribution boundaries, independent regions are determined; based on the independent regions and physical connection relationships, the real-time acquisition data of several independent device entities are tagged to obtain tagged data.

[0059] The power grid GIS database can be a database that stores the geographical coordinates, topological connection relationships, and attribute information (such as voltage level, device type) of power equipment.

[0060] The physical connection relationship can be an energy transmission path formed between devices through physical lines (such as cables, busbars), which can reflect the power grid topological structure.

[0061] An independent region can be a logical unit divided according to the spatial distribution or function of devices, used to limit the Agent collaboration scope and reduce communication latency.

[0062] Specifically, based on the device profile in the real-time data analysis results, the independent entities of multiple devices in the image or point cloud data are distinguished through a spatial clustering algorithm (such as DBSCAN). The device topology data stored in the power grid GIS (Geographic Information System) database is called, and the real-time device entities are spatially associated with the device nodes (such as transformers, circuit breakers) in the GIS database through a graph theory matching algorithm to determine the physical connection relationships between the independent device entities. Using the spatial distribution boundaries in the data analysis results, through spatial grid division (such as 50m×50m grid) or region aggregation algorithm (such as Voronoi diagram), the power grid devices are divided into several independent regions (such as Substation A Area, Transmission Corridor B Area). Combining the output of the foregoing steps, structured tags are added to the real-time data of each device entity, for example: Device-level tags: Device ID: Unique identifier (such as T1_SubstationA); Belonging region: Name of the independent region (such as "Substation A Area"); Connected devices: Upstream and downstream device IDs in the physical connection relationship (such as [CB2,L3]).

[0063] Data-level tags: Collection timestamp: Data generation time (such as 2024-05-01T14:30:00Z); Spatial boundary: Vertex coordinates of the device profile (such as [[116.404,39.915],[116.405,39.916]]); Associated defect type: Potential defects pre-associated according to the historical defect atlas (such as "insulation deterioration", "mechanical displacement").

[0064] Through the solution provided by this embodiment, by matching the device profile with the GIS library, it is ensured that the data corresponds one-to-one with the physical device, eliminating mislabeling of the device. The physical connection relationship label enables the Agent to understand the dependency relationship between devices, reducing the collaborative error rate caused by topological misunderstandings. The independent area division restricts the global competition of multiple Agents, reducing the resource conflict rate. The structured label data directly serves the subsequent D-S evidence theory classification and Agent decision-making, shortening the defect identification and handling link.

[0065] In some embodiments, the label data is analyzed to determine whether any of the real-time acquisition data belongs to multi-label data; if so, according to the D-S evidence theory and the preset historical defect atlas, the multi-label data is conflict resolved to obtain normal label data; the normal label data is analyzed to determine the label attributes, and the normal label data is classified according to the label attributes.

[0066] The multi-label data may be that the real-time data of the same device subject is simultaneously associated with multiple labels (such as "insulation deterioration" and "mechanical displacement") due to multi-dimensional analysis (such as voltage fluctuation, temperature anomaly).

[0067] The normal label data may be the unique or high-confidence label data retained after conflict resolution and can be directly used for subsequent classification and Agent decision-making. In a specific implementation manner, the data that does not belong to multi-label data does not need to be conflict resolved and can also be regarded as normal label data.

[0068] The label attributes may be dimensions describing the defect characteristics, including type, urgency, scope of influence, etc., and are used to guide the classification rules and handling priorities.

[0069] Specifically, each record in the label data is traversed to check whether it contains multi-label data (that is, the real-time data of the same device subject is associated with multiple label attributes, such as the simultaneous presence of "insulation deterioration" and "mechanical displacement"). If multi-label data exists, a basic probability assignment (BPA) is constructed: according to the historical defect probability of the same type of device in the preset historical defect atlas, an initial confidence level is assigned to each label (such as the confidence level of "insulation deterioration" is 0.7). The Dempster combination rule is calculated for multiple evidences (such as label A, label B) in the multi-label data, and then the normal label data after conflict resolution is output.

[0070] Analyze the normal label data (e.g., the name may reflect the defect type), classify it according to the evolution speed of the same type of defect in the historical defect atlas (such as "urgent", "high", "medium", "low"), and judge whether the defect is likely to spread to adjacent devices based on the physical connection relationship (such as "local", "regional level"), so as to determine the label attributes of the current normal label data. And classify the normal label data according to the label attributes, such as classifying the data into a preset classification pool according to the defect type (such as "insulation defect", "mechanical defect"); or generating a priority label according to the urgency and influence range (such as "insulation degradation - urgent - regional level").

[0071] Through the solution provided by this embodiment, the confidence levels of multiple labels are synthesized by the D-S evidence theory, redundant or contradictory labels are eliminated, the Agent is prevented from misjudging the defect type due to data conflicts, and the collaborative error is reduced. Classify the data according to the label attributes (such as "insulation type", "mechanical type"), making the classification results more in line with the actual scenario. Automated conflict resolution and classification replace traditional manual review, shortening the disposal response time.

[0072] In some embodiments, based on the physical connection relationship, analyze the device distribution in the independent area, and determine the associated devices belonging to the same electrical circuit in the independent area; and merge the environmental labels of the associated devices; after determining the associated devices, according to the device spatial distribution in the independent area, determine whether there are isolated devices; if there are isolated devices, generate device-level labels separately; according to the merger result and the device-level labels, obtain the label data.

[0073] Associated devices can be devices directly or indirectly connected in the same closed electrical circuit, and their operating states affect each other.

[0074] Environmental labels can be labels describing the environmental state of the device, including numerical types (such as temperature, humidity) and state types (such as "excessive electromagnetic interference").

[0075] Isolated devices can be independently operating devices that do not form a closed circuit with other devices in the independent area, and their data needs to be processed separately.

[0076] Device-level labels can be independent labels generated for isolated devices, including their exclusive state parameters (such as vibration, insulation resistance).

[0077] Specifically, construct an adjacency matrix between devices through the physical connection relationship (such as cable, bus connection). Based on the graph traversal algorithm (such as depth-first search), identify the set of devices forming a closed circuit and mark them as associated devices of the same electrical circuit.

[0078] Take the maximum or average value of the environmental parameters in the associated devices (e.g., take the highest value for "temperature = 45°C") to merge the environmental tags of the associated devices. If any device triggers an abnormal state (e.g., "humidity exceeds the limit"), then merge them into loop-level tags (e.g., "Loop_001: humidity abnormal").

[0079] After determining the associated devices, count the remaining devices not covered by the list of associated devices. Check whether the remaining devices have the ability to operate independently (such as energy storage devices, independent sensors). If they have the ability to operate independently, then identify them as isolated devices.

[0080] Tags generated only for the independent monitoring data of isolated devices, that is, device-level tags. Integrate loop-level tags (merged associated device data) with device-level tags (isolated device data) into tag data in a unified format.

[0081] Through the solution provided in this implementation, identify the set of devices in the same electrical loop through the physical connection relationship, aggregate the scattered device data into loop-level tags, and avoid the logical fragmentation caused by the isolated analysis of single-device data. Reduce the need to repeatedly label the same environmental parameters (such as loop temperature, humidity) for multiple devices within the same loop, and compress the original data volume. Detect independent devices that do not form a closed loop through the device spatial distribution, and avoid their data being misclassified or omitted. Generate exclusive device-level tags for isolated devices to ensure that their independent operating status is completely recorded and tracked. Through tag merging and classification, compress PB-level original data into high-value tag data, reducing the data volume and computing load processed by the Agent.

[0082] In some embodiments, parse the power grid GIS library to determine the device ID and device coordinate information of independent devices; based on the device ID, quickly match through the spatial index algorithm to obtain the adjacent devices of the independent devices; according to the device coordinate information, determine the device spacing between the adjacent devices and the independent devices; compare the device spacing with the preset interval threshold; if the device spacing is less than the preset interval threshold, then determine that there is a physical connection relationship.

[0083] The device ID can be a code that uniquely identifies an independent device in the power grid (such as "Substation A_Circuit Breaker_001").

[0084] The device coordinate information can be the coordinates of the device in space, which can be determined according to the established coordinate system.

[0085] The adjacent device can be a device adjacent to the independent device.

[0086] The device spacing can be the actual physical distance between devices, calculated based on coordinates, with the unit of meter (m).

[0087] The preset interval threshold can be the maximum physical connection distance allowed between devices defined in the grid code, and the threshold varies depending on the device type (e.g., for cables ≤ 20m, for busbars ≤ 5m).

[0088] Specifically, extract the device attribute table from the GIS library to obtain the device ID (unique identifier) and device coordinate information (latitude and longitude or three-dimensional coordinates) of each independent device. Convert the coordinate information to a unified geographic coordinate system (such as WGS-84 or the local engineering coordinate system).

[0089] Based on the device ID, use an R-tree or a quadtree to establish a spatial index for the device coordinates to accelerate the query of neighboring devices. Taking the current device as the center, quickly retrieve candidate devices within a certain range through the spatial index. Screen out the devices among the candidate devices that belong to the same voltage level or device type as the current device to obtain the adjacent devices of the independent device.

[0090] Use the Euclidean distance formula (two-dimensional or three-dimensional) to calculate the straight-line distance between devices to obtain the device spacing between the adjacent device and the independent device.

[0091] According to the device type and the maximum allowable physical connection distance defined in the grid code (such as the cable connection threshold ≤ 20m), that is, the preset interval threshold. Compare the device spacing with the preset interval threshold; if the device spacing is less than the preset interval threshold, it is determined that there is a physical connection relationship.

[0092] Through the solution provided in this embodiment, by parsing the grid GIS library, extracting the mapping relationship between the device ID and the coordinate information, ensuring that the geographical location of each independent device is unique and traceable, providing basic data support for subsequent physical connection analysis. Quickly screen out the list of neighboring devices of the target device through the spatial index algorithm, reducing the time complexity of neighboring device queries and adapting to the efficient processing requirements of PB-level grid data. Calculate the actual spacing based on the device coordinates, combined with the preset interval threshold, excluding the invalid associations of long-distance devices, ensuring that the connection relationship conforms to the actual engineering constraints. Clarify the electrical circuit attribution among device groups, providing a topological basis for subsequent defect identification and task assignment in the Agent system.

[0093] In some embodiments, obtain the device information of the adjacent device; according to the device information, determine the voltage level; compare the voltage level with the voltage level of the independent device; if the device spacing is less than the preset interval threshold and the voltage level is the same as the voltage level of the independent device, it is determined that there is a physical connection relationship.

[0094] The device information can be the attribute data of grid devices, including device ID, type, voltage level, port configuration, installation time, etc.

[0095] The voltage level can be the classification of the operating voltage range allowed by the design of grid equipment (such as 10 kV, 35 kV, 110 kV), which is defined by grid specifications.

[0096] Specifically, extract the device information of adjacent devices (including attributes such as device ID, device type, voltage level, port configuration, etc.) from the grid GIS database or device attribute table. Extract the "voltage level" field from the device information, ignore other attributes, and obtain the corresponding voltage level. Compare the voltage level with that of the independent device; if the device spacing is less than the preset interval threshold and the voltage levels are the same, it is determined that there is a physical connection relationship.

[0097] Through the solution provided in this embodiment, voltage level matching is introduced to ensure that the physical connection relationship complies with grid safety specifications and avoid the one-sidedness of relying solely on spatial distance. Double-condition filtering reduces the misjudgment of invalid connection relationships.

[0098] In some embodiments, according to the preset historical defect map, determine the label confidence of multi-label data; analyze several labels of the multi-label data to determine whether there are any two mutually exclusive label attributes; if so, based on the D-S evidence theory, according to the label confidence, resolve the conflict between the two labels with mutually exclusive attributes to obtain normal label data.

[0099] Specifically, based on the preset historical defect map, identify the matching frequency of the label with the real defect in the historical data, and calculate the label confidence of the multi-label data according to the matching frequency (for example, if the accuracy rate of the overload label in history is 90%, the confidence is 0.9).

[0100] Since there cannot be logical contradictions in the labels of the same device at the same time (such as "overload" and "low load"). Therefore, based on grid physical rules and device operation logic, predefined mutually exclusive label pairs are defined (such as overload ↔ low load, normal insulation ↔ deteriorated insulation). Traverse the label set to check whether there are predefined mutually exclusive label pairs, so as to determine whether there are any two mutually exclusive label attributes.

[0101] If so, based on the D-S evidence theory, according to the label confidence, and according to the conflict resolution scheme provided in the above embodiment, obtain the normal label data.

[0102] Through the solution provided in this embodiment, by detecting and eliminating tags with mutually exclusive attributes, it is ensured that the tag data of the same device has no contradictions at the physical or logical level, avoiding misjudgment by the defect recognition Agent caused by tag contradictions, thereby reducing the collaborative error rate. Based on the tag confidence of the historical defect map, reduce the repeated handling work orders caused by low-quality tags, improve the accuracy of defect reports, and prevent secondary defects from evolving into system failures. Alleviate the interference of tag noise in PB-level data to the system, and enhance the robustness of defect recognition under the new power system (high complexity, multi-parameter coupling).

[0103] In some embodiments, sort the tag confidences of two tags with mutually exclusive attributes. According to the sorting result, determine the tag with low confidence, and based on the D-S evidence theory, eliminate the tag with low confidence to obtain normal tag data.

[0104] Specifically, directly compare the confidence values of the two tags and arrange them in descending order (e.g., overload (0.9) > low load (0.3)). Select the tag with lower confidence in the sorting result (e.g., the confidence of "low load" is 0.3). And based on the D-S evidence theory, eliminate the tag with low confidence to obtain normal tag data.

[0105] Through the solution provided in this embodiment, directly resolve conflicts through confidence sorting, avoid complex calculations, and forcibly eliminate mutually exclusive tags to prevent the system from falling into logical chaos due to contradictory data.

[0106] In some embodiments, determine the data source according to the real-time collected data; determine the confidence weights of multi-tag data according to the data source; and determine the tag confidence of multi-tag data according to the preset historical defect map and confidence weights.

[0107] The data source can be a combination of the generation location and device type of the real-time collected data (e.g., "high-voltage equipment area - temperature monitoring").

[0108] Specifically, the real-time collected data includes monitoring data (such as voltage fluctuation values, temperature sensor readings) and its metadata (such as sensor ID, device location, collection time). Analyze the real-time collected data to determine the device location (e.g., "Substation A - Transformer Bank 1"), sensor type (e.g., "infrared thermometer"), and collection timestamp contained in the metadata to obtain the data source.

[0109] Based on the statistics of tag accuracy for different data sources in the preset historical defect map (e.g., the historical accuracy rate of the "high-voltage equipment area - temperature monitoring" tag is 85%), generate a confidence weight table, and then extract the corresponding confidence weight from the weight table according to the current data source tag (e.g., "high-voltage equipment area - temperature monitoring") to determine the tag confidence of multi-tag data.

[0110] Through the solution provided in this embodiment, based on the objective statistics of the preset historical defect map and the real-time data source, the subjective interference of manual experience is avoided. The confidence calculation process is transparent (historical frequency × weight), which is convenient for operation and maintenance personnel to trace the decision-making basis.

Claims

1. An Agent-based automated power grid defect identification system, characterized in that Including: Obtain real-time acquisition data, analyze the real-time acquisition data to obtain a data analysis result, and perform tagging processing on the real-time acquisition data according to the data analysis result to obtain tag data; Based on the D-S evidence theory, classify the tag data according to a preset historical defect map; Use a defect recognition Agent to identify defects in the classified tag data, generate a defect report according to the recognition result, and feedback it to the device maintenance system.

2. The system according to claim 1, wherein, The analyzing the real-time acquisition data to obtain a data analysis result includes: Parse the real-time acquisition data to determine an acquisition image; Analyze the acquisition image to determine a number of pixel features; Analyze the number of pixel features to determine the device contour and the spatial distribution boundary; Determine the device contour and the spatial distribution boundary as the data analysis result.

3. The system according to claim 2, wherein The performing tagging processing on the real-time acquisition data according to the data analysis result to obtain tag data includes: Determine a number of independent device entities in the data according to the device contour; Obtain a power grid GIS database, and determine the physical connection relationship between independent device entities according to the power grid GIS database; Determine an independent area according to the spatial distribution boundary; Perform tagging processing on the real-time acquisition data of a number of independent device entities according to the independent area and the physical connection relationship to obtain tag data.

4. The system according to claim 1, characterized in that, The classifying the tag data according to the D-S evidence theory and a preset historical defect map includes: Analyze the tag data to determine whether any real-time acquisition data belongs to multi-tag data; If so, perform conflict resolution on the multi-tag data according to the D-S evidence theory and the preset historical defect map to obtain normal tag data; Analyze the normal tag data to determine the tag attributes, and classify the normal tag data according to the tag attributes.

5. The system according to claim 3, characterized in that, The performing tagging processing on the real-time acquisition data of a number of independent device entities according to the independent area and the physical connection relationship to obtain tag data includes: Based on the physical connection relationship, analyze the device distribution in the independent area to determine associated devices belonging to the same electrical circuit in the independent area; And merge the environmental tags of the associated devices; After determining the associated devices, determine whether there are isolated devices according to the device spatial distribution in the independent area; If there are isolated devices, generate device-level tags separately; Obtain tag data according to the merging result and the device-level tags.

6. The system according to claim 3, wherein The determining the physical connection relationship between independent device entities according to the power grid GIS database includes: Parse the power grid GIS database to determine the device ID and device coordinate information of the independent device; Based on the device ID, quickly match through a spatial indexing algorithm to obtain adjacent devices of the independent device; Determine the device spacing between the adjacent devices and the independent device according to the device coordinate information; Compare the device spacing with a preset interval threshold; If the device spacing is less than the preset interval threshold, determine that there is a physical connection relationship.

7. The system according to claim 6, characterized in that, Before determining that there is a physical connection relationship if the device spacing is less than the preset interval threshold, it further includes: Obtaining the device information of the adjacent devices; Determining the voltage level according to the device information; Comparing the voltage level with the voltage level of the independent device; If the device spacing is less than the preset interval threshold and the voltage level is the same as the voltage level of the independent device, it is determined that there is a physical connection relationship.

8. The system according to claim 4, characterized in that, The conflict resolution of the multi-label data according to the D-S evidence theory and the preset historical defect map to obtain normal label data includes: Determining the label confidence of the multi-label data according to the preset historical defect map; Analyzing several labels of the multi-label data to determine whether there are any two mutually exclusive label attributes; If so, based on the D-S evidence theory and according to the label confidence, the conflict resolution of the two labels with mutually exclusive attributes is performed to obtain normal label data.

9. The system according to claim 8, wherein The conflict resolution of the two labels with mutually exclusive attributes based on the D-S evidence theory and according to the label confidence to obtain normal label data includes: Sorting the label confidences of the two labels with mutually exclusive attributes, determining the label with low confidence according to the sorting result, and based on the D-S evidence theory, resolving the label with low confidence to obtain normal label data.

10. The system according to claim 8, wherein Determining the label confidence of the multi-label data according to the preset historical defect map includes: Determining the data source according to the real-time collected data; Determining the confidence weight of the multi-label data according to the data source; Determining the label confidence of the multi-label data according to the preset historical defect map and the confidence weight.

Images