A method, system, and medium for risk assessment of power grid operations

Abstract

The application discloses a power grid operation risk assessment method, system and medium, and belongs to the technical field of smart grids. The method comprises the following steps: acquiring an electronic work ticket text of a power grid operation and analyzing the same, generating a plurality of action units and integrating the same into a power grid operation action sequence; extracting a current execution action unit, a standard position coordinate and a standard operation time thereof from the sequence, acquiring personnel actual position and operation behavior data, and determining an operation execution characteristic vector; determining environment abnormal information according to environment monitoring data, determining a space risk region according to matching between the standard position coordinate and a dangerous region set, and generating a risk constraint vector according to the space risk region and the environment abnormal information; and determining a comprehensive risk value of the current power grid operation according to the operation execution characteristic vector and the risk constraint vector. Through implementation of the application, the problem of rigid risk dimension in power grid operation risk assessment can be solved, and the accuracy of risk assessment is improved.

Description

Technical Field

[0001] This application relates to the field of smart grid technology, and in particular to a risk assessment method, system and medium for power grid operations. Background Technology

[0002] The power grid operating environment is complex and variable, involving various risk factors such as falls from heights, safe distances from live conductors, accumulation of toxic gases in confined spaces, and severe weather. Traditional manual monitoring methods rely on the on-site observation and experience of monitoring personnel, which is difficult to achieve full-time, all-round coverage, and is also costly and slow in response. Therefore, accurate risk assessment for power grid operations is of great significance for ensuring the safety of workers.

[0003] Currently, power grid operation risk assessment typically employs reasoning methods based on static rule bases. These methods pre-load fixed risk dimensions according to the operation type and set corresponding threshold rules. In practice, this reveals inherent limitations: once the risk dimensions used for assessment are preset, they cannot be changed. When unplanned location deviations, process adjustments, or sudden environmental changes occur during actual operations, the system cannot dynamically add or switch corresponding risk dimensions, nor can it distinguish the risk significance of the same behavioral deviation in different environmental contexts. This results in alarms being triggered by location deviations or operation timeouts in scenarios without actual hazards such as high-altitude edges or energized areas, but failing to generate targeted safety warnings. Conversely, in scenarios with actual hazards, the contribution of behavioral deviations to the overall risk lacks precise quantification, leading to a disconnect between assessment accuracy and actual on-site risks. Therefore, how to dynamically adjust risk assessment dimensions and the actual evolution of operations to improve the accuracy of power grid operation risk assessment is a pressing technical problem that needs to be solved. Summary of the Invention

[0004] This application provides a method, system, and medium for risk assessment of power grid operations, which solves the problem of rigid risk dimensions in existing power grid operation risk assessments and improves the accuracy of power grid operation risk assessments.

[0005] This application provides a risk assessment method for power grid operations, including: Obtain the electronic work order text for power grid operations; parse the electronic work order text to generate several action units; and integrate each action unit with standard location coordinates obtained from a geographic information system and a preset standard operation time into a power grid operation action sequence. Extract the currently executing action unit, the corresponding standard position coordinates, and the standard operation time from the power grid operation action sequence, and obtain the actual position data and operation behavior timing data of the personnel in the currently executing action unit; determine the operation execution feature vector based on the actual position data, the operation behavior timing data, the standard position coordinates, and the standard operation time; Based on the acquired environmental monitoring data of power grid operations, environmental anomaly information is determined; spatial matching is performed between the standard position coordinates corresponding to the currently executing action unit and a preset set of dangerous areas to determine spatial risk areas; and a risk constraint vector is generated based on the spatial risk areas and the environmental anomaly information. The comprehensive risk value of the current power grid operation is determined based on the operation execution feature vector and the risk constraint vector.

[0006] This application parses electronic work order text into action units and integrates them into an action sequence by associating them with standard location coordinates and standard operation time. This transforms the work process from static text into a quantifiable spatiotemporal benchmark, providing a structured framework for subsequent real-time comparison. The benchmark granularity for risk assessment is refined from the entire work process to each smallest execution unit. Standard data for the current action unit is extracted from the action sequence, along with the actual location and timing data of personnel's operational behavior. This data is compared with the benchmark to obtain a work execution feature vector, quantifying the quality of work behavior into numerical evidence that can be used in risk calculation. Environmental anomalies are identified based on environmental monitoring data, and spatial risk areas are determined by matching standard location coordinates with hazardous areas. These two are then fused to generate a risk constraint vector, achieving dynamic coupling between sudden changes in the on-site environment and static geographical constraints. This allows external hazards to be structurally incorporated into the assessment input. The work execution feature vector and risk constraint vector are fused to determine a comprehensive risk value, enabling behavioral quality and environmental constraints to work together within the same framework, outputting a risk level adapted to the current work dynamics. Different action units can correspond to different risk focuses, achieving automatic switching of risk dimensions with different work stages, rather than preset fixed dimensions, thus solving the problem of rigid risk dimensions. Compared with existing technologies, this application transforms risk assessment from static rule matching to action sequence-driven point-by-point quantification and multi-source evidence fusion, thereby improving the accuracy of power grid operation risk assessment.

[0007] Further, determining the job execution feature vector based on the actual location data of the personnel, the timing data of the operation behavior, the standard location coordinates, and the standard operation time includes: Calculate the positional offset between the actual position data of the personnel and the standard position coordinates, and generate a spatial deviation feature value based on the positional offset; Calculate the duration deviation rate between the actual duration of the operation behavior timing data and the standard operation time, and generate a time deviation feature value based on the duration deviation rate; Calculate the matching similarity between the time series of the velocity change rate of the actual personnel location data and the time series of the action switching frequency of the operation behavior time series data, and determine the behavior coordination feature value based on the matching similarity; The spatial deviation feature value, the temporal deviation feature value, and the behavioral coordination feature value are combined to obtain the job execution feature vector.

[0008] Spatial deviation feature values ​​are generated by calculating the offset between the actual position and the standard position, quantifying the degree of positional deviation into a numerical indicator that can be used for subsequent evaluation. Temporal deviation feature values ​​are generated by calculating the deviation rate between the actual duration of the operation and the standard duration, independently extracting the degree of time-consuming loss of control from the operation sequence. Behavioral coordination feature values ​​are determined by calculating the matching similarity between the rate of change of the personnel's movement speed and the frequency of hand operation switching, converting the temporal matching relationship between limb movement and fine manipulation into a quantitative score. The matching similarity can be quantified through cross-correlation calculations to determine the degree of synchronization between the two types of actions, showing high sensitivity to risks such as posture instability in high-altitude operations and decreased hand-eye coordination in live-line operations. The spatial deviation feature values, temporal deviation feature values, and behavioral coordination feature values ​​are combined into a work execution feature vector, enabling the quality of work behavior to be comprehensively characterized from three orthogonal dimensions: positional deviation, time-consuming loss of control, and movement coordination, providing fine-grained quantitative evidence input at the action level for comprehensive risk assessment.

[0009] Further, the process of parsing the electronic work order text to generate several action units, and then associating each action unit with standard location coordinates obtained from the geographic information system and a preset standard operation time, integrates them into a power grid operation action sequence, including: The electronic work order text is parsed using natural language to extract structured work instructions and generate an initial work instruction set; Based on the initial set of operation instructions, a time-series logic completion is performed using a preset power grid operation process knowledge graph to generate an enhanced operation instruction sequence. The enhanced operation instruction sequence is broken down into several indivisible action units; Obtain the standard location coordinates and preset standard operation time of each action unit in the geographic information system, and associate the standard location coordinates and the standard operation time with the corresponding action unit; integrate the action units into a power grid operation action sequence according to the power grid operation execution order.

[0010] By extracting structured work instructions from electronic work order text through natural language parsing, unstructured work text is transformed into a computable set of instructions, providing structured input for subsequent automated processing. A pre-defined power grid operation process knowledge graph is used to complete the temporal logic of the initial instruction set, automatically filling in missing pre- or post-step steps, ensuring the logical closure of the work instruction sequence and avoiding evaluation blind spots caused by omissions in work order descriptions. The enhanced instruction sequence is broken down into indivisible action units, and complex operations are decomposed into atomic sequences of single actions, laying a granular foundation for independent evaluation of each smallest execution link. The standard location coordinates and pre-defined standard operation time of each action unit in the geographic information system are obtained and associated with the corresponding action unit, then integrated into an action sequence according to the execution order. This provides each action unit with a spatial and temporal reference anchor, transforming the power grid operation process into a structured spatiotemporal trajectory that can be compared node by node.

[0011] Further, generating a risk constraint vector based on the spatial risk area and the environmental anomaly information includes: A spatial risk weight matrix is ​​generated based on the risk type corresponding to the spatial risk region; An environmental risk weight matrix is ​​generated based on the types and degrees of environmental parameters contained in the environmental anomaly information. The spatial risk weight matrix and the environmental risk weight matrix are weighted and fused to generate the risk constraint vector.

[0012] By mapping the risk types corresponding to spatial risk areas to a spatial risk weight matrix, the physical danger zones where personnel are located are quantified into numerical forms that can participate in the fusion calculation. The degree of anomaly of each parameter type in environmental anomaly information is mapped to an environmental risk weight matrix, quantifying sudden environmental changes such as wind speed, temperature, and humidity as independent risk contribution indicators. The spatial risk weight matrix and the environmental risk weight matrix are weighted and fused to generate a risk constraint vector, enabling the joint expression of geospatial constraints and environmental change constraints within the same mathematical framework, achieving a structured coupling of two types of heterogeneous risk factors.

[0013] Further, determining the comprehensive risk value of the current power grid operation based on the operation execution feature vector and the risk constraint vector includes: Based on the risk constraint vector and the timestamp information of the currently executing action unit, a joint confidence score is calculated; risk types whose joint confidence scores exceed a preset activation threshold are identified as activation risk types. For each activated risk type, the risk assessment model corresponding to the activated risk type is called from the risk model library, and the feature weight vector built into the risk assessment model is obtained. The spatial deviation feature value, the temporal deviation feature value, and the behavioral coordination feature value are weighted and summed according to the feature weight vector to obtain the first risk value of the activation risk type; The first risk values ​​of all the activation risk types are weighted and fused according to the risk intensity corresponding to each activation risk type to obtain the comprehensive risk value.

[0014] The joint confidence score is calculated based on the risk constraint vector and the timestamp information of the current action unit to filter risk types that exceed the preset activation threshold. This accurately locates the risk dimensions that need to be assessed at the current work node in the spatiotemporal dimension, avoiding interference from irrelevant risks. For each activated risk type, the corresponding risk assessment model is called from the risk model library, and the feature weight vector built into the risk assessment model is obtained to establish differentiated behavioral impact quantification benchmarks for different risk types. The spatial deviation feature value, temporal deviation feature value, and behavioral coordination feature value are weighted and summed according to the feature weight vector to obtain the first risk value for each activated risk type. The work execution quality feature vector is mapped to the local risk response of each risk type. The first risk values ​​of all activated risk types are weighted and fused according to their corresponding risk intensities to obtain a comprehensive risk value. Multi-dimensional risk contributions are integrated to output a quantitative indicator reflecting the overall degree of danger, thereby achieving a comprehensive risk judgment that dynamically adjusts the risk assessment dimensions based on the work context and integrates multi-source information.

[0015] Furthermore, determining environmental anomaly information based on the acquired environmental monitoring data of power grid operations includes: Extract wind speed, humidity, and temperature sequences from the environmental monitoring data; Determine whether each sampled value in the wind speed sequence, the humidity sequence, and the temperature sequence exceeds the corresponding preset safety threshold; When any of the sampled values ​​exceeds the corresponding preset safety threshold, environmental anomaly information is determined; wherein, the environmental anomaly information includes the sampled value and the data type corresponding to the sampled value.

[0016] By extracting wind speed, humidity, and temperature sequences from environmental monitoring data, the raw sensor data is decomposed into three independent environmental parameter dimensions, providing a structured input basis for subsequent anomaly detection. Each sequence's sampled value is then judged to determine whether it exceeds a corresponding preset safety threshold, transforming continuous environmental monitoring data into discrete threshold-exceeding judgment results, thus converting environmental states from analog quantities to logical flags. When any sampled value exceeds the preset safety threshold, environmental anomaly information is identified, and the anomaly sampled value and its data type are output, capturing environmental mutations as risk events with type identifiers, providing clear environmental evidence input for the subsequent generation of risk constraint vectors.

[0017] Further, the step of determining the spatial risk area by spatially matching the standard position coordinates corresponding to the currently executing action unit with a preset set of danger zones includes: The standard location coordinates are compared with the spatial locations of each danger zone in the set of danger zones to determine whether the standard location coordinates are located within the danger zone. When a dangerous area containing the standard location coordinates exists, the dangerous area is identified as a spatial risk area.

[0018] By comparing the standard location coordinates with the spatial locations of each hazardous area in the hazardous area set, it is determined whether the standard location coordinates are located within each hazardous area. This establishes a spatial topological relationship between the planned work point and the predefined hazardous area, enabling automatic determination of spatial affiliation. When a hazardous area containing the standard location coordinates exists, the hazardous area is identified as a spatial risk area. This allows the inherent geographical risks such as the proximity zone of charged objects and the critical zone for high-altitude falls to be identified during the work planning stage, providing a quantitative input of spatial dimensions for the subsequent risk constraint vector.

[0019] Furthermore, this application also includes: Obtain the historical risk trend data sequence from the power grid operation sequence; Generate a baseline risk evolution curve based on historical risk trend data sequences; Based on the baseline risk evolution curve and the comprehensive risk value, determine the deviation magnitude and rate of change difference of the currently executed action unit; based on the deviation magnitude and the rate of change difference, determine the risk abnormal deviation index; Calculate the risk value change range between the currently executing action unit and the adjacent action units, and determine whether the comprehensive risk value is an outlier based on the risk anomaly deviation index and the risk value change range; wherein, when the risk anomaly deviation index exceeds a preset deviation threshold and the risk value change range is less than a preset change tolerance, the comprehensive risk value is determined to be an outlier. If the overall risk value is an outlier, the electronic work order text fragment and the actual location data of the personnel corresponding to the currently executing action unit are re-parsed to correct the power grid operation action sequence.

[0020] By acquiring historical risk trend data sequences from the power grid operation sequence, historical reference data is provided for the risk verification of the current action unit during the operation evolution process. A benchmark risk evolution curve is generated based on the historical risk trend data sequence, constructing a spatiotemporal baseline reflecting the risk change pattern under normal operation logic. The deviation amplitude and rate of change of the current execution action unit are determined based on the benchmark risk evolution curve and the comprehensive risk value, quantifying the degree of deviation and fluctuation rate of the actual risk judgment result relative to the normal evolution trajectory. Risk anomaly deviation indicators are determined based on the deviation amplitude and the rate of change difference, aggregating multi-dimensional deviation information into a single, identifiable anomaly intensity measure. The risk value change amplitude between the current execution action unit and adjacent action units is calculated, and the preceding and following sections are extracted. The smoothness of the transition of the risk value; judging whether the comprehensive risk value is an outlier based on the risk anomaly deviation index and the magnitude of the risk value change, wherein when the risk anomaly deviation index exceeds the preset deviation threshold and the magnitude of the risk value change is less than the preset change tolerance, it is determined to be an outlier. By jointly judging the anomaly intensity and context smoothness, the true risk mutation is distinguished from the false peak caused by sensor noise or algorithm misjudgment; if the comprehensive risk value is an outlier, the electronic work ticket text fragment and the actual location data of the personnel corresponding to the current execution action unit are re-parsed to correct the power grid operation action sequence. By tracing back the original data, the action semantics and spatial coordinates of the abnormal node are reconstructed to keep the risk assessment result consistent with the operation evolution logic in a closed loop.

[0021] Another embodiment of the present invention provides a risk assessment system for power grid operations, which is used to perform the risk assessment method for power grid operations as described in this application.

[0022] Another embodiment of the present invention provides a computer-readable storage medium, including: a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the risk assessment method for power grid operations as described in this application. Attached Figure Description

[0023] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 A flowchart illustrating one embodiment of the risk assessment method for power grid operations provided in this application; Figure 2This is a schematic diagram of the job execution feature vector generation process of an embodiment of the method provided in this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0027] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0028] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0029] In power grid operation environments, existing risk assessment technologies generally rely on static rule bases, preset indicator systems, or expert-weighted models. These assessment dimensions are rigid and struggle to adapt to the dynamic evolution of work processes and real-time changes in the field environment. Especially in high-risk scenarios such as wind farm maintenance, high-voltage indoor and outdoor operations, and live-line inspections, the complex coupling relationships between work types, personnel behavior sequences, equipment status, and environmental constraints lead to frequent missed and false alarms in risk identification, resulting in low matching between assessment results and actual on-site needs. To address this, this application proposes a risk assessment method for power grid operations to improve the accuracy of risk assessment.

[0030] Please refer to Figure 1 , Figure 1This is a flowchart illustrating an embodiment of the risk assessment method for power grid operations provided in this application. To address the problems of rigid assessment dimensions, poor scenario adaptability, and insufficient matching between risk determination and operation evolution logic in existing technologies, an embodiment of this application provides a risk assessment method for power grid operations. This method includes steps S1 to S6, each step as follows: Step S1: Obtain the electronic work order text for power grid operation; parse the electronic work order text to generate several action units, and integrate each action unit with the standard location coordinates obtained from the geographic information system and the preset standard operation time into a power grid operation action sequence. Step S2: Extract the currently executing action unit and its corresponding standard position coordinates and standard operation time from the power grid operation action sequence, and obtain the actual position data and operation behavior time sequence data of the personnel executing the action unit; determine the operation execution feature vector based on the actual position data, operation behavior time sequence data, standard position coordinates, and standard operation time. Step S3: Determine environmental anomaly information based on the acquired environmental monitoring data of power grid operations; Step S4: Perform spatial matching between the standard position coordinates corresponding to the currently executing action unit and the preset set of dangerous areas to determine the spatial risk area; Step S5: Generate a risk constraint vector based on spatial risk areas and environmental anomaly information; Step S6: Determine the comprehensive risk value of the current power grid operation based on the operation execution feature vector and risk constraint vector.

[0031] An action unit is an indivisible basic execution unit representing a single work step, obtained by semantically parsing the electronic work order text. Each action unit is associated with at least the standard location coordinates and standard operation time required to execute that step. Multiple action units are linked according to the logical order of work execution, forming a semantic trajectory chain that represents the complete work process.

[0032] The job execution feature vector is a comprehensive evaluation index used to quantify the operational standardization and execution quality of personnel in the current action unit. This index is obtained by comprehensively matching and calculating the actual spatial movement trajectory and operational behavior sequence of the personnel during the job process with preset standard job requirements.

[0033] The risk constraint vector is a multi-dimensional structured data object used to quantitatively describe the strength of the comprehensive risk constraints faced by the current work action unit in the spatial and environmental dimensions. The comprehensive risk value is used to quantitatively characterize the overall danger level of the current execution action unit under the coupling effect of multiple environmental constraints and work execution quality; the higher the value, the greater the comprehensive risk of the current power grid work node.

[0034] In some embodiments, step S1 parses the electronic work order text to generate several action units. Each action unit is then associated with standard location coordinates obtained from a geographic information system and a preset standard operation time, and integrated into a power grid operation action sequence, including steps S11 to S14. The specific steps are as follows: Step S11: Perform natural language parsing on the electronic work order text, extract structured work instructions, and generate an initial work instruction set; Ontology mapping processing of the power grid operation domain is performed on the electronic work order text sequence. The original text is input into a dedicated ontology semantic parser, which performs semantic matching based on a predefined set of ontology nodes for operation type, equipment category, and safety measures, generating preliminary parsing results containing ontology node indices and semantic weights. Simultaneously, the text is segmented and named entity recognition is performed using a domain dictionary, annotating action verbs, equipment names, safety measure phrases, and operation type keywords in the text, and removing irrelevant stop words and non-business text fragments to form a domain-labeled lexical sequence, i.e., an entity sequence.

[0035] Syntactic dependency analysis is performed on the obtained entity sequence to construct a dependency tree between operation verbs, target devices, and safety measures, extracting key nodes from the instruction pattern. Then, these key nodes are encapsulated into fields to form a structured task metadata object containing job type, device number, and safety measure fields, and a unique identifier is established through metadata indexing. Semantic tagging is then performed on this metadata object: task category tags are generated based on job type, device attribute tags are generated based on device number, and safety policy tags are generated based on safety measures, resulting in a set of task instructions with multiple semantic tags.

[0036] Based on the dependency tree in the task instruction set, semantic role annotation is performed on action verbs, target devices, and safety measures: action verbs are identified as the action execution role, device names are identified as the target device role, and safety measures are identified as the safety constraint role, forming standardized job instruction triples (action—device—safety measure). Redundancy elimination and ambiguity resolution are performed by calculating semantic similarity through the context window to remove duplicate or contradictory triples, ensuring that each instruction is semantically unique and consistent. The above set of job instruction triples is then structured and encapsulated, assembling an initial job instruction set.

[0037] In one optional embodiment, the electronic work order text for a power grid maintenance task is: "Power outage maintenance of the 10KV busbar switch room for circuit breaker replacement. The operation requires voltage testing, grounding wire installation, and wearing insulated gloves." The system loads a domain dictionary containing terms such as "power outage maintenance," "busbar," "circuit breaker," "voltage testing," "grounding wire," and "insulated gloves," and obtains a word sequence after ontology mapping and word segmentation. Syntactic dependency analysis constructs a dependency tree, identifying candidate triples for "replace → circuit breaker," "voltage testing → grounding wire," and "wear → insulated gloves." Semantic role labeling sets "replace" as an action, "circuit breaker" as equipment, and "wear insulated gloves" as a safety measure, generating an initial work instruction set containing: {action: replace, equipment: circuit breaker, safety measure: wear insulated gloves}, {action: voltage testing, equipment: grounding wire, safety measure: wear insulated gloves}, and {action: install, equipment: grounding wire, safety measure: wear insulated gloves}.

[0038] Step S12: Based on the initial job instruction set, use the preset power grid operation process knowledge graph to complete the timing logic and generate an enhanced job instruction sequence; Based on the initial job instruction set, sequential logic reasoning and implicit condition completion operations are performed. The pre-trained power grid operation process knowledge graph is used to automatically infer and fill in the missing pre-verification steps or post-recovery steps, generating an enhanced job instruction sequence with a logical closed loop.

[0039] Step S13: Divide the enhanced operation instruction sequence into several indivisible action units; The enhanced job instruction sequence execution performs fine-grained semantic boundary recognition and atomized decomposition processing, cutting composite instructions into the smallest indivisible action units based on action duration thresholds and spatial transfer distance constraints.

[0040] Step S14: Obtain the standard location coordinates and preset standard operation time of each action unit in the geographic information system, and associate the standard location coordinates and standard operation time with the corresponding action unit; integrate the action units into a power grid operation action sequence according to the power grid operation execution order.

[0041] For the action unit generated in step S13, its original attribute fields of operation type, expected time consumption, and standard operation location are parsed sequentially to form a basic data object. One-hot encoding is performed on the operation type field to construct a category vector, which is then mapped to a dense representation using a domain-specific semantic embedding matrix to obtain a category-dense vector. Normalization is performed on the preset standard operation time field (i.e., expected time consumption), using the observation range of maximum and minimum time consumption as the normalization interval to transform the time span of different actions into comparable dimensionless features, resulting in normalized time features. Projection coordinate transformation is performed on the standard operation location field, mapping the input longitude, latitude, and elevation data to the power grid geographic information coordinate system to form a three-dimensional location vector of the standard location coordinates, resulting in a spatial location vector. The above category-dense vector, normalized time features, and spatial location vector are concatenated to form a high-dimensional feature vector containing multi-domain information. To ensure the unique traceability of timestamps and spatial coordinates, the time index and spatial index of the current action unit are appended to the high-dimensional feature vector to construct a structured, standardized semantic node data stream. Therefore, each action unit is associated with its standard position coordinates and preset standard operation time.

[0042] Based on the standardized semantic node data stream described above, the execution order of tasks between nodes is parsed, and a serialized node index table is extracted. A state transition determination algorithm is used to perform dependency detection on each pair of adjacent nodes to determine whether there are necessary process connection conditions, thus obtaining a set of connectable node pairs. For connectable node pairs, weighted directed edge objects are constructed based on the timestamp difference and spatial coordinate change between the start and end nodes, where the timestamp difference serves as the temporal weight and the spatial displacement serves as the spatial weight. The set of edge objects is input into a chain-based topology builder, and the nodes are sequentially chained in ascending order of their indices to form a directed graph structure with directionality and continuity. A topology sorting algorithm is used to ensure that the directed graph is acyclic and satisfies the strict constraints of the task execution order. Finally, a semantic trajectory chain with timestamps and spatial coordinates is generated, i.e., the power grid operation action sequence.

[0043] In one optional embodiment, in a high-voltage line maintenance scenario, the action unit set includes the "grounding wire connection" action unit. Its operation type field is encoded to obtain a dense vector, with an expected time (standard operation time) of 32 minutes and a normalized result of approximately 0.235. The standard operation location coordinates are (longitude 113.264385, latitude 23.129112, elevation 18 meters), which are projected to form a three-dimensional location vector. An additional time index "2024-06-18 09:32:15" and spatial index node ID "NODE_0447" are added to generate a standardized semantic node data object. Similar processing of other action units yields eight standardized semantic node data objects. Through topology construction, these objects are sequentially connected according to timestamp intervals (12s, 24s, etc.) and spatial displacement (0.0005 degrees, etc.) to form a power grid operation action sequence, representing the complete operation evolution logic.

[0044] In some embodiments, please refer to Figure 2 , Figure 2 This is a schematic diagram of the job execution feature vector generation process according to an embodiment of the method provided in this application. Step S2 determines the job execution feature vector based on the actual location data of personnel, the timing data of operation behavior, the standard location coordinates, and the standard operation time, including steps S21 to S24, each of which is as follows: Step S21: Calculate the positional offset between the actual position data of the personnel and the standard position coordinates, and generate a spatial deviation feature value based on the positional offset; In an optional embodiment, step S21 first acquires the actual location data of the personnel and the standard location coordinates. The actual location data is derived from the spatiotemporal gridding mapping of the personnel positioning heatmap time series data, and the standard location coordinates are derived from the standard operating position coordinates associated with the current action unit in step S14.

[0045] Specifically, based on the equipment number and safety measure information in the task instruction set generated in step S11, a spatiotemporal window partitioning benchmark parameter set is established: the equipment number is mapped to the standard operation location coordinates in the geographic information system, serving as the spatial window reference center point; the operation step timestamps involved in the safety measure information are used as the start and end boundary values ​​of the time window. Timestamp parsing is performed on the personnel positioning heatmap time series data, mapping the sampling point time to the relative time axis position of the standard operation step, and extracting the corresponding centroid coordinate sequence in conjunction with the spatial window reference center point; for operation behavior time series data, for example, operation gesture time series data, frame sequence time parsing is performed, aligning the start and end frame times of the action to the same relative time axis, and mapping the action event coordinates to the spatial window reference. Bidirectional interpolation synchronization processing is performed based on the two time-aligned sequences, ensuring that both types of data have spatial coordinate values ​​at the same time slice, guaranteeing time benchmark consistency. The synchronized time series and spatial offset values ​​are combined into the original signal stream for action execution. Subsequently, the personnel positioning heatmap data in this signal stream is loaded into the spatiotemporal mapping module to establish a unified spatiotemporal grid base index. The raster-based centroid calculator is invoked to divide the location thermal matrix into grid cells according to a preset spatial resolution, and the centroid coordinates are solved in each time slice: in, Using the centroid coordinates, The thermal value weight of the grid cell. The resulting centroid coordinate sequence is reconstructed into a continuous three-dimensional coordinate trajectory. After interpolation to fill in missing slices, it is converted into a spatial path vector, which is the actual location data of the personnel.

[0046] The actual personnel location data and the standard work position coordinates of the current action unit in the semantic trajectory chain are processed to unify the data structure, ensuring coordinate system consistency and dimensional matching. Each centroid coordinate point in the measured path vector is arranged in timestamp order, and its longitude, latitude, and elevation components are extracted to form a three-dimensional measured position vector sequence. The standard work position coordinates are mapped to the same time reference and parsed into a three-dimensional standard position vector sequence. Euclidean distance difference calculation is performed on each time slice to obtain the positional offset between the actual personnel position and the standard position. : in, These represent the longitude, latitude, and elevation components of the measured location vector, respectively. These are the corresponding components of the standard position vector. The distance differences of each time slice are used to construct a spatial deviation curve based on the time series. A moving average filter is used to eliminate high-frequency jitter components, and the filtered curve is normalized to map the deviation to a uniform scale range, outputting the spatial deviation characteristic value. For example, in a transmission line surge arrester replacement task, the equipment number in the task instruction set corresponds to GIS coordinates (120.1580, 30.2551), with a time window from 08:30 to 09:45. The personnel positioning heatmap sampling frequency is 1Hz, with 08:30 as the zero time base; the smart safety helmet gesture data frame rate is 25fps, synchronized to the same zero time base. After bidirectional interpolation and centroid calculation, the actual personnel location data is obtained. The measured coordinates at the 10th second are (120.156789, 30.256321, 15.2), and the standard position coordinates are (120.156700, 30.256200, 15.0). The calculated spatial deviation is approximately 0.2000001 meters. After normalization, the mapped value yields a spatial deviation characteristic value of 0.04.

[0047] Step S22: Calculate the duration deviation rate between the actual duration of the operation behavior time series data and the standard operation time, and generate a time deviation feature value based on the duration deviation rate; In an optional embodiment, step S22 acquires the operation gesture timing data from step S21. The timing data is then processed for action start and end frame detection and duration statistics: the start and end frames of each action unit are identified, the time difference between them is calculated, and the actual time consumed by the current action unit is extracted. Simultaneously, the standard operation time associated with the current action unit, i.e., the expected time baseline, is read from the power grid operation action sequence. The ratio difference between the actual time consumed and the standard operation time is calculated to generate a duration deviation rate, which characterizes the duration deviation rate. The duration deviation rate can be calculated as follows: Among them, the duration deviation rate is the characteristic value of time deviation.

[0048] Step S23: Calculate the matching similarity between the time series of the velocity change rate of the actual personnel location data and the time series of the action switching frequency of the operation behavior time series, and determine the behavior coordination feature value based on the matching similarity; In an optional embodiment, step S23 acquires the actual position data of the personnel and calculates its velocity change rate time series to obtain the spatial movement dynamic characteristic curve of the worker during the execution of the current action unit. It also acquires operation behavior time series data, such as operation gesture time series data, performs inter-frame action switching frequency determination, extracts the start and end points of actions and the number of switching within the same time window, and obtains a time series signal characterizing the frequency of hand operations. The velocity change rate sequence... With action switching frequency sequence Input the cross-correlation analysis module, aligning the sampling points by timestamp index. Use a normalized cross-correlation function to quantify the two at different lag times. The degree of relevance : in, and They are respectively and The mean. The location of the correlation peak. The value reflects the temporal synchronization offset between spatial movement and hand operations. The peak correlation coefficient is used as the matching similarity, i.e., the behavioral coordination feature value. This feature value quantifies the degree of temporal synchronization matching between spatial movement and hand operations for subsequent fusion with spatial deviation feature values ​​and temporal deviation feature values. For example, in a power grid tower maintenance task, the velocity change rate curve is sampled at a frequency of 10Hz, with a maximum change rate of 0.35m / s²; the action switching frequency signal switches an average of 15 times within a 30-second window. Normalized cross-correlation calculations yield... The peak correlation coefficient per second was 0.82, which means the matching similarity was 0.82, and was used as the behavioral collaboration feature value.

[0049] Step S24: Combine the spatial deviation feature value, the time deviation feature value, and the behavioral coordination feature value to obtain the job execution feature vector.

[0050] Obtain the spatial deviation feature value generated in step S21, the temporal deviation feature value generated in step S22, and the behavioral coordination feature value generated in step S23. Perform a combination operation on the deviation features of the above three dimensions to construct a structured data object containing three-dimensional deviation features, thereby obtaining the job execution feature vector.

[0051] In some embodiments, step S3 determines environmental anomaly information based on the acquired environmental monitoring data of power grid operations, including steps S31 to S33, each of which is as follows: Step S31: Extract wind speed, humidity, and temperature sequences from environmental monitoring data; To acquire multimodal environmental monitoring data for anomaly detection, spatial registration and keyframe extraction are performed on the UAV inspection video frame sequence, and outlier filtering is applied to the micro-meteorological sensor array monitoring data to generate multimodal sensing data segments. Specifically, the spatial coordinate reference in the original signal stream of the action execution output in step S21 is used to obtain a three-dimensional position vector as a spatial registration benchmark. Based on this vector, the region of interest is located and each frame is cropped to extract the focal region video frames related to the current action. The inter-frame differential cumulative energy method is used to identify frames whose changes exceed a threshold and extract a set of keyframes. Simultaneously, moving average and median filtering are applied to the micro-meteorological sensor array monitoring data (wind speed, humidity, temperature) to remove outliers. Finally, the keyframes are synchronized with the timestamps of the filtered environmental data and encapsulated into a multimodal sensing data segment. This segment is the required environmental monitoring data for subsequent extraction of wind speed, humidity, and temperature sequences and determination of safety thresholds. The time-series data collected by the micro-meteorological sensor array is parsed from the environmental parameter sequence. Instantaneous wind speed, relative humidity, and local temperature sequences are extracted according to data type identifiers. The original timestamp index is retained for each sequence to ensure temporal consistency in subsequent judgments.

[0052] Step S32: Determine whether each sampled value in the wind speed sequence, humidity sequence, and temperature sequence exceeds the corresponding preset safety threshold; For wind speed sequences, a preset safe wind speed threshold (e.g., 5 m / s) is used; for humidity sequences, a preset safe humidity threshold (e.g., 85% RH) is used; and for temperature sequences, a preset safe temperature threshold (e.g., 40℃) is used. Each sampled value in each sequence is compared to its corresponding safe threshold, and any exceedances are recorded. Simultaneously, a sliding difference method can be used to detect abnormal fluctuations (such as sudden wind speed changes exceeding a set rate of change), and these are also marked as exceeding the threshold.

[0053] Step S33: When any sampled value exceeds the corresponding preset safety threshold, determine the environmental anomaly information; wherein, the environmental anomaly information includes the sampled value and the data type corresponding to the sampled value.

[0054] If any sampled value in the wind speed, humidity, or temperature sequence exceeds its corresponding preset safety threshold, the current environment is determined to be abnormal. The determined environmental anomaly information includes: the specific value exceeding the threshold, the data type of that value (wind speed / humidity / temperature), and the timestamp of that value. If no sampled value exceeds the threshold, the environmental anomaly information is empty, indicating that the current environment is in a safe state. To further identify the abrupt change characteristics of environmental parameters, abrupt change feature extraction is performed on the instantaneous wind speed, relative humidity, and local temperature sequences in the micro-meteorological sensor array monitoring data. A sliding differential threshold detection algorithm is used to identify abnormal fluctuation points exceeding the normal operating threshold, generating an environmental abrupt change marker list characterizing the environmental abrupt change state. This abrupt change marker list can serve as a supplement to the environmental anomaly information and is used to refine the construction of risk constraint vectors.

[0055] In an optional embodiment, during a high-altitude surge arrester replacement operation, the spatial coordinates referenced are longitude 121.4800, latitude 31.2300, and height 15 meters, used to locate the tower high-altitude work platform, generating a 640×480 pixel region of interest identification matrix, with a correlation coefficient threshold of 0.85. The video frame rate is 30fps, and the inter-frame differential energy threshold is... Twelve keyframes were extracted. The micro-meteorological data sampling frequency was 1Hz, with a sliding window of 15 seconds and a median filtering window of 5 seconds, removing three outliers. After alignment, two blurry frames were removed, and the remaining 10 keyframes were encapsulated with stable environmental data into an environmental parameter sequence. The extracted wind speed sequence showed a sample value of 6.2 m / s at time t=120 seconds, exceeding the preset safe wind speed threshold of 5 m / s. Simultaneously, humidity and temperature were normal, thus identifying the environmental anomaly as {Data type: wind speed, Anomaly value: 6.2 m / s, Timestamp: 120s}. Simultaneously, the sliding differential method detected an instantaneous change in wind speed exceeding 5 m / s, generating an environmental abrupt change marker list containing this event.

[0056] In some embodiments, step S4 performs spatial matching between the standard position coordinates corresponding to the currently executing action unit and a preset set of danger zones to determine the spatial risk zone, including steps S41 to S42, each of which is as follows: Step S41: Compare the standard location coordinates with the spatial locations of each hazardous area in the hazardous area set to determine whether the standard location coordinates are located within a hazardous area. The standard position coordinates associated with the current action unit in the power grid operation sequence are analyzed to extract longitude, latitude, and elevation components, forming a three-dimensional position vector. Optionally, a coordinate system consistency transformation can be performed on this three-dimensional position vector (e.g., using a transformation matrix between WGS-84 and the target GIS coordinate system), and the elevation component can be adjusted using a gravity reference datum to ensure matching with the coordinate system of the geographic information system map. The three-dimensional position vector corresponding to the standard position coordinates is input into the geographic information system map query engine to load a pre-stored set of hazardous areas, including but not limited to safe distance zones for live conductors, critical fall zones, and confined space areas. A spatial topological inclusion judgment algorithm is invoked to calculate the geometric overlap relationship between the three-dimensional position vector corresponding to the standard position coordinates and each hazardous area. Specifically, the spatial distance from the standard position coordinates to the center point of each hazardous area can be calculated using Euclidean distance. : in, For the spatial components of standard position coordinates, These are the coordinate components of the center point of the danger zone. This refers to the spatial distance. The calculated spatial distance... Safety radius of the corresponding danger zone The inclusion criterion is used for comparison: like If the standard position coordinates are found to be within the corresponding danger zone, then it is determined that the standard position coordinates are located within the corresponding danger zone.

[0057] Step S42: When there is a dangerous area containing spatial coordinates, the dangerous area is identified as a spatial risk area.

[0058] Based on the comparison results above, hazardous areas containing spatial coordinates are identified as spatial risk areas, and the corresponding feature sets of these spatial risk areas are determined to generate a spatial risk area feature set. For each hazardous area type, if the inclusion determination is true, a corresponding proximity identifier (such as "proximity identifier for live conductors," "proximity identifier for falls from heights," or "proximity identifier for confined spaces") is added to the spatial risk area feature set. For example, the actual spatial coordinates of a personnel's location at a certain work node are longitude 121.4737, latitude 31.2304, and elevation 15.2 meters. The geographic information system retrieves the coordinates of the center point of the safe distance zone for live conductors (121.4730, 31.2300, 15.0), with a safe radius of 2.0 meters, and calculates the spatial distance. If the distance is less than 2.0 meters, it is considered true and a "near charged body" label is added. Simultaneously, if the distance between the center point of the high-altitude fall threshold zone and the location is found to be 1.5 meters, which is less than its radius of 1.8 meters, a "near high-altitude fall" label is added. Finally, the spatial risk area is determined to include the safe distance zone from charged bodies and the high-altitude fall threshold zone, and the corresponding spatial risk area feature set is output.

[0059] In some embodiments, step S5 generates a risk constraint vector based on spatial risk areas and environmental anomaly information, including steps S51 to S53, each of which is detailed below: Step S51: Generate a spatial risk weight matrix based on the risk type corresponding to the spatial risk area; Obtain a set of spatial risk area features, which includes identified spatial risk area markers, such as markers for proximity to charged objects, proximity to falls from heights, and proximity to confined spaces. Perform type encoding conversion on the spatial risk area feature set, mapping different risk types to quantitative spatial risk weight components, and generate a spatial risk weight matrix based on the risk intensity curve defined by the risk type level.

[0060] Step S52: Generate an environmental risk weight matrix based on the types and degrees of anomalies of environmental parameters contained in the environmental anomaly information; Based on the environmental anomaly information determined in step S33 and the environmental mutation marker list generated in step S34, the degree of anomaly of each type of environmental parameter (wind speed, humidity, temperature) in the environmental anomaly information is encoded, and the sudden changes in instantaneous wind speed, humidity jump, local temperature rise, etc. are mapped into environmental risk weight components. An environmental risk weight matrix is ​​generated based on the coupling relationship between anomaly magnitude and duration.

[0061] Step S53: Perform weighted fusion of the spatial risk weight matrix and the environmental risk weight matrix to generate a risk constraint vector.

[0062] The spatial risk weight matrix and the environmental risk weight matrix undergo feature dimension alignment and normalization to ensure dimensional consistency. The spatial risk weight matrix may contain weight components corresponding to multiple spatial risk types, such as proximity to charged objects or proximity to high-altitude falls. These components must be merged into a single spatial risk intensity component using predefined rules, such as taking the maximum value, weighted average, or maximum value. Similarly, the abnormal weight components of multiple environmental parameters in the environmental risk weight matrix, such as sudden changes in wind speed, sudden increases in humidity, and local temperature rises, are merged into a single environmental risk intensity component. A weighted association rule model is adopted, coupling the spatial risk weight matrix and the environmental risk weight matrix according to the risk correlation coefficient. The weight coefficient can be determined by the maximum correlation and minimum redundancy analysis of historical risk samples. The comprehensive risk intensity is calculated through the risk correlation formula. in, For spatial constraint weights, For the spatial risk intensity component, For environmental constraint weights, This represents the environmental risk intensity component. A nonlinear risk intensity mapping is performed on each element of the coupling matrix, using the Sigmoid function to limit the risk value to the interval between 0 and 1, highlighting the sensitive response of high-risk segments. A structured multidimensional risk constraint feature vector is output as the comprehensive spatial-environmental risk representation of the current work node. The identifiers and their weighted components corresponding to each spatial risk type, the identifiers and their weighted components corresponding to each environmental anomaly type, the risk intensity timestamp index, and the unique identifier of the action unit are structurally encapsulated to form a risk constraint vector. This risk constraint vector is output to the model scheduling layer for subsequent joint confidence judgment logic calls.

[0063] In an optional embodiment, in a high-altitude live-line work scenario, the spatial risk area feature set includes a live conductor safety distance zone identifier coded as 0.8 and a high-altitude fall threshold zone identifier coded as 0.6. Optionally, the spatial risk intensity component is obtained by fusing the components using the maximum value rule. The environmental abrupt change marker list includes instantaneous wind speed abrupt changes coded as 0.7 and local temperature rise coded as 0.5. Optionally, the maximum value rule is used for fusion to obtain the environmental risk intensity component. Preset spatial weights Environmental weight Then the risk intensity value for: The quantification of risk intensity is obtained by mapping using the Sigmoid function. : This value is the quantitative result of the risk intensity of this operation node.

[0064] In some embodiments, step S6, determining the comprehensive risk value of the current power grid operation based on the operation execution feature vector and risk constraint vector, includes steps S61 to S64, each step as follows: Step S61: Calculate the joint confidence level based on the risk constraint vector and the timestamp information of the currently executing action unit; identify risk types whose joint confidence level exceeds the preset activation threshold as activation risk types; Specifically, the risk intensity corresponding to each spatial risk region (e.g., risk intensity of 0.8 for the safe distance zone from charged objects, risk intensity of 0.6 for the high-altitude fall threshold zone) and the anomaly degree code corresponding to each environmental anomaly parameter (e.g., wind speed mutation code 0.7, local temperature rise code 0.5) are extracted from the risk constraint vector generated in step S5. Based on the timestamp information of the currently executing action unit, the spatial risk intensity and environmental anomaly degree are normalized to obtain the spatial proximity index. and time synchronization index For each candidate risk type For risks such as falls from heights and distance risks from charged objects, the original risk trigger probability score for this type of risk is calculated using a product-based fusion method. in Spatial weighting coefficient, For time weighting coefficients, These are the coefficients of the coupling terms. The joint confidence score is then obtained through a Sigmoid function mapping. Will With preset activation threshold (e.g., 0.75) Comparison. If the joint confidence level is greater than or equal to the activation threshold, the risk type is determined to be an activated risk type; otherwise, it is not activated.

[0065] Step S62: For each activated risk type, call the risk assessment model corresponding to the activated risk type from the risk model library, and obtain the feature weight vector built into the risk assessment model; For each activated risk type, the risk assessment model corresponding to that activated risk type is retrieved from the risk model library, and the feature weight vector built into the risk assessment model is obtained. For example, if an activated risk type is "charged body distance risk", the corresponding risk assessment model is retrieved from the risk model library based on its type identifier, and the feature weight vector is extracted from the initialization parameter configuration set of that model. This vector contains weight coefficients for three dimensions: spatial bias, temporal bias, and behavioral coordination.

[0066] Step S63: The spatial deviation feature value, the temporal deviation feature value, and the behavioral coordination feature value are weighted and summed according to the feature weight vector to obtain the first risk value of the activation risk type; Obtain spatial deviation score Time Deviation Scoring Behavioral coordination score For each activation risk type, its feature weight vector is... Perform a dot product operation with the three scores: Obtain the first risk value for this risk type. .

[0067] Step S64: Weight and merge the first risk values ​​of all activated risk types according to the risk intensity corresponding to each activated risk type to obtain the comprehensive risk value.

[0068] The first risk value of all activated risk types is determined according to the risk intensity corresponding to each risk type in the risk constraint vector. The weighted fusion is performed to obtain the comprehensive risk value. : The denominator is the sum of the risk intensities of each activation risk type. Fusion Results This represents the overall risk value for the current power grid operation. If no risk type is activated, the overall risk value can be set to the default value.

[0069] In one optional embodiment, in a certain high-altitude live-line operation scenario, if both "distance risk to live conductor" (risk intensity R=0.98, first risk value V=0.76) and "fall risk from height" (risk intensity R=0.92, first risk value V=0.75) are activated simultaneously, then the comprehensive risk value = (0.98×0.76+0.92×0.75) / (0.98+0.92)≈0.755. This value is output as the comprehensive risk value of the current power grid operation and is used for subsequent risk warning and operation scheduling.

[0070] In some embodiments, the risk assessment method for power grid operations of this application further includes step S7, which includes steps S71 to S74, and the specific details of each step are as follows: Step S71: Obtain the historical risk trend data sequence from the power grid operation action sequence; generate a benchmark risk evolution curve based on the historical risk trend data sequence; Specifically, historical risk trend data sequences of the current action unit and its adjacent nodes are extracted from the power grid operation sequence. A sliding time-series window is constructed, and a local risk time-series window containing the current node and a preset number of neighboring nodes (e.g., 3 nodes before and after) is built based on timestamp alignment logic, forming a standardized risk data set for consistency comparison. Linear interpolation and smoothing filtering operations are performed on the historical risk trend data sequences in this standardized risk data set to eliminate minor fluctuations caused by sensor jitter, generating a benchmark risk evolution curve that characterizes the expected risk change pattern under the operation evolution logic. For example, the local risk time-series window of a power grid maintenance task contains 7 time slices, and the historical risk confidence sequence is [0.15, 0.18, 0.20, 0.22, 0.21, 0.24, 0.23]. After interpolation and filtering, a smooth benchmark risk evolution curve is generated, reflecting the normal upward trend of risk values ​​during the operation.

[0071] Step S72: Based on the baseline risk evolution curve and the comprehensive risk value, determine the deviation magnitude and rate of change difference of the current execution action unit; based on the deviation magnitude and rate of change difference, determine the risk abnormal deviation index; The comprehensive risk value output by the current action unit is compared with the value of the baseline risk evolution curve at the current time slice, and the difference between the two is calculated as the deviation magnitude. At the same time, the difference between the rate of change of the current risk value and the risk value of the previous node and the corresponding rate of change of the baseline curve is calculated as the rate of change difference. The deviation magnitude and the rate of change difference are combined (e.g., by weighted summation or product) to generate a single risk anomaly deviation index that quantifies the degree to which the current risk judgment deviates from the normal evolution trajectory.

[0072] For example, the overall risk value of the current action unit is 0.82, the baseline risk evolution curve value at the current time slice is 0.55, and the deviation is 0.27; the risk value of the previous node is 0.60, the change rate of the current node is (0.82-0.60) / 0.60≈0.367, the corresponding change rate of the baseline curve is 0.08, and the difference in change rates is 0.287. By weighting the deviation and the difference in change rates equally, we obtain the risk anomaly deviation index δ=0.5×0.27+0.5×0.287=0.2785.

[0073] Step S73: Calculate the risk value change range between the currently executed action unit and the adjacent action units, and determine whether the comprehensive risk value is an outlier based on the risk anomaly deviation index and the risk value change range; wherein, when the risk anomaly deviation index exceeds the preset deviation threshold and the risk value change range is less than the preset change tolerance, the comprehensive risk value is determined to be an outlier. The historical risk confidence values ​​of the current action unit's preceding and following adjacent nodes are extracted from the local risk time sequence window. The absolute difference or gradient magnitude between the current node's risk value and the risk values ​​of adjacent nodes is calculated as the risk value change magnitude. A preset deviation threshold and change tolerance are set. When the risk anomaly deviation index exceeds the deviation threshold and the risk value change magnitude is less than the change tolerance, it is judged as an outlier (i.e., isolated high risk).

[0074] For example, if the overall risk value of the current action unit is 0.82, the risk value of the previous node is 0.60, and the risk value of the next node is 0.63, then the magnitude of the risk value change is the maximum of the changes before and after: max(|0.82-0.60|, |0.82-0.63|) = 0.22. The preset deviation threshold is 0.25, and the preset tolerance for change is 0.15. Since the risk anomaly deviation index δ = 0.2785 exceeds the deviation threshold of 0.25, but the magnitude of the risk value change of 0.22 is greater than the tolerance for change of 0.15, the condition of "magnitude of change less than tolerance for change" is not met, so the overall risk value is determined not to be an outlier. If in another scenario the risk anomaly deviation index is 0.35, and the magnitude of the risk value change is 0.10 (less than 0.15), then it is determined to be an outlier. This joint discrimination mechanism effectively distinguishes between real risk mutations and false peaks caused by sensor noise or algorithm misjudgment.

[0075] Step S74: If the comprehensive risk value is an outlier, the electronic work order text fragment and the actual location data of the personnel corresponding to the current execution action unit are re-parsed to correct the power grid operation action sequence.

[0076] When the comprehensive risk value is determined to be an outlier, a secondary semantic re-parsing process is triggered. First, the isolated high-risk judgment markers in the power grid operation action sequence integrity verification results are logically parsed to extract the control enable signal that triggers the secondary semantic re-parsing process, generating a re-parsing task instruction set containing the index of nodes to be corrected. Then, based on the index of nodes to be corrected in this instruction set, the electronic work order text sequence fragments and personnel location heatmap time-series data corresponding to the original multi-source heterogeneous data stream are extracted backtrackingly to construct a local spatiotemporal context augmented dataset. Next, a domain-aware sequence encoder is used to perform fine-grained action segmentation and reconstruction processing on this augmented dataset, re-identifying missing action units and correcting drifting spatial coordinates to generate a corrected local power grid operation action sequence. Finally, this corrected power grid operation action sub-sequence is mapped to the global power grid operation action sequence to replace the original abnormal nodes, updating the association between the action execution quality label and the risk constraint vector, thereby generating intermediate data for global environmental risk assessment that has undergone consistency calibration, completing the correction of the power grid operation action sequence.

[0077] In one optional embodiment, during a high-altitude operation, transient sensor interference caused the overall risk value of a certain node to be incorrectly judged as 0.85 (the actual value should be 0.45). After joint verification of deviation indicators and change amplitude, the risk anomaly deviation indicator was 0.35 (exceeding the deviation threshold of 0.25), and the risk value change amplitude was 0.08 (less than the change tolerance of 0.15), thus identifying it as an outlier. The system backtracked the electronic work order segment ("climbing the tower - installing the grounding wire") and personnel positioning heatmap data corresponding to the node's time window. Re-analysis revealed that the actual action sequence was consistent with the original power grid operation action sequence, with only sensor noise causing the falsely high risk. After correction, the corrected local power grid operation action sub-sequence replaced the original abnormal node, adjusting the node's risk value to 0.46, and updating the global power grid operation action sequence, thus avoiding false alarms and ensuring that the risk assessment results and the operation evolution logic maintained a closed-loop consistency.

[0078] In some embodiments, one example of the power grid operation risk assessment system provided in this application includes: the power grid operation risk assessment system is used to perform the power grid operation risk assessment method as described in this application.

[0079] It is understood that the above system embodiments correspond to the method embodiments of this application, and can implement the power grid operation risk assessment method provided by any of the above method embodiments of this application.

[0080] Based on the above-described method embodiments, another embodiment of this application provides a computer-readable storage medium, including: a stored computer program, wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute the risk assessment method for power grid operations as described in any of the above-described method embodiments of this application.

[0081] The modules / units integrated in the device / terminal equipment, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0082] The above are preferred embodiments of this application. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles of this application, and these improvements and modifications are also considered to be within the scope of protection of this application.

Claims

1. A risk assessment method for power grid operations, characterized in that, include: Obtain the electronic work order text for power grid operations; The electronic work order text is parsed to generate several action units. Each action unit is then associated with standard location coordinates obtained from the geographic information system and a preset standard operation time, and integrated into a power grid operation action sequence. Extract the currently executing action unit, the corresponding standard position coordinates, and the standard operation time from the power grid operation action sequence, and obtain the actual position data and operation behavior timing data of the personnel in the currently executing action unit; determine the operation execution feature vector based on the actual position data, the operation behavior timing data, the standard position coordinates, and the standard operation time; Based on the acquired environmental monitoring data of power grid operations, environmental anomaly information is determined; spatial risk areas are determined by spatial matching between the standard position coordinates corresponding to the currently executing action unit and a preset set of dangerous areas. Based on the spatial risk area and the environmental anomaly information, a risk constraint vector is generated; The comprehensive risk value of the current power grid operation is determined based on the operation execution feature vector and the risk constraint vector.

2. The risk assessment method for power grid operations according to claim 1, characterized in that, The process of determining the job execution feature vector based on the actual location data of the personnel, the timing data of the operation behavior, the standard location coordinates, and the standard operation time includes: Calculate the positional offset between the actual position data of the personnel and the standard position coordinates, and generate a spatial deviation feature value based on the positional offset; Calculate the duration deviation rate between the actual duration of the operation behavior timing data and the standard operation time, and generate a time deviation feature value based on the duration deviation rate; Calculate the matching similarity between the time series of the velocity change rate of the actual personnel location data and the time series of the action switching frequency of the operation behavior time series data, and determine the behavior coordination feature value based on the matching similarity; The spatial deviation feature value, the temporal deviation feature value, and the behavioral coordination feature value are combined to obtain the job execution feature vector.

3. The risk assessment method for power grid operations according to claim 1, characterized in that, The process involves parsing the electronic work order text to generate several action units. Each action unit is then associated with standard location coordinates obtained from a geographic information system and a preset standard operation time, and integrated into a power grid operation action sequence, including: The electronic work order text is parsed using natural language to extract structured work instructions and generate an initial work instruction set; Based on the initial set of operation instructions, a time-series logic completion is performed using a preset power grid operation process knowledge graph to generate an enhanced operation instruction sequence. The enhanced operation instruction sequence is broken down into several indivisible action units; Obtain the standard location coordinates and preset standard operation time of each action unit in the geographic information system, and associate the standard location coordinates and the standard operation time with the corresponding action unit; integrate the action units into a power grid operation action sequence according to the power grid operation execution order.

4. The risk assessment method for power grid operations according to claim 1, characterized in that, The step of generating a risk constraint vector based on the spatial risk region and the environmental anomaly information includes: A spatial risk weight matrix is ​​generated based on the risk type corresponding to the spatial risk region; An environmental risk weight matrix is ​​generated based on the types and degrees of environmental parameters contained in the environmental anomaly information. The spatial risk weight matrix and the environmental risk weight matrix are weighted and fused to generate the risk constraint vector.

5. The risk assessment method for power grid operations according to claim 2, characterized in that, Determining the comprehensive risk value of the current power grid operation based on the operation execution feature vector and the risk constraint vector includes: Based on the risk constraint vector and the timestamp information of the currently executing action unit, a joint confidence score is calculated; risk types whose joint confidence scores exceed a preset activation threshold are identified as activation risk types. For each activated risk type, the risk assessment model corresponding to the activated risk type is called from the risk model library, and the feature weight vector built into the risk assessment model is obtained. The spatial deviation feature value, the temporal deviation feature value, and the behavioral coordination feature value are weighted and summed according to the feature weight vector to obtain the first risk value of the activation risk type; The first risk values ​​of all the activation risk types are weighted and fused according to the risk intensity corresponding to each activation risk type to obtain the comprehensive risk value.

6. The risk assessment method for power grid operations according to claim 1, characterized in that, The step of determining environmental anomaly information based on the acquired environmental monitoring data of power grid operations includes: Extract wind speed, humidity, and temperature sequences from the environmental monitoring data; Determine whether each sampled value in the wind speed sequence, the humidity sequence, and the temperature sequence exceeds the corresponding preset safety threshold; When any of the sampled values ​​exceeds the corresponding preset safety threshold, environmental anomaly information is determined; wherein, the environmental anomaly information includes the sampled value and the data type corresponding to the sampled value.

7. The risk assessment method for power grid operations according to claim 1, characterized in that, The step of determining spatial risk areas by spatially matching the standard position coordinates corresponding to the currently executing action unit with a preset set of danger zones includes: The standard location coordinates are compared with the spatial locations of each danger zone in the set of danger zones to determine whether the standard location coordinates are located within the danger zone. When a dangerous area containing the standard location coordinates exists, the dangerous area is identified as a spatial risk area.

8. The risk assessment method for power grid operations according to claim 1, characterized in that, Also includes: Obtain the historical risk trend data sequence from the power grid operation sequence; Generate a baseline risk evolution curve based on historical risk trend data sequences; Based on the baseline risk evolution curve and the comprehensive risk value, determine the deviation magnitude and rate of change difference of the currently executed action unit; based on the deviation magnitude and the rate of change difference, determine the risk abnormal deviation index; Calculate the risk value change range between the currently executing action unit and the adjacent action units, and determine whether the comprehensive risk value is an outlier based on the risk anomaly deviation index and the risk value change range; wherein, when the risk anomaly deviation index exceeds a preset deviation threshold and the risk value change range is less than a preset change tolerance, the comprehensive risk value is determined to be an outlier. If the overall risk value is an outlier, the electronic work order text fragment and the actual location data of the personnel corresponding to the currently executing action unit are re-parsed to correct the power grid operation action sequence.

9. A risk assessment system for power grid operations, characterized in that, The power grid operation risk assessment system is used to perform the power grid operation risk assessment method as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, include: A stored computer program, wherein, when the computer program is executed, the device containing the computer-readable storage medium is controlled to perform the risk assessment method for power grid operations as described in any one of claims 1-8.

Images