Distribution line first-aid repair operation risk assessment method, system and equipment based on multi-source sensing data, and medium

Abstract

The invention discloses a distribution line first-aid repair operation risk assessment method, system and equipment based on multi-source perception data and a medium, and belongs to the technical field of first-aid repair operation risk assessment. The method comprises the steps that multi-source data of perception data, operation data and environment data are collected and subjected to time-space alignment, quality verification and missing repair are carried out, and a first-aid repair result is obtained; generating fusion features with credibility weights; constructing a dynamic graph structure taking the space nodes and the participation roles as vertexes in the operation scene; a physical consistency index constraint used for reflecting an operation safety boundary is introduced into the graph structure, and an interpretable risk characterization mechanism is formed; carrying out distribution robustness and interval calibration on the risk characterization result; carrying out anti-fact simulation on the plurality of candidate coping schemes based on digital twin modeling, and triggering the sensing equipment to execute active supplementary mining operation according to an information value maximization strategy; sensing and simulation results are fed back to the model for updating and iteration. According to the invention, intelligent risk identification and adaptive control of the whole operation process of the power transmission line are realized.

Description

Technical Field

[0001] This invention relates to the field of emergency repair operation risk assessment technology, specifically to a method, system, equipment, and medium for assessing the risk of power line emergency repair operations based on multi-source sensing data. Background Technology

[0002] In existing technologies, some methods for risk assessment of power distribution line emergency repair operations analyze meteorological monitoring data, historical accident records, or geographic information data, which can reflect the impact of the working environment on repair safety to a certain extent. However, these technologies mostly rely on single or limited data sources and lack comprehensive utilization of multi-source sensing information, making it difficult for the assessment results to fully reflect the actual risk situation on site. Furthermore, existing methods are generally based on historical statistical data, failing to respond in real time to sudden environmental changes, secondary equipment failures, or abnormal personnel conditions during operations, resulting in assessment lag. In addition, some assessment models still use experience-based scoring or linear weighting, making it difficult to reveal the coupling relationships and temporal evolution characteristics between multiple factors, resulting in insufficient prediction and early warning capabilities and difficulty in timely alerting of potential risks. Moreover, existing technologies often build models for specific scenarios, with limited adaptability and universality, failing to meet the repair needs under different geographical environments and climatic conditions. Summary of the Invention

[0003] In view of the above-mentioned problems, the present invention provides a method, system, equipment and medium for risk assessment of emergency repair operations of power distribution lines based on multi-source sensing data.

[0004] Therefore, the technical problem solved by this invention is the defects of existing risk assessment methods for emergency repair of power distribution lines, such as single data source, insufficient information dimensions, delayed risk assessment, crude model, and lack of predictability and adaptability.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a risk assessment method for power line emergency repair operations based on multi-source sensing data, comprising: collecting and spatiotemporally aligning multi-source data including sensing data, operational data, and environmental data; performing quality verification and missing data repair; generating fusion features with credibility weights; constructing a dynamic graph structure in the operational scenario with spatial nodes and participating roles as vertices, with graph edges established based on structural connections, spatial proximity, and operational process dependencies; introducing physical consistency index constraints to reflect operational safety boundaries into the graph structure, constructing causal propagation channels, distinguishing between external disturbance paths and internal state paths, and forming an interpretable risk characterization mechanism; implementing split-bar and interval calibration on the risk characterization results to obtain uncertainty intervals that meet coverage targets, and adaptively adjusting the warning threshold based on state feedback; performing counterfactual simulations of multiple candidate response schemes based on digital twin modeling, and triggering the sensing equipment to perform active supplementary data collection operations based on the information value maximization strategy; outputting the risk level, dominant factors, and expected trigger time, and feeding the sensing and simulation results back to the model for updates and iterations.

[0006] As a preferred embodiment of the risk assessment method for power line emergency repair operations based on multi-source sensing data described in this invention, the following steps are taken: The multi-source data (collecting and spatiotemporally aligning sensing data, operational data, and environmental data), undergoing quality verification and missing data repair, and generating fusion features with credibility weights, include: synchronizing the multi-source data based on a unified operational time benchmark; aligning asynchronous data at the event level by identifying cross-modal key event anchor points to form spatiotemporally consistent data segments; correcting data segments that fail to identify anchor points based on their time offset from aligned segments; performing consistency checks on the aligned data segments to identify anomalous segments; supplementing anomalous segments using spatiotemporal interpolation of neighboring data, and supplementing segments that cannot be interpolated using a generative model; and calculating the credibility weight of each data segment based on source reliability, freshness, and consistency to generate fusion features.

[0007] As a preferred embodiment of the risk assessment method for power distribution line emergency repair operations based on multi-source sensing data described in this invention, the step of constructing a dynamic graph structure with spatial nodes and participating roles as vertices in the operation scenario, and establishing graph edges based on structural connectivity, spatial proximity, and operational process dependencies, includes: identifying spatial nodes and participating role nodes in the operation scenario, constructing a vertex set of the dynamic graph; setting structural connection edges based on facility topology to represent physical connections; setting spatial proximity edges based on geographic spatial distance to represent location associations; setting operational process edges based on the order of the operation process to represent process dependencies; setting initial edge weights for the graph edges, and periodically updating them according to environmental conditions and operational behaviors.

[0008] As a preferred embodiment of the risk assessment method for power distribution line emergency repair operations based on multi-source sensing data described in this invention, the step of introducing physical consistency index constraints to reflect the safety boundary of the operation in the graph structure, constructing causal propagation channels, distinguishing external disturbance paths from internal state paths, and forming an interpretable risk characterization mechanism includes: setting physical consistency indices to constrain propagation behavior in the graph structure to reflect the safety boundary of the operation; performing propagation suppression processing on graph edges that do not meet the physical consistency indices; identifying external disturbance paths and internal state paths according to data source attributes; setting the propagation direction of external disturbance paths, propagating only to internal state nodes and risk nodes; setting the propagation range of internal state paths, limiting propagation to equipment nodes, work unit nodes, and personnel nodes; and completing the construction of the risk characterization mechanism based on the paths and boundaries.

[0009] As a preferred embodiment of the risk assessment method for power line emergency repair operations based on multi-source sensing data described in this invention, the following steps are included: implementing robust and interval calibration on the risk characterization results to obtain an uncertainty interval that meets the coverage target, and adaptively adjusting the warning threshold based on state feedback. This includes: after generating fused features and completing graph structure construction, grouping the output results of risk nodes in the graph structure risk characterization according to the category of the source data; applying a robust modeling method based on minimum performance guarantees to each group of risk nodes so that the most extreme node in each group can still be identified; calculating structural and distance indicators based on the temporal stability and distribution deviation of each fused feature segment; when any indicator exceeds the preset judgment range, marking the corresponding fused feature segment as data outside the training distribution; expanding the interval prediction range of the marked fused feature segment for risk nodes in the graph structure and resetting the upper and lower limit positions; the upper and lower limits are determined by a pre-set coverage target, which is adjusted according to the risk level; after the uncertainty interval is expanded, increasing the alarm level of the corresponding risk node in the graph structure by one level according to the preset level mapping relationship.

[0010] This preferred solution groups the sources of fused features and introduces a robust processing method based on minimum performance guarantees into the output of risk nodes in the graph structure, which can fully preserve the risk expression of extreme samples. By combining the stability of fused features in the time dimension and the degree of distribution deviation, samples outside the distribution are marked, and the range of uncertainty interval is expanded accordingly. This allows the warning level adjustment to be based on the actual deviation behavior of the segment, enabling alarm adjustment with a greater ability to perceive distribution boundaries in the graph structure.

[0011] As a preferred embodiment of the risk assessment method for power line emergency repair operations based on multi-source sensing data described in this invention, the method involves: performing counterfactual simulations of multiple candidate response schemes based on digital twin modeling, and triggering the sensing equipment to perform proactive supplementary data collection based on an information value maximization strategy. This includes: after constructing the graph structure and determining the uncertainty interval, constructing a digital twin model based on the current graph node status and fusion features. The digital twin model covers the operation path, time period, personnel, and equipment information in the graph structure. Within the digital twin model, intervention operations are sequentially applied, including operation path adjustment, operation time change, personnel replacement, operation speed setting, and operation interruption. For each intervention operation, calculation is performed. The system considers the changes in the state of risk nodes caused by the graph structure, the duration of cross-node operations, and the required number of equipment and personnel resources. Based on set weight parameters, intervention operations are weighted and scored using multiple indicators, and operation paths that conflict with constraint edges in the graph structure are eliminated. When multiple intervention operations have the same score, the one with the highest correlation to the dominant factor corresponding to the current risk node and involving the fewest roles is selected. When the credibility of a data segment in the fused features is lower than a threshold, and the segment is located in a high-risk area of ​​the graph structure, an active supplementary data collection operation is triggered at the corresponding location of the sensing device. Whether the sensing device performs supplementary data collection is determined by the alarm level, the remaining available sensing resources, and the weather label of the area where the graph node is located.

[0012] This preferred solution constructs a digital twin model covering operation paths, time periods, personnel, and equipment. It can perform path-level evaluation of multiple intervention operations in a graph structure and identify key difference paths based on multi-indicator scoring. When the credibility of the fused feature fragments in the graph structure is low and the node risk level is high, the sensing device is triggered to locate and perform supplementary sampling to ensure that only candidate paths with complete sensing and intervention adaptation relationships are retained in the graph structure.

[0013] As a preferred embodiment of the risk assessment method for power line emergency repair operations based on multi-source sensing data described in this invention, the following steps are taken: The output of risk level, dominant factor, and expected trigger time feeds the sensing and simulation results back to the model for updates and iterations. This includes: based on the marked risk node states in the graph structure, outputting the risk level for the current time period and recording the corresponding fused feature positions; in the graph structure, calculating the attention allocation score and the risk value change after disturbance input for each node feature path; disturbance processing, based on stable data within a historical time window, applying a small increment to each input in the fused features and calculating the risk level change; when a feature shows a certain impact on wind speed in three consecutive time windows... When the risk level has a significant impact, the marker feature is the dominant factor of the corresponding node in the current graph structure; the criteria for determining the dominant factor comes from the historical graph node replay data and is updated according to a preset period; when the weather label in the fused features indicates a strong external disturbance, the threshold for determining the dominant factor is lowered and the interval prediction range of the corresponding node is expanded; the risk level, dominant factor marker, expected trigger time, and intervention execution result of each node in the current graph structure are uniformly written into the status record table of the graph structure; for intervention schemes identified as failed in the digital twin model, including paths that violate the physical boundaries of the operation, combinations of operation nodes with incorrect order, and execution paths with mismatched role qualifications, all are recorded in the exclusion subgraph of the graph structure.

[0014] This preferred scheme can identify the direct impact characteristics of risk level changes by perturbing the node characteristic paths in the graph structure and combining them with stable inputs within the historical window. The dominant factor is marked with a continuous time window. When the change of the dominant factor encounters a strong external disturbance, the judgment threshold is adjusted. At the same time, the node status, dominant factor and contingency plan results within the period are written into the graph structure status record area, and the failed intervention path is written into the exclusion subgraph. This can avoid the repeated generation of invalid paths during the evaluation and enhance the causal identification integrity of the graph structure in subsequent updates and iterations.

[0015] This invention provides a risk assessment system for emergency repair operations of power distribution lines based on multi-source sensing data.

[0016] To address the aforementioned technical problems, this invention provides the following technical solution: a risk assessment system for power line emergency repair operations based on multi-source sensing data, comprising: a data fusion module, a graph structure construction module, a risk modeling module, an uncertainty control module, a supplementary data collection module, and an output module; the data fusion module is used to collect and spatiotemporally align multi-source data including sensing data, operational data, and environmental data, perform quality verification and missing data repair, and generate fusion features with credibility weights; the graph structure construction module is used to construct a dynamic graph structure in the operational scenario with spatial nodes and participating roles as vertices, and the graph edges are established based on structural connections, spatial proximity, and operational process dependencies; the risk modeling module is used to... A physical consistency index constraint reflecting the operational safety boundary is introduced, and a causal propagation channel is constructed to distinguish between external disturbance paths and internal state paths, forming an interpretable risk characterization mechanism. The uncertainty control module is used to perform sub-bar and interval calibration on the risk characterization results to obtain an uncertainty interval that meets the coverage target, and adaptively adjusts the warning threshold based on state feedback. The supplementary sampling module is used to perform counterfactual simulation of multiple candidate response schemes based on digital twin modeling, and triggers the sensing device to perform active supplementary sampling operation according to the information value maximization strategy. The output module is used to output the risk level, dominant factor, and expected trigger time, and feeds the sensing and simulation results back to the model for updating and iteration.

[0017] The present invention provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the method for risk assessment of power distribution line emergency repair operations based on multi-source sensing data.

[0018] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the method for risk assessment of power distribution line emergency repair operations based on multi-source sensing data.

[0019] The beneficial effects of this invention are as follows: This invention achieves intelligent risk identification and adaptive control throughout the entire process of power transmission line operations by constructing a full-chain architecture that connects multi-source trusted alignment, dynamic spatiotemporal graphs, causal constraint propagation, uncertainty calibration, digital twin counterfactual, and proactive perception closed loop. First, it significantly improves the authenticity and alignability of multi-source data through causal anchors, sensor fingerprints, and triple consistency quality control. Second, it introduces physical safety channels and differentiable rule layers, ensuring that risk propagation is simultaneously constrained by structure, physics, and procedures, eliminating false compliance and abnormal drift. Third, it constructs interpretable uncertainty intervals using sub-Bruker bars and interval calibration, avoiding threshold distortion. Fourth, it utilizes digital twins and shadow cost surfaces to screen optimal contingency plans in a virtual environment, achieving an optimal balance between risk reduction and resource utilization. Fifth, through an information value-driven proactive perception mechanism, it forms a closed-loop supplementary data collection and dynamic correction, thereby achieving a unified optimal balance between risk, cost, and enforceability. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 The above is a flowchart of a risk assessment method for emergency repair operations of power distribution lines based on multi-source sensing data, provided as an embodiment of the present invention. Detailed Implementation

[0022] To make the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0023] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides a method for risk assessment of emergency repair operations on power distribution lines based on multi-source sensing data, including: S1. Collect and align multi-source data, including spatiotemporal sensing data, operational data, and environmental data, perform quality verification and missing data repair, and generate fusion features with credibility weights.

[0024] S2. Construct a dynamic graph structure in the work scenario with spatial nodes and participating roles as vertices. The graph edges are established based on structural connections, spatial proximity, and operational process dependencies.

[0025] S3. Introduce physical consistency index constraints to reflect the safety boundary of the operation into the graph structure, construct causal propagation channels, distinguish between external disturbance paths and internal state paths, and form an interpretable risk characterization mechanism.

[0026] S4. Perform sub-bar and interval calibration on the risk characterization results to obtain the uncertainty interval that meets the coverage target, and adaptively adjust the early warning threshold according to the state feedback.

[0027] S5. Based on digital twin modeling, counterfactual simulations are performed on multiple candidate response schemes, and the sensing devices are triggered to perform active supplementary data collection based on the information value maximization strategy.

[0028] S6 outputs the risk level, dominant factors, and expected trigger time, and feeds the perception and simulation results back to the model for updates and iterations.

[0029] Through steps S1-S6 described in this embodiment, a comprehensive risk assessment method covering data acquisition, structural modeling, risk propagation, judgment output, and closed-loop update is formed, encompassing multi-source data fusion, dynamic graph structure construction, risk identification and assessment, intervention plan simulation and information triggering, as well as result writing and risk iteration in emergency repair scenarios. Each processing path involved is based on fusion features and graph structure status, with clear input data paths, processing logic structures, and output node expression methods. It can provide risk identification and intervention support for dynamic operation processes in scenarios where data perception at the emergency repair site is complex, process paths are unstable, and risk influencing factors are varied.

[0030] Example 2, an embodiment of the present invention, provides a method for risk assessment of power line emergency repair operations based on multi-source sensing data, based on the previous embodiment, including: In this embodiment, the fusion feature in step S1 can be achieved by collecting multi-source data, including sensing data, operational data, and environmental data, and synchronizing them with a unified operational time base; performing event-level alignment on asynchronous segments from different data sources based on cross-modal key event anchors; correcting data segments without identified anchors based on their time offset from aligned segments; performing consistency detection on aligned segments to identify abnormal segments, and using spatiotemporal interpolation of neighboring data to fill in the abnormal segments, supplementing them with a generative model when interpolation is not possible; and calculating source reliability, freshness, and consistency to form credibility weights, and using the weighted expression of all processed segments as the fusion feature.

[0031] In one alternative implementation, the fusion features can also use the output of fixed sensing units deployed in the work area as the main data source and use work log events as time alignment anchors; for missing segments, stable window data of spatially nearest nodes are used to fill in the gaps first; if the gap span exceeds the length of the stable window, historical segments in the same area are used as the basis for extrapolation; the calculation of the credibility weight is dominated by the data source type, and different initial weight coefficients are assigned to behavioral information and environmental information to adapt to the regional differences of the work scenario.

[0032] In another alternative implementation, the fusion features can also use continuous flow measurement data from mobile sensing terminals as the main input, and time alignment can use process events triggered by continuous action recognition as anchor points. For misaligned segments, the optimal time offset is determined by cross-source correlation within a sliding window for correction. The backfilling method is mainly based on streaming generation, with neighborhood interpolation used for short time gaps. In this method, the confidence weight mainly depends on the freshness score of the data to reflect the timeliness changes of mobile terminals in different inspection spans.

[0033] In the embodiments of this application, the present invention eliminates the temporal and spatial offsets of different data sources by synchronizing, aligning, correcting, inspecting and replenishing multi-source data, identifies and replaces abnormal segments, and calculates the reliability, freshness and consistency of the source to form the segment credibility weight. The fusion feature is composed of weighted segments.

[0034] Furthermore, in step S1, multi-source data including spatiotemporally aligned sensing data, operational data, and environmental data are collected, quality verification and missing data repair are performed, and fusion features with confidence weights are generated, including the following steps A1-A6: A1. Time synchronization of multi-source data based on a unified operation time benchmark.

[0035] A2. By identifying cross-modal key event anchor points, asynchronous data is aligned at the event level to form spatiotemporally consistent data fragments.

[0036] A3. For data segments where anchor points could not be identified, corrections were made based on the time offset from the aligned segments.

[0037] A4. Perform a consistency check on the aligned data segments to identify any abnormal segments.

[0038] A5. For abnormal segments, use the spatiotemporal interpolation method of neighborhood data to fill in the gaps. For segments that cannot be interpolated, use the generative model to fill in the gaps.

[0039] A6. Calculate the credibility weight of each data segment based on source reliability, freshness, and consistency, and generate fusion features.

[0040] In this embodiment, the abnormal segments in step A4 can be evaluated for stability through a consistency detection process after the spatiotemporal alignment of multi-source data is completed. Specifically, this includes sequence difference calculation within a sliding window. If the median absolute deviation of the difference sequence exceeds a set threshold, it is considered a mutation and the data segment is marked as an abnormal segment. For abnormal segments, backfilling processing is performed based on the spatial neighborhood and temporally adjacent segments of the node where it is located. If the neighborhood data is insufficient, a lightweight generation model is called to perform interpolation prediction to ensure that the abnormal segments can be used for subsequent fusion feature generation after processing.

[0041] In an optional implementation, abnormal segments can also be identified by a consistency test method based on statistical features after the perceived data is aligned. The basis for anomaly judgment is the angle between data from different sources within the same time window. If the angle is greater than a set threshold, it is marked as a multi-source contradiction and regarded as an abnormal segment. Such abnormal segments do not directly participate in the confidence weight calculation, but are statically replaced by stable data from nearby historical periods. The replaced segments are then labeled with uncertainty.

[0042] In another optional implementation, during the alignment of operational data and environmental data, if a sudden gradient change in the temperature sequence, inconsistency between the operational flow status and sensor signals, or a sharp jump in the position signal is detected, the corresponding segment is marked as an anomalous segment. Such anomalous segments are preferentially corrected by weighted interpolation using reliable segments of spatially adjacent nodes within the graph structure. If the interpolation error exceeds the verification threshold, the backfilling is not performed, and the segment is marked as unusable and excluded from the fusion feature calculation process.

[0043] This invention performs consistency analysis on multi-source data before fusion, identifies and locates abnormal segments formed under temporal changes, cross-source contradictions or physical mutations, and combines interpolation and generation models to perform replenishment or removal of abnormal segments, so that the constituent data of the fused features have a higher data consistency foundation.

[0044] Specifically, a serial process of causal time warp alignment, sensor fingerprint identification, and triple consistency quality control is implemented before the data enters the spatiotemporal graph: First, using the unified time scale of the operation as the root, cross-modal causal time warp is used to align data of different sampling frequencies with event-level anchor points. The anchor points come from the instantaneous moments of electrical interruption, current surge, and action recognition such as voltage detection / grounding / restoration. When a source is missing an event anchor point, the cross-correlation peak shift with the aligned source is used as the correction amount. If the absolute value of the shift exceeds 2 seconds, it is defined as drift and is reversed. Then, a fingerprint (clock jitter, noise floor spectral shape, quantization step, pose inertia features) is extracted for each sensor source and compared with the registered fingerprint. If any two of the three items deviate from the registered confidence boundary, it is marked as suspected counterfeiting or unauthorized access and is downgraded. Then, a triple quality control process is performed: temporal consistency (the median absolute deviation of the sliding window first-order difference exceeding the steady-state threshold is considered a jump), cross-source consistency (the angle between feature vectors from the same location and window is greater than 60° is considered a contradiction), and physical consistency (outside the mechanism domains such as temperature-sag-tension, wind speed-yaw acceleration, etc., are considered suspicious). Suspicious and missing data are first imputed using spatiotemporal KNN; if the neighborhood is insufficient, a lightweight streaming generation model is used for imputation. Unrecoverable long gaps are replaced with the historical nearest neighbor of the same segment and labeled with uncertainty. A credibility weight is calculated for each segment for subsequent fusion; the calculation formula is: in, Indicates the first Segment credibility weight; The source reliability score is based on calibration reports and offline assessment of mismatch rates, and ranges from 0 to 1. This represents freshness score, which decays exponentially over time and ranges from 0 to 1. This represents the combined score for alignment and consistency, ranging from 0 to 1. This represents the three normalization coefficients, whose values ​​are set according to the importance of the three dimensions. .

[0045] In this embodiment, the graph structure in step S2 can be constructed using spatial entities (including towers, cable segments, span nodes, etc.) and participating roles (including operators, monitors, patrol units, etc.) identified in the work scenario as the vertex set of the graph, thus creating a dynamic graph structure. The graph edges include three types: first, structural connection edges, constructed based on the physical connection relationships of electrical equipment; second, spatial proximity edges, established based on the actual geographical distance and layout relationships between equipment; and third, operational flow edges, directed edges connected based on the dependencies in the work procedures, used to represent the sequence of process stages. After setting initial edge weights, the graph edges are periodically updated according to real-time changes in environmental conditions and work behavior, maintaining the dynamism and adaptability of the dynamic graph structure.

[0046] In an optional implementation, the graph structure can also be based on the geographic information system coordinates of the power distribution line repair area, and a set of nodes can be formed by parsing the inspection points and cable segment numbers in the work plan table; the structural connection of the graph edges is automatically generated based on the cable topology map, the edge weights of spatially adjacent edges are adjusted according to the geographical distance between the two nodes and the obstacle factor, and the operation flow edges are determined by the sequential relationship set in the work instruction code; the edge weights are refreshed once every fixed time window based on weather conditions, personnel progress and work behavior logs.

[0047] In another alternative implementation, the graph structure can also use the tower number and personnel positioning tag identified from the UAV inspection image as the graph node input, the structural connection edge is determined by the insulator string connection relationship matched by the image recognition algorithm; the spatial proximity edge is generated according to the Euclidean distance of the visual coordinates, and the operation flow edge is established according to the order of operation commands extracted by the voice recognition system.

[0048] This invention constructs a dynamic graph structure consisting of spatial nodes and participating roles. The graph edges represent physical connections, spatial proximity relationships, and operational process dependencies, respectively, enabling the graph structure to express real operational space and temporal characteristics. The graph edges have real-time update capabilities and can be refreshed synchronously with changes in environmental conditions and operational behaviors.

[0049] Furthermore, in step S2, a dynamic graph structure with spatial nodes and participating roles as vertices is constructed in the work scenario. The graph edges are established based on structural connectivity, spatial proximity, and operational process dependencies, including the following steps B1-B5: B1. Identify spatial nodes and participating role nodes in the work scenario, and construct a vertex set for the dynamic graph.

[0050] B2. Set structural connection edges based on facility topology to represent physical connections.

[0051] B3. Set spatial proximity edges based on geographic distance to indicate location association.

[0052] B4. Set operation flow edges according to the work process sequence to indicate process dependencies.

[0053] B5. Set initial edge weights for the graph edges and update them periodically according to environmental conditions and operational behaviors.

[0054] when (This value can be preset; the number here uses a conventional setting.) When this occurs, the segment is not directly entered into the graph, but is replaced by the neighborhood confidence lower bound statistic. The so-called triggering of order disruption refers to the actual process sequence violating the partial order constraint in the knowledge graph: power off—power test—grounding—marking—operation—review—power restoration. The criterion is the existence of at least one pair of reverse orders or the absence of any necessary process. The reverse order is obtained by comparing the topological sorting of the work log timestamps and video action recognition segments. Once this determination is triggered, the evidence for that time period is marked as unusable and only used for anomaly interpretation, not for risk aggregation. By placing the source authenticity and causal event anchor points before alignment and quality control, systematic deviations caused by traditional one-sided repairs relying solely on timestamps and interpolation are avoided.

[0055] The integrated refinement of dynamic spatiotemporal graph and differentiable rules—physical mask—micro-meteorological pushdown constructs a dynamic graph with towers, spans, work units and personnel as nodes. Edge classes include electrical topology, spatial proximity and process dependence; edge weights are updated in real time with micro-meteorology, load and human factors.

[0056] To address the shortcomings of existing graph structures lacking physical awareness and procedural consciousness, this invention integrates micro-meteorological pushdown, physical safety channels, and differentiable rule layers into the graph ontology: First, station-level meteorology and DEM topography are pushed down to the span scale via a graph neural network downsampling model, generating fine-grained fields of local wind speed, wind direction, and icing thickness; then, safety distance, sag limit, and icing load threshold are mapped as attention-gated safety channels, with those not meeting the requirements being shielded from propagation by binary gating me=0; the threshold is derived from enterprise safety regulations and design manuals, and adaptively expanded according to local estimation variance to control false positives; simultaneously, a differentiable rule layer is formed from a knowledge graph composed of work tickets, safety regulations, tool status, and qualification relationships, with strong conflicts (such as re-energizing without grounding) mapped to negative infinite penalty terms for strong blocking, and weak conflicts (such as slightly lower qualifications but with a tutor accompanying the student) mapped to a propagation weight reduction of 0.5, and rule hits are pre-checked with SAT / SMT to eliminate ambiguity before being soft-injected.

[0057] The edge weight update cycle is 30 seconds, which is shortened to 5 seconds when order disruption or distribution drift is detected to quickly isolate the abnormal influence domain; the spatial proximity edge weight is dynamically adjusted according to the wind corridor direction, and the proximity weight is increased when the angle between the wind direction and the span azimuth is less than 30° to reflect wind-induced coupling.

[0058] Among these, micrometeorological downscaling refers to mapping station-level observations to the tower-span scale through terrain and historical distribution learning; safety pipeline refers to transforming physical safety constraints into structured gating of attention and edge propagation; and differentiable rule layer refers to embedding clause constraints into the model's structured layer using differentiable penalties or masking. The advantage of this integrated design lies in simultaneously constraining information propagation from three dimensions: structure, physics, and procedures, reducing the illusion of meaningless cross-edge attention and procedural compliance without physical safety.

[0059] Furthermore, in step S3, a physical consistency index constraint reflecting the operational safety boundary is introduced into the graph structure, and a causal propagation path is constructed to distinguish between external disturbance paths and internal state paths, forming an interpretable risk characterization mechanism, including the following steps C1-C6: C1. Set physical consistency indicators in the graph structure to constrain propagation behavior and reflect the operational safety boundary.

[0060] C2. Perform propagation suppression processing on graph edges that do not meet the physical consistency index.

[0061] C3. Identify external disturbance paths and internal state paths according to data source attributes.

[0062] C4. Set the propagation direction of external disturbance paths, so that they only propagate to internal state nodes and risk nodes.

[0063] C5. Set the propagation range of internal status paths to be limited to propagation between device nodes, work unit nodes, and personnel nodes.

[0064] C6. Construct a risk characterization mechanism based on paths and boundaries.

[0065] In this embodiment, the external disturbance path in step C3 can be classified into various types of edges in the graph based on the attribute labels of the data source after the graph structure is constructed. All graph edges generated by environmental data or external sensing devices are uniformly identified as external disturbance paths. This path is used to represent the direction of influence of non-operational internal factors such as environmental changes and third-party interference. In the propagation mechanism, the external disturbance path is set to propagate in one direction, allowing propagation only from environmental sensing nodes to equipment nodes, operation nodes, or personnel nodes. At the same time, in order to prevent the external anomaly from being amplified indefinitely, an edge weight attenuation mechanism is set in the path propagation so that the disturbance impact weakens step by step as the number of path layers increases.

[0066] In one alternative implementation, after establishing the graph structure, the external disturbance path can also be filtered by selecting nodes and graph edges corresponding to all information sources related to the work site, such as weather tags, traffic events, geographical obstacles, and third-party construction. Data sources that are not directly causally related to internal operations are considered as external disturbance sources. Their connecting edges are marked as external disturbance paths. These paths only participate in the background disturbance input of the current graph state and do not trigger the active update of state variables within the graph structure, ensuring that external changes only affect the risk score within a reasonable range.

[0067] In another alternative implementation, external disturbance paths can also be defined by analyzing the co-occurrence relationships of various graph edges in historical samples. Edges that are associated with non-operational nodes (such as weather stations, road sensors, and third-party communication nodes) and do not appear in the internal operation flow are all defined as external disturbance paths. These paths only play a triggering role at the outermost layer of the graph structure during the graph propagation process and do not participate in the causal update of intermediate state variables. In addition, to enhance the risk interpretability of the graph model, a disturbance impact factor is set for the external disturbance paths, which participates in the final risk level mapping process.

[0068] This invention marks data paths originating from non-operational internal systems in a graph structure as external disturbance paths and sets specific propagation directions and edge weight attenuation strategies. This effectively isolates the influence boundary between external environmental fluctuations and internal operational states. At the same time, it enables the risk representation mechanism in the graph structure to have path-level causal differentiation capabilities, thereby improving the interpretability of the final risk prediction results and the accuracy of intervention operations.

[0069] Specifically, to alleviate the shortcomings of traditional black-box models such as attribution misalignment and cross-stage crosstalk, this invention separately encodes external disturbance channels (weather, road conditions, external alarms) and internal state channels (electrical parameters, equipment status, work behavior, human factors), and applies targeted gating. External channels only point to internal and risk nodes, while internal channels are limited to loops within the equipment-work-person subgraph. All actions triggered by system policies are marked with intervention tags to prevent confusion with natural observations. Stage locks use the triple consistency of work order progress, ticket status, and video action recognition as boundaries; any causal propagation must not cross these boundaries, and crossing them is weakened to zero by the rule layer. The dominant factor is scored jointly by attention score and offline perturbation response. The offline perturbation response is obtained by applying small increments to historical stable windows using a shadow model, without on-site trial and error. If the combined contribution of three consecutive windows exceeds a threshold, it is marked as a dominant factor. The threshold is derived from the Youden points of the historical truth replay set and is calibrated quarterly. When strong exogenous perturbations such as typhoon warning level upgrades occur, the threshold is lowered by 10% to improve sensitivity, while the uncertainty interval is expanded in the next step to suppress false alarms. Definitions: Causal path refers to a directional feature propagation path; stage lock refers to a mechanism that constrains propagation by stage boundaries; dominant factor refers to an external or internal factor that contributes stably and significantly to the increase in risk. The advantage of this design is that it unifies causal direction, stage boundaries, and evidence strength into a computable paradigm, preventing cross-stage misattribution.

[0070] Furthermore, in step S4, the risk characterization results are subjected to split bar and interval calibration to obtain the uncertainty interval that meets the coverage target, and the early warning threshold is adaptively adjusted according to the state feedback, including the following steps D1-D7: D1. After generating the fusion features and completing the graph structure construction, the output results of the risk nodes in the graph structure risk representation are grouped according to the category of the source data.

[0071] D2. Apply a robust modeling method based on minimum performance guarantee to each group of risk nodes so that the most extreme node in each group can still be identified.

[0072] D3. Calculate the structural index and distance index based on the temporal stability and distribution deviation of each fused feature segment.

[0073] D4. When any indicator exceeds the preset judgment range, the corresponding fused feature fragment is marked as data outside the training distribution.

[0074] D5. For the labeled fusion feature fragments, expand their interval prediction range for risk nodes in the graph structure and reset the upper and lower limit positions.

[0075] D6. The upper and lower limits are determined by the pre-set coverage targets, which are adjusted according to the risk level.

[0076] D7. After the uncertainty interval expands, the alarm level of the corresponding risk node in the graph structure is raised by one level according to the preset level mapping relationship.

[0077] Specifically, to address the pain points of existing systems such as distortion in cross-domain instability intervals and arbitrary thresholds, this invention employs robust training to stabilize the performance of the worst-case group. During operation, out-of-sample detection is performed using both population stability index and energy distance as criteria. Any instance exceeding the threshold is considered an out-of-sample situation. The threshold is derived from the ROC equilibrium point of the cross-regional migration experiment and is updated semi-annually.

[0078] Interval calibration uses stratified leave sets to perform quantile calibration, outputting risk and TTE intervals in the form of... in, Indicates the lower bound of the interval; Indicates the upper bound of the interval; This represents the lower quantile estimated on the holdout set; This indicates that the target coverage is half and is coupled with the alarm strategy, with a value ranging from 0 to 0.5; Indicates the upper quantile position; when triggered out of sample, Automatically increase by 0.05 to widen the range and simultaneously raise the alarm level by one level. This can be adjusted based on preset values ​​and step sizes from historical data.

[0079] To transform uncertainty into actionable steps, a risk budget mapping is introduced: the interval width, out-of-sample indicators, and consistency with the dominant factor are mapped to the search radius of the contingency plan library, the delay of automatic power-on, and the intensity of secondary confirmation. When the interval width exceeds the historical P90 or the dominant factor conflicts with the procedure, unsupervised power-on is automatically prohibited, and the process is forced to proceed to the next counterfactual simulation. P90 is the 90th percentile, indicating that approximately 90% of the data points are below this value.

[0080] Out-of-sample refers to situations where the operating distribution deviates significantly from the training distribution; risk budgeting refers to converting the magnitude of uncertainty into resource thresholds and policy quotas. The advantage of this design lies in using an interval control strategy to avoid misleading decisions based on single-point estimations.

[0081] Furthermore, in step S5, counterfactual simulations are performed on multiple candidate response schemes based on digital twin modeling, and the sensing device is triggered to perform active supplementary data collection based on the information value maximization strategy, including the following steps E1-E7: E1. After the graph structure is constructed and the uncertainty range is determined, a digital twin model is constructed based on the current graph node status and fusion characteristics. The digital twin model covers the work path, time period, personnel and equipment information in the graph structure.

[0082] E2. In the digital twin model, intervention operations are applied sequentially, including adjusting the work path, changing the work time, replacing personnel, setting the work speed, and interrupting the work.

[0083] E3. For each intervention operation, calculate the resulting changes in the state of risk nodes in the graph structure, the duration of cross-node operations, and the required amount of equipment and personnel resources.

[0084] E4. Based on the set weight parameters, perform multi-index weighted scoring on the intervention operation and eliminate operation paths that conflict with the constraint edges in the graph structure.

[0085] E5. When there are multiple interventions with the same score, select the intervention that has the highest correlation with the dominant factor corresponding to the current risk node and involves the fewest roles.

[0086] E6. When the confidence level of a data segment in the fused features is lower than the threshold and the segment is located in a high-risk area of ​​the graph structure, the active supplementary sampling operation of the corresponding location of the sensing device is triggered.

[0087] E7. Whether the sensing device performs supplementary data collection is determined by the alarm level, the remaining available sensing resources, and the weather label of the area where the map node is located.

[0088] To avoid on-site trial and error and reliance on experience-based decisions, this invention applies intervention simulations to candidate contingency plans such as route modification, time period modification, shift replacement, speed limit, and temporary parking within a digital twin. It calculates a three-dimensional index of risk reduction, recovery time, and resource cost, and ranks them using a comprehensive utility function. in, Indicates the overall effectiveness of the contingency plan; The three weights are derived from the Delphi method score of the company's annual safety and power supply KPIs and normalized. This indicates the risk reduction relative to the current situation, expressed in risk points. Indicates the estimated recovery time, in minutes; This represents the cost of resources, which is converted into a unified cost component including man-hours, vehicles, tools, and opportunity costs.

[0089] To improve feasibility and interpretability, a shadow cost surface is introduced: an approximately differentiable cost-benefit surface is established for each type of intervention, derived from historical mission replays and small-scale controlled experiments, and used as a regularizer in simulations to avoid selecting theoretically sound but infeasible contingency plans in practice. The selection criteria are as follows: And it satisfies the strong constraints of the rule layer. Use the lower quartile of history to ensure conservatism; if multiple factors are tied, choose the contingency plan that most directly impacts the dominant factor and has the least burden on human factors.

[0090] The simulation also generates failure modes (violation of physics, disruption of order, and exceeding of human factors limits), and writes back the rule layer to form negative samples to patch blind spots. Definitions: Digital twin refers to a jointly computable model of equipment, terrain, microclimate, process, and human factors; shadow cost surface refers to a differentiable cost-benefit approximation derived from historical data; the advantage of this design lies in replacing on-site trial and error with virtual trial and error, and achieving usability and usability through cost surface constraints.

[0091] To address the problems of blind and inefficient traditional re-mining operations, this invention uses the reduction in uncertainty per unit cost and the gain in risk coverage as information value, and considers team safety and arbitrary time-limited planning that can be interrupted at any time: the value of a single re-mining operation is defined as: in, This indicates the value of the proactive perception. This indicates the expected risk prior to supplementary extraction; This represents the expected conditional risk after performing this perception. This represents the cost of re-mining, converted into a uniform cost unit.

[0092] when Time-triggered supplementary sampling; The alarm level is determined by the available resources and weather window, and automatically decreases by 20% at yellow and above. Path planning considers drone routes, vehicle inspections, and secondary personnel checks, and is constrained by no-fly zones, road accessibility, and personnel safety buffer radius. The algorithm uses a sub-modal greedy heuristic to obtain a 1-1 / e approximation, and uses integer linear programming for hot start to ensure that it can be interrupted and resumed at any time under any time limit. The planning incorporates team safety constraints: those with high fatigue cannot be assigned to high-risk nodes, and those with insufficient qualifications can only enter low-risk areas with their mentors.

[0093] The most efficient approach for disconnected fields is edge-based degradation inference. The degradation model consists of regional nearest neighbor statistics and strong constraints at the rule layer, outputting conservative conclusions. The process of incorporating new observations from re-sampled data into the initial step achieves alignment quality control, and the calculation... Then, a dynamic graph is injected to narrow down the interval and may lower the alarm level or change the sorting.

[0094] Among these, information value refers to the reduction in uncertainty and the gain in risk coverage per unit cost; arbitrary time-limited planning refers to a planning method that can produce the current optimal feasible solution at any point in time; team safety refers to hard constraints on personnel allocation such as fatigue, qualifications, and consecutive shift length. The advantage of this design is that it makes it clear where, when, and who collects evidence, and it can be interrupted and resumed at any time.

[0095] Furthermore, in step S6, the risk level, dominant factor, and expected trigger time are output, and the perception and simulation results are fed back to the model for updating and iteration, including the following steps F1-F8: F1. Based on the marked risk node status in the graph structure, output the risk level for the current time period and record the corresponding fused feature location.

[0096] F2. In the graph structure, calculate the attention allocation score and the change in risk value after perturbation input for all node feature paths.

[0097] F3. Disturbance processing applies a small increment to each input in the fusion feature based on stable data within a historical time window to calculate the change in risk level.

[0098] F4. When a feature shows a significant impact on the risk level in three consecutive time windows, the feature is marked as the dominant factor of the corresponding node in the current graph structure.

[0099] F5. The criteria for determining the dominant factor are derived from historical graph node replay data and are updated according to a preset cycle.

[0100] F6. When the weather label in the fused features indicates a strong external disturbance, lower the threshold for determining the dominant factor and expand the interval prediction range of the corresponding node.

[0101] F7. Write the risk level, dominant factor label, expected trigger time, and intervention results of each node in the current graph structure into the graph structure's status record table.

[0102] F8. All intervention schemes identified as failed in the digital twin model, including paths that violate the physical boundaries of the operation, combinations of operation nodes with incorrect order, and execution paths with mismatched roles, are recorded in the exclusion subgraph of the graph structure.

[0103] Example 3 is an embodiment of the present invention. This embodiment provides a risk assessment system for emergency repair operations of power distribution lines based on multi-source sensing data, including a data fusion module, a graph structure construction module, a risk modeling module, an uncertainty control module, a supplementary data acquisition module, and an output module.

[0104] The data fusion module is used to collect and align multi-source data, including spatiotemporal sensing data, operational data, and environmental data, to perform quality verification and missing data repair, and generate fusion features with credibility weights.

[0105] The graph structure construction module is used to build dynamic graph structures with spatial nodes and participating roles as vertices in the operation scenario. The graph edges are established based on structural connections, spatial proximity, and operational process dependencies.

[0106] The risk modeling module is used to introduce physical consistency index constraints that reflect the safety boundary of operations into the graph structure, construct causal propagation channels, distinguish between external disturbance paths and internal state paths, and form an interpretable risk representation mechanism.

[0107] The uncertainty control module is used to perform sub-bar and interval calibration on the risk characterization results to obtain the uncertainty interval that meets the coverage target, and adaptively adjust the warning threshold according to the status feedback.

[0108] The supplementary data acquisition module is used to perform counterfactual simulations of multiple candidate response schemes based on digital twin modeling, and to trigger the sensing devices to perform active supplementary data acquisition operations based on the information value maximization strategy.

[0109] The output module is used to output the risk level, dominant factors, and expected trigger time, and to feed the perception and simulation results back to the model for updates and iterations.

[0110] This embodiment also provides an electronic device applicable to the risk assessment method for emergency repair operations of power distribution lines based on multi-source sensing data, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the risk assessment method for emergency repair operations of power distribution lines based on multi-source sensing data proposed in the above embodiment.

[0111] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the risk assessment method for emergency repair operations of power distribution lines based on multi-source sensing data as proposed in the above embodiments.

[0112] The storage medium proposed in this embodiment and the method for risk assessment of power line emergency repair operations based on multi-source sensing data proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0113] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0114] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A risk assessment method for emergency repair operations of power distribution lines based on multi-source sensing data, characterized in that: include, Collect and align multi-source data, including spatiotemporal sensing data, operational data, and environmental data, perform quality verification and missing data repair, and generate fusion features with credibility weights. In the work scenario, a dynamic graph structure is constructed with spatial nodes and participating roles as vertices. The graph edges are established based on structural connections, spatial proximity, and operational process dependencies. Introduce physical consistency index constraints to reflect the safety boundary of operations into the graph structure, construct causal propagation channels, distinguish between external disturbance paths and internal state paths, and form an interpretable risk characterization mechanism. The risk characterization results are subjected to split bar and interval calibration to obtain the uncertainty interval that meets the coverage target, and the early warning threshold is adaptively adjusted according to the state feedback. Counterfactual simulations of multiple candidate response schemes were performed based on digital twin modeling, and the sensing devices were triggered to perform active supplementary data collection based on the information value maximization strategy. Output the risk level, dominant factors, and expected trigger time, and feed the perception and simulation results back to the model for updates and iterations.

2. The method for risk assessment of power line emergency repair operations based on multi-source sensing data as described in claim 1, characterized in that: The multi-source data, including the collected and spatiotemporally aligned sensing data, operational data, and environmental data, undergoes quality verification and missing data repair to generate fusion features with confidence weights, including... Time synchronization of multi-source data is performed based on a unified operation time benchmark; By identifying cross-modal key event anchor points, asynchronous data is aligned at the event level to form spatiotemporally consistent data fragments; For data segments where anchor points could not be identified, corrections were made based on the time offset from the aligned segments. Perform a consistency check on the aligned data segments to identify any abnormal segments; For abnormal segments, spatiotemporal interpolation of data within the neighborhood is used to fill in the missing segments; for segments that cannot be interpolated, a generative model is used to fill in the missing segments. The credibility weight of each data segment is calculated based on source reliability, freshness, and consistency to generate fusion features.

3. The method for risk assessment of power line emergency repair operations based on multi-source sensing data as described in claim 2, characterized in that: The process of constructing a dynamic graph structure in the work scenario, with spatial nodes and participating roles as vertices, and establishing graph edges based on structural connectivity, spatial proximity, and operational process dependencies, includes: Identify spatial nodes and participating role nodes in the task scenario, and construct a vertex set for a dynamic graph; Structural connection edges are set according to the facility topology to represent physical connections; Spatial proximity edges are set based on geographic distance to indicate location association; Set operation flow edges according to the work process sequence to indicate process dependencies; Set initial edge weights for the graph edges and update them periodically according to environmental conditions and operational behaviors.

4. The method for risk assessment of power line emergency repair operations based on multi-source sensing data as described in claim 3, characterized in that: The process involves introducing physical consistency index constraints to reflect operational safety boundaries into the graph structure, constructing causal propagation channels, distinguishing between external disturbance paths and internal state paths, and forming an interpretable risk characterization mechanism. In the graph structure, physical consistency indicators are set to constrain propagation behavior and reflect the operational safety boundary; For graph edges that do not meet the physical consistency index, propagation suppression processing is performed; External disturbance paths and internal state paths are identified according to data source attributes; Set the propagation direction of external disturbance paths to propagate only to internal state nodes and risk nodes; Set the propagation range of the internal state path to be limited to propagation between device nodes, work unit nodes, and personnel nodes; The risk characterization mechanism is constructed based on the path and boundary.

5. The method for risk assessment of power line emergency repair operations based on multi-source sensing data as described in claim 4, characterized in that: The risk characterization results are subjected to split-bar and interval calibration to obtain an uncertainty interval that meets the coverage target, and the early warning threshold is adaptively adjusted based on state feedback. include, After the fusion features are generated and the graph structure is constructed, the output results of the risk nodes in the graph structure risk representation are grouped according to the category of the source data; A robust modeling method based on minimum performance guarantees is applied to each group of risk nodes, so that the most extreme node in each group can still be identified; Based on the temporal stability and distribution deviation of each fused feature fragment, structural and distance indices are calculated respectively. When any indicator exceeds the preset judgment range, the corresponding fused feature fragment is marked as data outside the training distribution; For the labeled fusion feature fragments, expand their interval prediction range for risk nodes in the graph structure and reset the upper and lower limit positions; The upper and lower limits are determined by pre-set coverage targets, which are adjusted according to the risk level. After the uncertainty range expands, the alarm level of the corresponding risk node in the graph structure is raised by one level according to the preset level mapping relationship.

6. The method for risk assessment of power line emergency repair operations based on multi-source sensing data as described in claim 5, characterized in that: The process of performing counterfactual simulations of multiple candidate responses based on digital twin modeling, and triggering the sensing device to perform proactive data acquisition based on an information value maximization strategy, includes: After the graph structure is constructed and the uncertainty interval is determined, a digital twin model is constructed based on the current graph node status and fusion characteristics. The digital twin model covers the work path, time period, personnel and equipment information in the graph structure. In the digital twin model, intervention operations are applied sequentially, including adjusting the work path, changing the work time, replacing personnel, setting the work speed, and interrupting the work. For each intervention operation, calculate the resulting change in the state of the risk nodes in the graph structure, the duration of cross-node operations, and the required amount of equipment and personnel resources; Based on the set weight parameters, the intervention operation is scored using multiple weighted indicators, and operation paths that conflict with the constraint edges in the graph structure are eliminated. When there are multiple interventions with the same score, select the intervention that has the highest correlation with the dominant factor corresponding to the current risk node and involves the fewest roles. When the confidence level of a data segment in the fused features is lower than the threshold and the segment is located in a high-risk area of ​​the graph structure, the active supplementary sampling operation of the corresponding location of the sensing device is triggered. Whether a sensing device performs supplementary data collection is determined by the alarm level, the remaining available sensing resources, and the weather tags of the area where the map node is located.

7. The method for risk assessment of power line emergency repair operations based on multi-source sensing data as described in claim 6, characterized in that: The output risk level, dominant factor, and expected trigger time will feed the perception and simulation results back to the model for updates and iterations, including... Based on the marked risk node status in the graph structure, output the risk level for the current time period and record the corresponding fused feature location; In the graph structure, the attention allocation score and the magnitude of risk value change after perturbation input are calculated for each node feature path; The perturbation process applies a small increment to each input in the fused feature based on stable data within a historical time window to calculate the risk level change. When a feature shows a significant impact on the risk level in three consecutive time windows, the feature is marked as the dominant factor of the corresponding node in the current graph structure. The criteria for determining the dominant factor are derived from historical graph node replay data and are updated according to a preset cycle; When the weather label in the fused features indicates a strong external disturbance, the threshold for determining the dominant factor is lowered, and the interval prediction range of the corresponding node is expanded. The risk level, dominant factor label, expected trigger time, and intervention results of each node in the current graph structure are uniformly written into the status record table of the graph structure. For intervention schemes identified as failed in the digital twin model, including paths that violate the physical boundaries of the operation, combinations of operation nodes with incorrect order, and execution paths with mismatched roles, all are recorded in the exclusion subgraph of the graph structure.

8. A risk assessment system for emergency repair operations of distribution lines based on multi-source sensing data, employing the risk assessment method for emergency repair operations of distribution lines based on multi-source sensing data as described in any one of claims 1 to 7, characterized in that, include: The system includes a data fusion module, a graph structure construction module, a risk modeling module, an uncertainty control module, a supplementary data collection module, and an output module. The data fusion module is used to collect and align multi-source data, including spatiotemporal sensing data, operational data, and environmental data, to perform quality verification and missing data repair, and to generate fusion features with credibility weights. The graph structure construction module is used to construct a dynamic graph structure with spatial nodes and participating roles as vertices in the operation scenario. The graph edges are established based on structural connections, spatial proximity, and operational process dependencies. The risk modeling module is used to introduce physical consistency index constraints that reflect the safety boundary of the operation into the graph structure, construct causal propagation channels, distinguish between external disturbance paths and internal state paths, and form an interpretable risk representation mechanism. The uncertainty control module is used to perform sub-bar and interval calibration on the risk characterization results to obtain the uncertainty interval that meets the coverage target, and adaptively adjust the warning threshold according to the state feedback. The supplementary data acquisition module is used to perform counterfactual simulations of multiple candidate response schemes based on digital twin modeling, and to trigger the sensing device to perform active supplementary data acquisition operations according to the information value maximization strategy. The output module is used to output the risk level, dominant factor and expected trigger time, and to feed the perception and simulation results back to the model for updating and iteration.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for risk assessment of power distribution line emergency repair operations based on multi-source sensing data as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for risk assessment of power distribution line emergency repair operations based on multi-source sensing data as described in any one of claims 1 to 7.

Images