Power transmission line inspection area dynamic risk grading evaluation method and system based on Conformer model

Abstract

The invention relates to the technical field of power transmission line inspection, and discloses a power transmission line inspection area dynamic risk grading evaluation method and system based on a Conformer model, and the method comprises the steps: building a multi-scale space-time grid based on a line topology and risk propagation model, and achieving the mechanism-guided data fusion; secondly, dynamically adjusting attention distribution of the model by utilizing a collaborative optimization controller, and outputting a risk quantized value and uncertainty; then, performing quality inspection and traceability on the result through a comprehensive credibility calculation formula; then, constructing a multi-objective optimization function to solve an optimal decision under resource constraints; and finally, realizing closed-loop evolution by adopting a three-layer optimization framework, driving system parameters to be adaptively updated, and forming a perception-evaluation-decision-optimization complete link. According to the method, the accuracy, credibility and response efficiency of risk assessment are remarkably improved, dynamic optimal configuration of operation and maintenance resources is achieved, meanwhile, the continuous self-evolution capacity is achieved, and a full-process solution is provided for intelligent operation and maintenance of the power transmission line.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power transmission line inspection, in particular to a power transmission line inspection area dynamic risk grading evaluation method and system based on a Conformer model. BACKGROUND

[0002] With the development of smart grids and digital twin technology, the field of power transmission line inspection and risk assessment has evolved from traditional manual mode to automation and intelligence. In the perception layer, existing technologies generally use unmanned aerial vehicles, satellite remote sensing, online monitoring devices, etc. to obtain visible light, infrared, laser point cloud and working condition parameters and other multi-source data, and perform image defect identification or time series anomaly detection through deep learning models (such as CNN, Transformer). In the evaluation layer, deep learning-based risk prediction models are widely used to fuse multi-source features and output risk probabilities or levels. In the decision support layer, some systems begin to try to combine rule engines or simple optimization models to map risk results to preliminary inspection recommendations. These technologies form the basis of current line intelligent operation and maintenance;

[0003] However, the fusion of multi-source data in existing technical systems mostly stays in simple spatiotemporal superposition or feature splicing, lacking deep spatiotemporal correlation modeling based on line physical topology and risk propagation mechanism, resulting in a disconnection between perception results and actual evolution of risks. Secondly, risk assessment models are mostly static "black boxes" with fixed attention mechanisms and decision logic, which cannot be dynamically adjusted and explained according to real-time changes in operation and maintenance goals (such as power preservation mode and disaster prevention mode). Thirdly, the chain from risk assessment to operation execution is broken, and a global optimization decision model that integrates safety, timeliness, cost and other multi-objective constraints has not been established, and resource allocation relies on manual experience, which is inefficient;

[0004] To solve the above problems, the present application proposes a power transmission line inspection area dynamic risk grading evaluation method and system based on a Conformer model. SUMMARY

[0005] The present application aims to provide a power transmission line inspection area dynamic risk grading evaluation method and system based on a Conformer model, mainly to solve the problem that the fusion of multi-source data in existing technical systems mostly stays in simple spatiotemporal superposition or feature splicing, lacking deep spatiotemporal correlation modeling based on line physical topology and risk propagation mechanism, resulting in a disconnection between perception results and actual evolution of risks, and a global optimization decision model that integrates safety, timeliness, cost and other multi-objective constraints has not been established, and resource allocation relies on manual experience, which is inefficient.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] The method and system for dynamic risk grading assessment of transmission line inspection area based on Conformer model comprises the following steps:

[0008] S1: spatio-temporal perception fusion, obtaining multi-source inspection data of the target transmission line corridor, constructing a multi-scale spatio-temporal grid based on the line topology and risk propagation model, and performing heterogeneous graph alignment and adaptive weighted fusion in the grid to generate a unified feature representation;

[0009] S2: collaborative optimization assessment, inputting the unified feature representation into the dynamic risk analysis model, and adjusting the model attention distribution in real time according to the real-time evaluation target through the internal collaborative optimization controller, and outputting the comprehensive risk quantitative value and uncertainty estimation of each inspection unit;

[0010] S3: quality evaluation and credible traceability, based on the uncertainty estimation, data source quality index and historical evaluation accuracy, calculating the comprehensive credibility score of the risk assessment result of each inspection unit; for the high-risk assessment result with credibility lower than the threshold, automatically tracing and visualizing the dominant risk factor and key evidence data;

[0011] S4: resource-constrained decision-making, constructing a multi-objective optimization function integrating evaluation accuracy, response delay and resource consumption, solving the function based on the comprehensive risk quantitative value, comprehensive credibility score and real-time resource state to obtain the dynamic risk level and corresponding differentiated operation and maintenance strategy;

[0012] S5: closed-loop evolution, executing the operation and maintenance strategy, collecting feedback data, and jointly optimizing the dynamic risk analysis model parameters, the fusion weights of the multi-scale spatio-temporal grid and the weights of the multi-objective optimization function based on the strategy effectiveness and prediction deviation.

[0013] Preferably, in the quality evaluation and credible traceability step, the "comprehensive credibility score" is calculated by the following formula: C = a ⋅ (1−Unorm) + β ⋅ Qdata + γ ⋅ Ahistory, wherein C is the comprehensive credibility score, Unorm is the normalized model uncertainty estimation value, Qdata is the comprehensive quality score of the input data, Ahistory is the accuracy rate of the historical assessment result of the unit, a, β, γ are adjustable weight coefficients, and a + β + γ = 1.

[0014] Preferably, in the spatio-temporal perception fusion step, the "multi-scale spatio-temporal grid based on line topology and risk propagation model" specifically comprises:

[0015] S11: basic grid construction, constructing a basic topological grid reflecting electrical connection and physical space risk conduction path according to tower coordinates, conductor sag and real-time meteorological field data;

[0016] S12: Risk model calling, for at least one risk factor to be evaluated in the mountain fire, icing, and external force damage, call its preset risk propagation model, which defines the diffusion speed, attenuation characteristics and spatial influence function of the factor;

[0017] S13: Multi-scale window fusion, on the basis of the topological grid, according to the dynamic influence range output by the called risk propagation model, a multi-scale spatio-temporal analysis window corresponding to each risk factor is generated in parallel, and these windows are superimposed and fused in the time and space dimensions, and finally a unified multi-scale spatio-temporal grid is formed.

[0018] Preferably, in the collaborative optimization evaluation step, the working mechanism of the "collaborative optimization controller" is: receiving an external input optimization target vector, generating a set of spatial attention modulation coefficients through a lightweight strategy network, and applying the coefficients to the key-value pairs of the global attention path in the dynamic risk analysis model to realize dynamic reweighting of the risk attention degree of different geographical areas.

[0019] Preferably, in the closed-loop evolution step, the "joint optimization" is realized by using a three-layer optimization framework:

[0020] Inner optimization: fixing the fusion weight and function weight, updating the model parameters by using the feedback data;

[0021] Middle optimization: based on the updated model, optimizing the dynamic weight configuration of the multi-objective function;

[0022] Outer optimization: evaluate the long-term system performance under different fusion weights, search and update the optimal fusion weight.

[0023] The power line inspection area dynamic risk grading evaluation system based on the Conformer model comprises:

[0024] A multi-scale spatio-temporal perception fusion module is used to perform the spatio-temporal perception fusion step, realize grid construction and multi-source heterogeneous data fusion based on risk propagation mechanism;

[0025] A collaborative optimization intelligent evaluation module is used to perform the collaborative optimization evaluation step, integrate a controllable deep learning model and an optimization controller, and output risk quantization values and uncertainties;

[0026] A quality evaluation and credible source module is used to perform the quality evaluation and credible source step, calculate the credibility of the evaluation results and perform attribution analysis on high-risk conclusions;

[0027] A multi-objective resource constraint decision module is used to perform the resource constraint decision step, and the optimal risk level and operation and maintenance strategy are solved under multiple constraints;

[0028] A perception-decision closed loop evolution module is configured to perform the closed loop evolution step, and automatically iterates and optimizes key parameters by system feedback.

[0029] Preferably, the multi-scale spatio-temporal perception fusion module comprises:

[0030] A topology and risk model library unit stores line electrical topology data and standardized propagation models of various risk factors.

[0031] A multi-scale grid dynamic construction unit is configured to call the model library and dynamically generate corresponding multi-scale spatio-temporal grids according to real-time risk factor types.

[0032] A graph fusion calculation unit is configured to represent multi-source data as node and edge features on a graph, and perform graph neural network fusion calculation on the dynamically constructed grid.

[0033] Preferably, the collaborative optimization intelligent evaluation module comprises:

[0034] A gated spatio-temporal feature extraction network unit is composed of parallel local convolution branches and global attention branches, and fuses them through a gating mechanism.

[0035] An embedded collaborative optimization controller unit is integrated in the feature extraction network, and generates an attention modulation coefficient in real time according to input.

[0036] An uncertainty quantification output unit is configured to synchronously output a comprehensive risk quantification value and its prediction uncertainty at the end of the model.

[0037] Preferably, the perception-decision closed loop evolution module comprises:

[0038] A feedback effectiveness collection and analysis unit is configured to collect strategy execution results and calculate actual disposal effectiveness indicators.

[0039] A three-layer collaborative optimization engine unit is configured to execute the three-layer optimization framework as claimed in claim 5, and coordinate the update of model parameters, function weights, and fusion weights.

[0040] A strategy knowledge graph self-evolution unit is configured to automatically convert verified efficient decision cases into new knowledge and store them in the strategy knowledge graph.

[0041] Preferably, the system further comprises a panoramic visualization and interactive cockpit module configured to:

[0042] In a digital twin scenario, live risk distribution, resource scheduling, credibility labeling, and traceability evidence chain are synchronously displayed.

[0043] An interactive interface is provided for operation and maintenance personnel to adjust and optimize targets, review system decisions, and inject domain knowledge.

[0044] Advantages:

[0045] (1) Perception fusion and target-adjustable collaborative evaluation based on risk propagation mechanism enhance the relevance of features and the pertinence of evaluation from the source, making high-risk identification earlier and more accurate. Quantitative credibility scoring and automatic traceability mechanism provide a credible basis and transparent explanation for decision-making, greatly improving the acceptability of the results;

[0046] (2) By constructing and real-time solving a multi-objective function that integrates safety, timeliness and cost, the system can automatically generate a comprehensive optimal differentiated strategy based on real-time risk value, credibility and resource state, realizing the leap from static risk alert to dynamic resource optimal scheduling, and significantly improving resource utilization efficiency;

[0047] (3) The closed-loop mechanism of joint iterative optimization of model parameters, fusion weights and decision preferences based on real feedback enables the system to adapt to environmental changes and data distribution drift, overcoming the performance degradation problem of traditional static models, ensuring long-term applicability and reducing long-term operation and maintenance costs;

[0048] (4) Each link is deeply coupled and mutually enhanced, mechanism perception improves evaluation quality, credible evaluation optimizes decision input, decision feedback drives system evolution, forming a self-reinforcing positive cycle. Ultimately, the core indicators of evaluation accuracy, decision intelligence, resource economy and system adaptability are simultaneously and significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0049] Fig. 1 The power transmission line inspection risk assessment method flowchart of the present application;

[0050] Fig. 2 The spatio-temporal perception fusion step flowchart of the present application;

[0051] Fig. 3 The power transmission line inspection risk assessment system flowchart of the present application;

[0052] Fig. 4 The multi-scale spatio-temporal perception fusion module flowchart of the present application;

[0053] Fig. 5 The collaborative optimization intelligent evaluation module flowchart of the present application;

[0054] Fig. 6 The perception-decision closed-loop evolution module flowchart of the present application. DETAILED DESCRIPTION

[0055] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application.

[0056] Please refer to Figs. 1-2 The power line inspection area dynamic risk grading evaluation method based on the Conformer model includes the following steps:

[0057] S1: spatiotemporal perception fusion, acquiring multi-source inspection data of the target power line corridor, constructing a multi-scale spatiotemporal grid based on the line topology and risk propagation model, and performing heterogeneous graph alignment and adaptive weighted fusion in the grid to generate a unified feature representation;

[0058] S2: collaborative optimization evaluation, inputting the unified feature representation into a dynamic risk analysis model, and adjusting the attention distribution of the model according to real-time evaluation of the target through the internal collaborative optimization controller, and outputting the comprehensive risk quantitative value and uncertainty estimation of each inspection unit;

[0059] S3: quality evaluation and credible traceability, based on the uncertainty estimation, data source quality index and historical evaluation accuracy, calculating the comprehensive credibility score of the risk evaluation result of each inspection unit; for the high-risk evaluation result with credibility lower than the threshold, automatically tracing and visualizing the dominant risk factor and key evidence data;

[0060] S4: resource-constrained decision-making, constructing a multi-objective optimization function that integrates evaluation accuracy, response delay and resource consumption, solving the function based on the comprehensive risk quantitative value, comprehensive credibility score and real-time resource state to obtain the dynamic risk level and the corresponding differentiated operation and maintenance strategy;

[0061] S5: closed-loop evolution, executing the operation and maintenance strategy, collecting feedback data, and based on the strategy effectiveness and prediction deviation, jointly optimizing the dynamic risk analysis model parameters, the fusion weights of the multi-scale spatiotemporal grid and the weights of the multi-objective optimization function.

[0062] An intelligent closed loop of "perception-evaluation-quality inspection-decision-evolution" is constructed. S1 dynamically constructs a perception grid according to a line physical topology and a risk propagation mechanism, and realizes mechanism and data fusion. S2 receives external target instructions through an internal controller, dynamically adjusts the attention distribution of the model, and makes the evaluation behavior controllable. S3 quantitatively inspects and attributes the evaluation results by fusing multi-dimensional information. S4 substitutes the risk, credibility and resource state into a multi-objective optimization model to solve the optimal decision. S5 uses real feedback to jointly iteratively optimize the perception, evaluation and decision whole-link parameters, solves the pain points of the prior art "static black box of evaluation model, disconnection of decision and resources, and lack of self-evolution of the system", realizes controllable, interpretable, optimal and adaptive risk evaluation, and improves the precision, intelligence level and long-term applicability of management as a whole.

[0063] More specifically, in the quality evaluation and credible traceability step, the "comprehensive credibility score" is calculated by the following formula: C = a * (1 - Unorm) + b * Qdata + g * Ahistory, wherein C is the comprehensive credibility score, Unorm is the normalized model uncertainty estimate, Qdata is the comprehensive quality score of the input data, Ahistory is the accuracy rate of the historical evaluation results of the unit, a, b, g are adjustable weight coefficients, and a + b + g = 1. In this way, the subjective "credibility" is converted into an objective and calculable index, which provides a key basis for decision priority ranking and manual review, and significantly enhances the acceptability and practicality of the AI system output results.

[0064] More specifically, in the spatio-temporal perception fusion step, the "construction of a multi-scale spatio-temporal grid based on line topology and risk propagation model" specifically includes:

[0065] S11: basic grid construction, according to the coordinates of the tower, the sag of the conductor and the real-time meteorological field data, a basic topological grid reflecting the electrical connection and the physical space risk conduction path is constructed;

[0066] S12: risk model calling, for at least one to-be-evaluated risk factor in the mountain fire, icing and external force damage, a pre-set risk propagation model of the risk factor is called, and the model defines the diffusion speed, attenuation characteristics and spatial influence function of the factor;

[0067] S13: multi-scale window fusion, on the basis of the topological grid, according to the dynamic influence range output by the called risk propagation model, a multi-scale spatio-temporal analysis window corresponding to each risk factor is generated in parallel, and these windows are superimposed and fused in the spatio-temporal dimension, and finally a unified multi-scale spatio-temporal grid is formed;

[0068] Thus, the data-aware grid is strictly matched with the risk physical propagation law, the relevance and rationality of feature fusion are improved from the source, the system's ability to capture the spatio-temporal evolution trend of potential risks is enhanced, and the false positives are reduced

[0069] More specifically, in the synergistic optimization evaluation step, the working mechanism of the "synergistic optimization controller" is as follows: receiving an external input optimization target vector, generating a set of spatial attention modulation coefficients through a lightweight strategy network, and applying the coefficients to the key-value pairs of the global attention path in the dynamic risk analysis model to achieve dynamic reweighting of the risk attention degree in different geographical regions. As a "reflex center", the controller decodes the external abstract optimization target vector into a set of spatial attention modulation coefficients through a lightweight network, and directly applies the coefficients to the key-value pairs of the model attention mechanism to achieve dynamic spatial reweighting of the evaluation focus. Thus, the same evaluation model can flexibly adjust its evaluation focus according to different real-time business targets (such as "power supply" or "disaster prevention"), achieving "one model, multiple strategies" and improving the flexibility and scenario adaptability of the evaluation.

[0070] More specifically, in the closed-loop evolution step, the "joint optimization" is implemented using a three-layer optimization framework:

[0071] Inner optimization: fixing the fusion weight and function weight, updating the model parameters using feedback data;

[0072] Middle optimization: based on the updated model, optimizing the dynamic weight configuration of the multi-objective function;

[0073] Outer optimization: evaluating the long-term system performance under different fusion weights, searching for and updating the optimal fusion weight;

[0074] Thus, the system parameters evolve from micro to macro in a coordinated and safe manner, avoiding suboptimal solutions or system instability that may be caused by single-level optimization, and ensuring the comprehensiveness and robustness of the closed-loop learning process.

[0075] Please refer to Figs. 2-6 , the power line inspection area dynamic risk grading evaluation system based on the Conformer model, comprising:

[0076] A multi-scale spatio-temporal perception fusion module is used to perform the spatio-temporal perception fusion step, implement grid construction based on risk propagation mechanism, and fuse multi-source heterogeneous data;

[0077] A synergistic optimization intelligent evaluation module is used to perform the synergistic optimization evaluation step, integrate a controllable deep learning model and an optimization controller, and output risk quantification values and uncertainties;

[0078] a quality assessment and credible traceability module configured to perform the quality assessment and credible traceability step, calculate the credibility of the assessment result, and perform attribution analysis on a high-risk conclusion;

[0079] a multi-target resource constraint decision-making module configured to perform the resource constraint decision-making step, and solve the optimal risk level and operation and maintenance strategy under multiple constraints;

[0080] a perception-decision closed-loop evolution module configured to perform the closed-loop evolution step, and realize automatic iteration and optimization of key parameters by using system feedback;

[0081] In this way, the components and function division of the system are defined, facilitating actual deployment, integration, and definition of the scope of protection.

[0082] More specifically, the multi-scale spatio-temporal perception fusion module comprises:

[0083] a topology and risk model library unit storing line electrical topology data and standardized propagation models of multiple risk factors;

[0084] a multi-scale grid dynamic construction unit configured to call the model library and dynamically generate corresponding multi-scale spatio-temporal grids according to real-time risk factor types;

[0085] a graph fusion calculation unit configured to represent multi-source data as node and edge features on a graph, and perform graph neural network fusion calculation on the dynamically constructed grid;

[0086] The multi-scale spatio-temporal perception fusion module comprises a knowledge base (model library), a grid construction engine (dynamic construction unit), and a fusion calculation engine (graph fusion unit), which sequentially complete knowledge calling, dynamic grid generation, and graph neural network fusion calculation. Through modular design, the specific implementation path of the perception link is defined, the core technical path of “knowledge-driven dynamic grid + graph fusion calculation” is protected, and the implementability of the system is enhanced.

[0087] More specifically, the collaborative optimization intelligent assessment module comprises:

[0088] a gated spatio-temporal feature extraction network unit composed of parallel local convolution branches and global attention branches, and fused through a gating mechanism;

[0089] an embedded collaborative optimization controller unit integrated in the feature extraction network, and configured to generate an attention modulation coefficient in real time according to an input;

[0090] an uncertainty quantification output unit configured to synchronously output a comprehensive risk quantification value and its prediction uncertainty at the end of the model;

[0091] The cooperative optimization intelligent evaluation module integrates a feature extraction network (a gated temporal unit), a behavior controller (an embedded controller unit) and a result calibrator (an uncertainty quantification unit) to jointly complete controllable feature extraction and risk assessment, protects the specific model architecture of the ''adjustable network + embedded controller'', and ensures the advancement and controllability of the evaluation link.

[0092] More specifically, the perception-decision closed loop evolution module includes:

[0093] A feedback performance collection and analysis unit is configured to collect strategy execution results and calculate actual disposal performance indicators.

[0094] A three-layer cooperative optimization engine unit is configured to execute the three-layer optimization framework as claimed in claim 5 and coordinate the update of model parameters, function weights and fusion weights.

[0095] A strategy knowledge graph self-evolution unit is configured to automatically convert verified efficient decision cases into new knowledge and store the new knowledge in the strategy knowledge graph.

[0096] The perception-decision closed loop evolution module includes a feedback collector (collection and analysis unit), an optimization executor (three-layer optimization engine unit) and a knowledge management library (self-evolution unit), realizes the whole process from effect collection, parameter optimization to knowledge sedimentation, clearly defines the specific execution unit of system self-evolution, protects the complete learning loop of ''feedback driving + three-layer optimization + knowledge evolution'', and ensures that the system can continuously accumulate experience and autonomously improve.

[0097] More specifically, the system further includes a panoramic visualization and interactive cockpit module configured to:

[0098] In the digital twin scene, live risk distribution, resource scheduling, credibility labeling and traceability evidence chain are synchronously displayed.

[0099] An interactive interface is provided for operation and maintenance personnel to adjust and optimize targets, review system decisions and inject domain knowledge.

[0100] The panoramic visualization and interactive cockpit module integrally and visually presents the internal state (risk, credibility, traceability chain), decision process (resource scheduling) and external environment of the system on the digital twin base, and provides a human-computer interaction interface, realizes the transparency of system running state and the visualization of decision process, builds a hybrid augmented intelligent interaction mode of ''man-in-loop'', and greatly improves the explainability, trustworthiness and artificial supervision efficiency of the system.

[0101] From the above, the specific implementation of the present application is as follows:

[0102] The dynamic, reliable and optimized management and autonomous evolution of the power transmission line risk is realized through five closely connected links. Firstly, in the space-time perception fusion link, the system dynamically constructs a multi-scale space-time analysis grid matched with the physical topology structure of the power transmission line and the propagation mechanism model of different risk factors (such as forest fire and icing), and then performs graph structure alignment and adaptive weighted fusion of the multi-source heterogeneous inspection data on the grid to generate a unified feature representation containing physical laws and space-time correlations, laying the physical foundation for data-driven;

[0103] Secondly, in the cooperative optimization evaluation link, the feature representation is input into a dynamic risk analysis model (such as Conformer) with an embedded “cooperative optimization controller”; the controller receives external real-time optimization goals (such as “power supply protection” or “disaster prevention”), generates spatial attention modulation coefficients through a lightweight strategy network, dynamically reweights the attention degree of the model to different areas of the line, thereby realizing on-demand adjustment of the evaluation strategy, and simultaneously outputs the risk quantization value and its uncertainty estimate;

[0104] Then, in the quality evaluation and reliable traceability link, the system calculates the comprehensive reliability score of the evaluation result by comprehensively considering the model uncertainty, data quality and historical accuracy, and automatically traces the dominant risk factor and key evidence chain for the low-reliability high-risk result, realizing the explainability of the result and the auditability of the decision. Then, in the resource-constrained decision-making link, the system inputs the risk value, reliability and real-time resource state into a multi-objective optimization function that integrates accuracy, response time and resource consumption to solve, and outputs the optimal dynamic risk level under the current constraints and the refined operation and maintenance strategy bound thereto, realizing the global optimal mapping from risk perception to resource scheduling;

[0105] Finally, in the closed-loop evolution link, the system executes the strategy and collects feedback, based on the actual performance and prediction bias, starts a three-layer joint optimization process: inner-layer optimization of model parameters, middle-layer adjustment of multi-objective function weights, and outer-layer search of optimal data fusion weights, thereby driving the cooperative self-adaptation of the perception, evaluation and decision-making whole-link parameters, enabling the system to continuously learn and evolve from real operation and maintenance feedback. The whole process forms a complete intelligent closed loop of “enhancing perception by physical mechanism, target-oriented regulation and evaluation, multi-dimensional result quality inspection, optimized decision-making under constraint conditions, and whole-link iteration driven by practice feedback”.

[0106] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can modify the technical solutions described in the foregoing embodiments or make equivalent replacements to some technical features. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for dynamic risk classification and assessment of transmission line inspection areas based on the Conformer model, characterized in that, Includes the following steps: S1: Spatiotemporal perception fusion, acquire multi-source inspection data of the target transmission line corridor, construct a multi-scale spatiotemporal grid based on the line topology and risk propagation model, and perform heterogeneous graph alignment and adaptive weighted fusion in the grid to generate a unified feature representation; S2: Collaborative optimization assessment, inputting the unified feature representation into the dynamic risk analysis model, and dynamically adjusting the model attention distribution according to the real-time evaluation target through its internal collaborative optimization controller, outputting the comprehensive risk quantification value and uncertainty estimate of each inspection unit; S3: Quality assessment and credible traceability: Based on the aforementioned uncertainty estimation, data source quality indicators, and historical assessment accuracy, calculate the comprehensive credibility score of the risk assessment results for each inspection unit. For high-risk assessment results with a credibility level below the threshold, their dominant risk factors and key evidence data are automatically traced and visualized. S4: Resource constraint decision-making. Construct a multi-objective optimization function that integrates assessment accuracy, response latency, and resource consumption. Solve the function based on the comprehensive risk quantification value, comprehensive credibility score, and real-time resource status to obtain the dynamic risk level and the corresponding differentiated operation and maintenance strategy. S5: Closed-loop evolution, execute the operation and maintenance strategy, collect feedback data, and jointly optimize the parameters of the dynamic risk analysis model, the fusion weights of the multi-scale spatiotemporal grid, and the weights of the multi-objective optimization function based on strategy effectiveness and prediction deviation.

2. The method for dynamic risk classification and assessment of transmission line inspection areas based on the Conformer model according to claim 1, characterized in that, In the quality assessment and credible tracing steps, the "overall credibility score" is calculated using the following formula: C = α⋅(1−Unorm) + β⋅Qdata + γ⋅Ahistory, where C is the overall credibility score, Unorm is the normalized model uncertainty estimate, Qdata is the overall quality score of the input data, Ahistory is the accuracy of the historical evaluation results of this unit, and α, β, γ are adjustable weight coefficients, and α + β + γ = 1.

3. The method for dynamic risk classification and assessment of transmission line inspection areas based on the Conformer model according to claim 1, characterized in that, In the spatiotemporal perception fusion step, the "construction of a multi-scale spatiotemporal grid based on the line topology and risk propagation model" specifically includes: S11: Basic grid construction: Based on tower coordinates, conductor sag and real-time meteorological field data, a basic topology grid is constructed to reflect the transmission path of electrical connection and physical space risks. S12: Risk model invocation. For at least one risk factor to be assessed, namely wildfire, ice cover, or external damage, the preset risk propagation model is invoked. The model defines the diffusion rate, decay characteristics, and spatial influence function of the factor. S13: Multi-scale window fusion: On the basic topological grid, based on the dynamic influence range output by the called risk propagation model, multi-scale spatiotemporal analysis windows corresponding to each risk factor are generated in parallel, and these windows are superimposed and fused in the spatiotemporal dimension to finally form a unified multi-scale spatiotemporal grid.

4. The method for dynamic risk classification and assessment of transmission line inspection areas based on the Conformer model according to claim 1, characterized in that, In the collaborative optimization evaluation step, the working mechanism of the "collaborative optimization controller" is as follows: it receives the optimization target vector from the external input, generates a set of spatial attention modulation coefficients through a lightweight policy network, and applies the coefficients to the key-value pairs of the global attention path in the dynamic risk analysis model to realize the dynamic reweighting of risk attention in different geographical areas.

5. The method for dynamic risk classification and assessment of transmission line inspection areas based on the Conformer model according to claim 1, characterized in that, In the closed-loop evolution step, the "joint optimization" is implemented using a three-layer optimization framework: Inner layer optimization: Fix the fusion weights and function weights, and update the model parameters using feedback data; Mid-level optimization: Based on the updated model, optimize the dynamic weight configuration of the multi-objective function; Outer layer optimization: Evaluate the long-term system performance under different fusion weights, and search for and update the optimal fusion weights.

6. A dynamic risk classification and assessment system for transmission line inspection areas based on the Conformer model for implementing the method of any one of claims 1 to 5, characterized in that, include: The multi-scale spatiotemporal perception fusion module is used to execute the spatiotemporal perception fusion steps to realize grid construction and multi-source heterogeneous data fusion based on risk propagation mechanism; The collaborative optimization intelligent assessment module is used to execute the collaborative optimization assessment steps, integrates a controllable deep learning model and an optimization controller, and outputs risk quantification values ​​and uncertainties. The quality assessment and credible traceability module is used to execute the quality assessment and credible traceability steps, calculate the credibility of the assessment results, and perform attribution analysis on high-risk conclusions. The multi-objective resource constraint decision module is used to execute the resource constraint decision steps and solve for the optimal risk level and operation and maintenance strategy under multiple constraints. The perception-decision closed-loop evolution module is used to execute the closed-loop evolution steps and utilize system feedback to achieve automatic iteration and optimization of key parameters.

7. The dynamic risk classification and assessment system for transmission line inspection areas based on the Conformer model according to claim 6, characterized in that, The multi-scale spatiotemporal perception fusion module includes: The topology and risk model library unit stores line electrical topology data and standardized propagation models for various risk factors; Multi-scale grid dynamic building blocks are used to call the model library and dynamically generate corresponding multi-scale spatiotemporal grids according to the real-time risk factor type. The graph fusion computing unit is used to represent multi-source data as graph node and edge features, and to perform graph neural network fusion computing on the dynamically constructed grid.

8. The dynamic risk classification and assessment system for transmission line inspection areas based on the Conformer model according to claim 6, characterized in that, The collaborative optimization intelligent evaluation module includes: The gated spatiotemporal feature extraction network unit consists of parallel local convolutional branches and global attention branches, which are fused through a gating mechanism. An embedded collaborative optimization controller unit is integrated into the feature extraction network to generate attention modulation coefficients in real time based on the input; The uncertainty quantification output unit is used to simultaneously output the comprehensive risk quantification value and its prediction uncertainty at the end of the model.

9. The dynamic risk classification and assessment system for transmission line inspection areas based on the Conformer model according to claim 6, characterized in that, The perception-decision closed-loop evolution module includes: The feedback performance collection and analysis unit is used to collect strategy execution results and calculate actual handling performance indicators. The three-layer collaborative optimization engine unit is used to execute the three-layer optimization framework as described in claim 5, and coordinate the updates of model parameters, function weights and fusion weights. The self-evolving unit of the strategy knowledge graph is used to automatically transform validated and efficient decision-making cases into new knowledge and store them in the strategy knowledge graph.

10. The dynamic risk classification and assessment system for transmission line inspection areas based on the Conformer model according to claim 6, characterized in that, The system also includes a panoramic visualization and interactive cockpit module, used for: In a digital twin scenario, real-time risk distribution, resource allocation, credibility labeling, and traceability evidence chain are displayed simultaneously; Provides an interactive interface for operations and maintenance personnel to adjust and optimize goals, review system decisions, and inject domain knowledge.

Images