10kV overhead line self-adaptive evaluation method and system for non-power-cut operation

Abstract

The invention discloses a 10kV overhead line adaptive evaluation method and system for non-power-cut operation, and the method comprises the following steps: obtaining the field data of a target line, and constructing the field data into a multi-dimensional evaluation feature vector according to a preset evaluation dimension system for non-power-cut operation; inputting the multi-dimensional evaluation feature vector into a non-power-cut operation adaptability evaluation model to obtain a comprehensive evaluation score and an optimization suggestion of the target line; outputting the comprehensive evaluation score and the optimization suggestion; wherein the non-power-cut operation adaptability evaluation model is constructed by the following steps: training to obtain an initial machine learning model; simulating a preset non-power-cut operation task flow in the digital twinborn body; and performing incremental training and parameter optimization on the initial machine learning model by using operation effect data fed back after actual operation of the simulation data vector. The problems of single evaluation dimension, model stiffness and the like in the prior art can be solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance and safe operation technology of power systems, specifically to an adaptive evaluation method and system for 10kV overhead lines for uninterrupted power supply operations. Background Technology

[0002] Live-line work on power distribution networks is a key technology for improving power supply reliability. The safety and efficiency of this work highly depend on accurate assessments of the line's own conditions, environmental conditions, and the work process itself. Traditional line evaluation methods are mostly based on manual experience or simple static procedural checks, which have the following prominent problems: 1) The evaluation dimensions are singular and static, mainly relying on design drawings and periodic inspection data; 2) The weights of evaluation indicators are usually set based on experience, and the evaluation methods are rigid and difficult to adapt to the differences in different regions and working conditions. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of existing technologies, such as single and static evaluation dimensions and rigid evaluation methods, and to provide an adaptive evaluation method and system for 10kV overhead lines for live-line work. This solution aims to deeply integrate professional domain knowledge, digital twin simulation verification, and incremental learning from machine learning to construct an evaluation system that can continuously evolve and provide quantitative decision support, thereby solving the problems of single evaluation dimensions, rigid models, and disconnect from practical work in existing evaluation methods.

[0004] On the one hand, this invention provides an adaptive evaluation method for 10kV overhead lines for uninterrupted power supply operations, comprising the following steps: S1: Acquire the field data of the target line, and construct the field data into a multi-dimensional evaluation feature vector based on the preset evaluation dimension system for live-line work; wherein, the evaluation dimension system includes at least the tower structure layer dimension, the electrical clearance layer dimension, the work-friendly layer dimension, and the environmental adaptation layer dimension. S2: Input the multi-dimensional evaluation feature vector into the live-line operation adaptability evaluation model to obtain the comprehensive evaluation score and optimization suggestions for the target line; S3: Output the comprehensive evaluation score and optimization suggestions; The live-line work adaptability evaluation model is constructed in the following way: S21: Based on the multidimensional evaluation feature vectors of the line samples and the corresponding expert evaluation labels, the initial machine learning model is trained; S22: For the target line, construct its digital twin, and simulate the preset uninterrupted power supply operation task process in the digital twin to generate a simulation data vector containing operation efficiency, operation complexity and virtual risk indicators. S23: Using the simulation data vector and the operation effect data fed back after the actual operation on the target line, the initial machine learning model is incrementally trained and the parameters are optimized to form the live-line operation adaptability evaluation model.

[0005] Furthermore, the characteristic parameters corresponding to the tower structure layer dimension include crossarm type, number of hanging points, and tower health; the characteristic parameters corresponding to the electrical clearance layer dimension include phase-to-phase distance, distance to ground, and safe working space margin index; the characteristic parameters corresponding to the work-friendliness layer dimension include hardware disassembly and assembly convenience index, work platform accessibility index, and standard shielding operation complexity index; and the characteristic parameters corresponding to the environmental adaptability layer dimension include wind speed and wind direction influence coefficient, humidity and condensation influence coefficient, and regional pollution level coefficient.

[0006] Furthermore, the acquisition of on-site data of the target route includes: obtaining static ledger parameters of the target route from the production management system, collecting dynamic environmental parameters of the target route in real time through sensors installed on poles and towers, and obtaining corridor status parameters of the target route through drone inspection and image recognition.

[0007] Furthermore, the acquisition of on-site data of the target route specifically includes: obtaining static ledger parameters of the route from the production management system through an application programming interface; collecting dynamic environmental parameters in real time through IoT sensors deployed on poles; and conducting inspections using visible light and lidar payloads carried by drones, and analyzing the acquired images and point cloud data based on computer vision models to extract route corridor status parameters.

[0008] Furthermore, in step S1, when constructing the field data into a multidimensional evaluation feature vector, a combined weighting method is used to determine the weights of each dimension and the underlying feature parameters. The combined weighting method integrates the subjective weights obtained based on the analytic hierarchy process and the objective weights obtained based on the entropy weighting method.

[0009] Furthermore, in step S21, the initial machine learning model adopts one of gradient boosting decision tree, random forest or deep neural network model; the expert evaluation label is the line live-line operation safety level determined based on the multi-expert Delphi method.

[0010] Furthermore, step S22 specifically includes: Based on the geometric and electrical parameters of the target line, a digital twin of the target line is constructed in the digital twin platform; The digital twin incorporates parameterized virtual models of standard operating personnel, insulating tools, and shielding equipment. The virtual model is driven to execute a preset uninterrupted power supply operation task process, and the tool path, operation time, virtual distance parameters, and potential risk events determined by the built-in rule engine are recorded simultaneously. Based on the recorded data, the operation efficiency, operational complexity, and virtual risk indicators are calculated, and a simulation data vector is formed.

[0011] Furthermore, the calculation of virtual risk indicators includes: Based on the simplified electric field calculation module integrated in the digital twin platform, virtual risk indicators are calculated; The simplified electric field calculation module is used to simulate and estimate the changes in electric field intensity at preset key locations during the operation. Furthermore, the operation performance data fed back after the actual operation on the target line in step S23 includes the actual operation time, unexpected difficulties encountered, and the subjective ratings of the operators on the effectiveness of the optimization suggestions provided by the model; the incremental training updates the model parameters using online learning or mini-batch retraining.

[0012] Furthermore, the output optimization suggestions refer to suggestions generated by integrating the comprehensive evaluation score, digital twin simulation results, and model decision-making basis. The optimization suggestions include at least: a list of unfriendly towers or sections and their specific risk causes, simulation-verified hardware modification or work tool configuration schemes, dynamic work risk warnings based on current environmental parameters, and recommended optimal work methods; Furthermore, step S23 also includes a model co-evolution step: by deploying a federated learning framework, multiple local live-line work adaptive evaluation models deployed in different power supply areas are incrementally trained on local data, and only the encrypted live-line work adaptive evaluation model parameters are updated and uploaded to the central server for secure aggregation, forming a globally optimized new generation of live-line work adaptive evaluation model and synchronizing it to each local system.

[0013] On the other hand, the present invention also discloses an adaptive evaluation system for 10kV overhead lines for uninterrupted power supply operations, used to implement the method described above, including: The data acquisition and feature construction module is used to acquire on-site data of the target route and construct multi-dimensional evaluation feature vectors based on a preset evaluation dimension system. An adaptive evaluation model engine is used to store and run the aforementioned uninterrupted power supply (UPS) adaptive evaluation model. The decision support and output module is used to output and display the comprehensive evaluation score and optimization suggestions. The construction and optimization of the evaluation model are achieved through the following sub-modules: Initial training unit, used to train an initial machine learning model based on historical data; The digital twin simulation and enhancement unit is used to construct a digital twin of the target line, simulate the operation process, and generate simulation data vectors. The incremental learning and optimization unit is used to integrate simulation data and actual operation feedback to perform incremental training and optimization of the model.

[0014] Furthermore, the digital twin simulation and enhancement unit also includes a configurable task library and a rule engine. The task library is used to define standard operating procedures, and the rule engine is used to determine operational compliance and risks in real time during the simulation process.

[0015] Furthermore, it also includes a federated learning client module, which is embedded in the model self-learning optimization module. After completing the training of the uninterrupted power supply operation adaptability evaluation model locally, it collaborates with the central server to complete the federated learning process.

[0016] Furthermore, the decision support and visualization output module also includes an augmented reality interface, which is used to convert key optimization suggestions, risk warnings, or simulation-verified work steps into guidance information that can be displayed on augmented reality devices.

[0017] Compared with the prior art, the present invention has at least the following beneficial effects: 1) By using a pre-defined four-dimensional evaluation system to guide data construction, the evaluation is ensured to fully cover the key factors affecting operational safety and efficiency. The evaluation system is more systematic and professional, fundamentally different from general evaluation methods.

[0018] 2) An adaptive evaluation model for uninterrupted power supply operations with dynamic evolution capability is constructed through a three-stage method of initial training, simulation enhancement, and feedback optimization. This model can continuously absorb new knowledge (digital twin simulation exploration) and new experience (actual operation feedback), realizing the leap from a static tool to a growth-type intelligent agent, and effectively avoiding the impact of potential degradation in model performance during use.

[0019] 3) Digital twin simulation is not only used for model enhancement, but also for visual and quantifiable virtual verification of optimization suggestions. Because decision support is quantifiable and verifiable, the output optimization suggestions are more specific, credible, and actionable, truly realizing a closed loop from evaluation to decision support.

[0020] 4) By deeply integrating professional knowledge in the power field (tower structure, electrical clearance, operation friendliness and environmental adaptability), computational electromagnetics (simulation), machine learning and incremental learning, federated learning and other cutting-edge technologies, a complete solution with high technical barriers has been formed to address the specific problem of rigidity in existing evaluation methods. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0022] Figure 1 This is a flowchart illustrating the steps of an adaptive evaluation method for 10kV overhead lines for uninterrupted power supply operations provided in an embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram illustrating the process principle of the adaptive evaluation method for 10kV overhead lines for uninterrupted power supply operations provided in this embodiment of the invention.

[0024] Figure 3 This diagram illustrates the three-stage construction and optimization of the adaptive evaluation model for live-line work in the 10kV overhead line adaptive evaluation method for live-line work provided in this embodiment of the invention.

[0025] Figure 4 This is a block diagram of the architecture of a comprehensive evaluation system for 10kV overhead lines for uninterrupted power supply operations, provided in an embodiment of the present invention.

[0026] Figure 5 This is a schematic diagram of the digital twin simulation interface in the adaptive evaluation method for 10kV overhead lines for uninterrupted power supply operations provided in an embodiment of the present invention. Detailed Implementation

[0027] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0028] It should be understood that the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention. Method Implementation Examples

[0029] refer to Figure 1 and Figure 2 This invention provides an adaptive evaluation method for 10kV overhead lines for uninterrupted power supply operations, comprising the following steps: S1: Acquire on-site data of the target line and construct a multi-dimensional evaluation feature vector based on the preset evaluation dimension system for live-line work; wherein, the evaluation dimension system includes at least the tower structure layer dimension, electrical clearance layer dimension, work-friendly layer dimension, and environmental adaptability layer dimension; S2: Input the multi-dimensional evaluation feature vector into the live-line operation adaptability evaluation model to obtain the comprehensive evaluation score and optimization suggestions for the target line; S3: Output the overall evaluation score and optimization suggestions.

[0030] The live-line work adaptability evaluation model is constructed in the following way: S21: Based on the multidimensional evaluation feature vectors of the line samples and the corresponding expert evaluation labels, the initial machine learning model is trained; S22: For the target line, construct its digital twin, and simulate the preset uninterrupted power supply operation task process in the digital twin to generate a simulation data vector containing operation efficiency, operation complexity and virtual risk indicators. S23: Using simulation data vectors and feedback data on the operation results after actual operation on the target line, incremental training and parameter optimization are performed on the initial machine learning model to form an adaptive evaluation model for live-line operation.

[0031] like Figure 2 This paper demonstrates the core process and self-optimization principle of the adaptive evaluation method for 10kV overhead lines for uninterrupted power supply operations. The schematic diagram presents a complete technical closed loop of data-driven, model evaluation, simulation verification, and feedback optimization in a cyclical evolution structure. The main evaluation process on the left (start → step S1 → step S2 → step S3 → end) describes a linear operational chain from field data collection and feature construction to model evaluation decision generation and final output results. The model construction and continuous optimization process on the right reveals the self-evolutionary mechanism supporting the evaluation function. It takes historical data, target line data, and actual operation feedback as inputs, and sequentially updates the initial machine learning model through three stages: S21 initial model training, S22 digital twin simulation, and S23 incremental learning optimization, ultimately forming an optimized evaluation model, which in turn strengthens the S2 stage in the main process. The diagram clearly shows, through arrows and data flow, that simulation data and actual feedback constitute the dual engines for continuous learning and optimization of the model. This enables the system to integrate expert experience, virtual verification, and actual operational feedback, achieving a leap from a static evaluation tool to a dynamic, evolving intelligent agent. This ensures that the live-line work adaptability evaluation model can autonomously adapt and optimize as the working environment, equipment status, and technology evolve.

[0032] Step S1 is based on a four-layer evaluation dimension system. In Step S1, a specialized evaluation dimension system for the specific scenario of live-line work is systematically constructed through the Delphi method, procedure analysis, and historical accident case mining. This dimension system consists of four dimensions: physical foundation (tower structure layer), electrical safety boundary (electrical clearance layer), process convenience (work-friendly layer), and external disturbance (environmental adaptation layer). This forms a complete multi-dimensional coverage from "things" to "people" to "environment," aiming to comprehensively cover all key dimensions affecting operational safety and efficiency.

[0033] In the four-layer evaluation system, the tower structure layer mainly includes crossarm type (assigned score), number of hanging points (counting index), and tower health (corrosion identification score based on images). The electrical clearance layer mainly includes phase-to-phase distance margin ratio, ground distance margin ratio, and three-dimensional safe working space margin index. The work-friendliness layer mainly includes the ease of hardware disassembly and assembly (expert scoring), accessibility of multi-path work platforms (optimal path score), and complexity of standard shielding operations (step and time estimation). The environmental adaptability layer mainly focuses on the influence coefficients of wind speed and wind direction (piecewise function), the influence coefficients of humidity and condensation (insulation reduction model), and the regional pollution level coefficient (table lookup).

[0034] In an embodiment of the present invention, the following example is taken: a municipal power supply company plans to conduct an assessment of the adaptability of uninterrupted power supply operations before carrying out the annual comprehensive overhaul of the 10kV "Binhu Line".

[0035] Step S1: Construct an evaluation dimension system, which mainly includes the following process in its implementation: 1) The provincial live-line work center took the lead in organizing an expert group of 10 people, including front-line team leaders, technical experts, and safety supervisors.

[0036] 2) Conduct discussions through at least two rounds of the Delphi method: In the first round, experts freely propose all factors that affect the safety and efficiency of live-line work; in the second round, the importance of the four major categories and fifteen specific indicators initially summarized are ranked and their definitions are clarified.

[0037] 3) Finally, a written "10kV Overhead Line Live-Line Operation Adaptability Evaluation Index System Specification" was formed.

[0038] In addition, the evaluation index system determined by the expert group for the Binhu Line is shown in the following example. In the embodiment of the present invention, the target line "Binhu Line" includes a total of 45 towers.

[0039] The structural layer dimension (C1) of the pole and tower mainly includes the crossarm type (C11), the number of hanging points (C12), and the pole and tower health (C13) indicators.

[0040] C11 crossarm type rating, for example, insulated crossarms: #1-#30 double angle iron, 0.7; #31-#45 single angle iron, 0.4; this means that in the crossarm type, #1 to #30 are double angle iron crossarms (assigned a value of 0.7), and #31 to #45 are single angle iron crossarms (assigned a value of 0.4).

[0041] The available hanging point index for C12 is as follows: #10, quantity ≥ 4, 1.0; #12, quantity = 3, 0.7; #25, quantity = 2, 0.4; #33, quantity ≤ 1, 0.1. This means that the available hanging point index counts the number of secure hanging points per pole that can be used to suspend insulated ropes or shielding tools. Specifically, #10 pole has ≥ 4 hanging points (index 1.0), #12 pole has 3 hanging points (index 0.7), #25 pole has 2 hanging points (index 0.4), and #33 pole has ≤ 1 hanging point (index 0.1).

[0042] C13 tower health rating (based on drone imagery corrosion identification and rating, 0-1 points). This rating is primarily based on last year's drone inspection report; for example, towers #5 and #18 showed moderate corrosion, with health ratings dropping to 0.6 and 0.55 respectively.

[0043] The electrical clearance layer dimension (C2) mainly includes the phase-to-phase distance margin ratio (C21), ground distance margin ratio (C22), and three-dimensional safe operating space margin (C23) indicators.

[0044] C21 phase-to-phase distance margin ratio (measured distance / minimum distance in the "Safety Regulations"), for example, the measured average phase-to-phase distance of the Binhu Line is 0.75 meters, compared with the "Safety Regulations" requirement of 0.7 meters, the margin ratio is 1.07.

[0045] C22 Ground Distance Margin Ratio, for example, through lidar measurement, it was found that due to tree growth, the minimum ground distance within the #22 and #23 spans is 4.3 meters (the specification requires 4.0 meters), and the margin ratio is 1.075.

[0046] C23 Three-dimensional safety work space margin is calculated using point cloud data or models to determine the minimum net distance. For example, for the "replacing a side phase insulator" operation, the space required for the operator (including the range of movement) is simulated in the three-dimensional model, and the minimum distance between the operator and the middle phase conductor is calculated to be 0.5 meters (meeting the requirements).

[0047] The operation-friendly layer dimension (C3) mainly includes the ease of hardware assembly and disassembly (C31), accessibility of multi-path operation platforms (C32), and standard shielded operation complexity (C33) index.

[0048] The ease of assembly and disassembly of typical C31 fittings (experts scored bolt orientation and tool space). For example, after on-site inspection, experts scored the orientation of typical tension clamp bolts and wrench operating space. #8 scored 0.8 (easy) and #35 scored 0.4 (difficult, requires special tools).

[0049] The C32 multi-path operation platform accessibility (evaluating the highest score for methods such as boom trucks and pole climbing) was assessed. For example, the assessment found that pole #15 was located by the fishpond and could not be approached by an insulated boom truck, so pole climbing was the only option, reducing the accessibility index to 0.5.

[0050] C33 standard shielding operation complexity (number of insulating shields required and estimated steps), for example, simulating "live disconnection of lead wires" operation, counting the number of insulating blankets and shields that need to be installed, #40 pole (tension pole) requires 7 pieces, the complexity index is 0.8 (high).

[0051] The environmental adaptation layer dimension (C4) mainly includes the wind speed and wind direction influence coefficient (C41), humidity and condensation influence coefficient (C42), and regional pollution level coefficient (C43).

[0052] C41 Wind speed and direction influence coefficient (piecewise function calculation). For example, according to local historical meteorological data, the area often experiences winds of level 3-5 (5.5-10.7 m / s) in the afternoon during spring, and the influence coefficient is preset to 0.7-0.9.

[0053] C42 Humidity and Condensation Influence Coefficient (Considering Insulation Reduction Due to Condensation). For example, given the line's location near a lake, the morning humidity is typically greater than 85%, resulting in a condensation influence coefficient of 0.6, requiring a reduction in insulation strength.

[0054] The pollution level coefficient for area C43 (obtained from the chart). For example, according to the pollution area distribution map, the lakeside area is in a Class D pollution area (with more industrial dust), and the pollution level coefficient is set at 0.7.

[0055] The core of step S1 is to acquire and integrate static and dynamic field data of the target line through multi-source data fusion technology. The target line is a 10kV overhead line designed for uninterrupted power supply operations. Specifically, the system collects multi-source heterogeneous data of the target line from different data sources through a hybrid sensing approach combining application programming interface (API) calls, Internet of Things (IoT) sensor networks, UAV inspections, and computer vision analysis, providing a unified quantitative input for subsequent feature construction. Static parameters (tower type, conductor sag, etc.) come from the Production Management System (PMS). For example, through the API interface, the system automatically retrieves the ledger data of all towers on the Binhu line from the company's PMS 2.0 system, forming a structured table. Dynamic parameters (such as wind speed and humidity) come from IoT sensors. For example, solar-powered miniature weather stations (including wind speed, wind direction, temperature, and humidity sensors) are installed on key towers such as #10, #25, and #40 on the Binhu line, and the data is transmitted back every 5 minutes via a 4G network. The status of the corridor (such as tree obstructions and building distances) is obtained from drone image recognition. For example, a multi-rotor drone equipped with a dual-light (visible light + LiDAR) pod is dispatched to conduct a detailed inspection of all 45 towers along the route. After the flight, the data is automatically uploaded to a cloud processing platform. The cloud processing platform uses a pre-trained deep learning model (such as YOLOv5) to automatically identify trees and buildings in the drone photos and measure their closest distances to the guide wire. Simultaneously, it calculates the percentage of projected vegetation cover area along the entire route.

[0056] The multi-source data fusion mentioned above is mainly reflected in three levels: 1) Source layer fusion. Static structured data comes from enterprise back-end databases such as production management systems (PMS) and asset management systems, and is extracted periodically or triggered by API interfaces (such as RESTful API) or database middleware. Dynamic time-series data comes from IoT sensor networks deployed on poles or lines, including micro-meteorological sensors (wind speed, wind direction, temperature and humidity), image sensors, tilt sensors, etc. Dynamic data is transmitted back to the cloud platform or edge server in real time or near real time via 4G / 5G or dedicated power wireless networks (such as RF mesh, LoRa). This is the variable for evaluation. Spatial imagery and point cloud data come from UAV automated inspection. UAVs are equipped with visible light cameras, infrared thermal imagers, and lidar (LiDAR) to collect high-resolution images and 3D point clouds. This is the "scenario" for evaluation. 2) Data transmission and protocol fusion. The system needs to be compatible with multiple communication protocols, such as MQTT / CoAP for sensor data, HTTP / HTTPS for inter-system interaction, and dedicated image transmission or high-speed download protocols for processing UAV data, so as to aggregate the data into a unified data platform or cloud platform. 3) Data processing and feature extraction fusion, mainly including data cleaning and alignment (timestamp synchronization, unit unification, and outlier handling for data of different frequencies and formats), model-based analysis (for unstructured UAV imagery and point cloud data, deep learning models are used to automatically identify tree obstacles, buildings, hardware status, rusted areas, etc., and the identification results are quantified into specific feature parameters such as distance, area, and health score), and feature vector construction (normalizing and combining static parameters from PMS, dynamic readings from sensors, and quantitative indicators generated from image recognition according to a preset evaluation dimension system to form a complete, structured, multi-dimensional evaluation feature vector).

[0057] In the embodiments of this invention, static and dynamic parameters as shown in Table 1 were obtained. The static parameters are ledger data of 85 towers along the Binhu Line exported from the PMS system, including model, coordinates, conductor type, sag, and hardware list. The dynamic parameters are obtained by installing miniature weather stations on key towers such as #10, #25, and #40, which transmit real-time data on wind speed (3.2 m / s), wind direction (southeast), temperature (25℃), and humidity (72%) to the cloud platform via 4G data transfer units (DTU). Orthophotos and lidar scans were performed along the entire line using drones, and three potential tree obstruction points (closest distance less than 1.05 meters) were identified using AI algorithms. The vegetation coverage rate along the entire line was calculated to be 22% based on the image recognition results.

[0058] Table 1

[0059] The static and dynamic data obtained in step S1 above provide concrete and quantifiable data support for the abstract indicators in the evaluation dimension system, shifting the evaluation from qualitative to quantitative. Static data is the baseline of the evaluation, while dynamic data is the variable. The combination of dynamic and static data breaks the limitations of traditional evaluations that rely solely on design drawings or periodic inspections. Simultaneously, the introduction of dynamic data enables the evaluation to reflect the real environmental risks at the moment of operation. Without dynamic parameters, the model will be unable to perceive the immediate risks brought about by sudden environmental changes, losing its adaptive premise; conversely, without accurate static parameters, the evaluation will lose its foundation of accuracy. The combination of the two is the cornerstone for achieving scenario-based and real-time evaluation. Here, dynamic parameters can be understood as numerical or categorical variables obtained from dynamic data through extraction, cleaning, transformation, and calculation, capable of quantifying a specific state, attribute, or risk of the system. Dynamic parameters are prepared to solve specific problems (such as evaluation).

[0060] Preferably, in step S1, when constructing a multidimensional evaluation feature vector from the field data, a combined weighting method is used to determine the weights of each dimension and the underlying feature parameters. This combined weighting method integrates subjective weights obtained based on the analytic hierarchy process (AHP) and objective weights obtained based on the entropy weighting method. The weights determined by the combined weighting method are used to construct the multidimensional evaluation feature vector. Specifically, when converting field data into a feature vector, different dimensions and underlying feature parameters need to be weighted and integrated to reflect the degree of influence of each dimension indicator on the comprehensive evaluation. The subjective and objective combined weights obtained by the combined weighting method are used for this weighting calculation, thereby forming a more scientific and adaptive feature representation, which serves as the input to the live-line work adaptability evaluation model. In this step, firstly, domain experts are invited to determine the subjective weights using the analytic hierarchy process (AHP). For example, using professional software (such as Yaahp, an auxiliary modeling and calculation software for AHP and fuzzy comprehensive evaluation), experts are asked to compare the four primary indicators pairwise (e.g., the electrical clearance layer dimension is significantly more important than the work-friendliness layer dimension). The software automatically calculates the weight vector Ws = [0.28, 0.38, 0.22, 0.12] and passes a consistency check (e.g., CR = 0.05 < 0.1). Subjective weights reflect industry experience and procedural requirements. Secondly, based on historical data, objective weights are calculated using the entropy weight method. For example, the system retrieves data from 100 completed evaluation routes within the province over the past two years as a sample. Each underlying indicator value is standardized, and its information entropy Ej is calculated. The smaller the information entropy Ej, the greater the difference between different routes, and the greater the weight should be assigned. The calculated objective weight vector Wo = [0.22, 0.41, 0.25, 0.12]. Objective weights reflect the inherent differences and information content of the data. Finally, game theory and other methods are used to optimize the combination of subjective and objective weights. For example, using game theory, an optimization model is established to find a combination coefficient α that minimizes the sum of squared deviations between the combined weight W = α * Ws + (1-α) * Wo and Ws and Wo. The solution yields α = 0.55. The final combined weights W = [0.25, 0.40, 0.24, 0.11]. The calculation results show that, for the lines in this region, the electrical clearance layer dimension is assigned the highest weight (weight 0.40), which is consistent with the actual situation of prominent tree obstruction problems in this region, reflecting the advantage of combining subjective and objective factors.

[0061] The above optimized steps overcome the limitations of relying solely on expert experience (which may be subjective) or purely data-driven approaches (which may ignore the hard constraints of regulations). The Analytic Hierarchy Process (AHP) weighting ensures that the evaluation complies with the rigid requirements of safety regulations, while the entropy weighting method allows the evaluation to adapt to the differences in data characteristics of different lines. Furthermore, the combined weights are not fixed; when evaluating different regions and types of lines, the objective weights calculated based on their data characteristics are dynamically adjusted. Therefore, this step solves the problem of how to scientifically determine weights in traditional evaluations. If fixed weights are used, the model will not be able to adapt to the diverse characteristics of lines; if it is entirely data-driven (such as the pure entropy weighting method), it may violate basic safety principles. Combined weighting is the specific manifestation of the adaptive characteristics of this invention at the weighting level.

[0062] The core of step S2 is an uninterrupted operation adaptability evaluation model based on three stages: initial training, simulation enhancement, and feedback optimization. Figure 2 and Figure 3 As shown, this step is also the core innovative step of this invention. Figure 3 This paper demonstrates the three-stage construction and optimization evolution mechanism of the adaptive evaluation model for live-line work in this invention. The schematic diagram employs a structure combining horizontal progression and vertical expansion, clearly revealing the complete lifecycle of the model from basic training to intelligent evolution. The initial training stage on the left, based on a historical line database and expert evaluation label library, uses feature engineering and machine learning algorithms to train and construct an initial evaluation model with basic assessment capabilities. The digital twin simulation enhancement stage in the middle uses target line data as input to construct a high-fidelity digital twin in a virtual environment and simulate typical work tasks, generating simulation data containing indicators such as work time and operational difficulty, injecting virtual practical experience into the model. The model fusion and incremental learning stage on the right integrates simulation data with actual work feedback (including work time records, difficulty scores, and feedback on the effectiveness of suggestions), continuously optimizing model parameters through incremental learning to form an optimized evaluation model with adaptive capabilities. Specifically, Figure 3 The following illustrates the federated learning extension mechanism. After multiple branch offices' local models evolve on their local data, they only upload encrypted parameter updates to the central server for secure aggregation, forming a globally optimized next-generation model, which is then synchronously transmitted back. This achieves cross-regional knowledge sharing and collaborative evolution while protecting data privacy, demonstrating the technological advancement of this invention, which possesses both local adaptability and global optimization capabilities.

[0063] Step S2.1 mainly involves the initial training of the live-line work adaptability evaluation model. The initial machine learning model employs one of the following: gradient boosting decision tree, random forest, or deep neural network. A supervised learning model (such as XGBoost, an efficient gradient boosting decision tree algorithm) is trained using structured data (historical static parameters and historical dynamic parameters) of historical lines and their expert-assessed safety level labels. For example, complete feature vectors of 300 historical lines (i.e., data processed through steps S1 to S3) and their final safety level labels (A / B / C / D levels, post-assessed by a provincial expert group) are extracted from the knowledge base. The XGBoost algorithm is used for training, which automatically handles non-linear relationships and provides a ranking of feature importance. After training, the initial model file model_v1.0.pkl is saved. Step S2.1 constructs a seed model with preliminary predictive capabilities, which learns from historical experience and expert knowledge.

[0064] Step S2.2 mainly includes digital twin simulation enhancement, constructing a digital twin of the target line, and accurately simulating standard operating procedures (such as insulator replacement) in this virtual environment, automatically generating simulation data (operation time, tool path, virtual risk events), such as... Figure 5 As shown. For example, firstly, a twin is constructed based on the tower coordinates, conductor sag table, and laser point cloud data of the Binhu Line. The entire 3D scene of the line is reproduced 1:1 in a 3D modeling engine (such as Unity 3D), and its 10kV electrical properties are assigned. Secondly, a task simulation is performed, selecting a high-frequency operation task—simulating phase-to-phase connection on tower #15 (a straight tower). This includes: calling up models of standard virtual workers (wearing full protective suits), a 10kV insulated bucket truck, insulated gloves, conductor shielding covers, etc., from the model library; and driving the virtual model to automatically execute the following steps according to the "Standardized Operation Instructions": vehicle positioning → bucket truck raising and approaching → voltage testing → shielding installation → piercing and connection → shielding removal → evacuation. Thirdly, simulation data is automatically generated during the simulation process and recorded by the system. The generated data includes: total virtual time T_sim = 25 minutes (operation time); number of tool changes N_tool_change = 6 times; shielding installation steps S_shield = 8 steps (tool path and operation complexity); and the electric field intensity at the operator's fingertips and tool tips is calculated in real time using an integrated simplified boundary element method electric field calculation module. Simulations show that when the boom rotates to a certain angle, the virtual minimum distance D_min between the operator's shoulder and the edge phase jumper is 0.35 meters (less than the safe distance of 0.4 meters), triggering a "virtual alarm" (virtual risk event). Simultaneously, the system estimates the electric field intensity at this location as E_field = 3.8 kV / cm, close to 80% of the air breakdown threshold (virtual risk indicator).

[0065] Step S2.2 addresses situations where real-world operational data is scarce or high-risk scenarios prevent experimentation. Through twin simulation, massive amounts of diverse "synthetic data" can be generated at low cost and with zero risk, significantly enriching the model's training samples. Furthermore, twin simulation not only outputs results but also records the process, revealing the reasons for a particular route's low score (e.g., excessive occlusion steps or tool path conflicts). This provides a basis for generating specific and interpretable optimization suggestions, rather than simply providing a score. Step S2.2 upgrades the evaluation from inductive reasoning based on historical data to deductive verification combined with virtual experiments, solving two major bottlenecks: lack of process verification for evaluation results and insufficient training data in high-risk scenarios. This makes the optimization suggestions more credible and actionable.

[0066] Step S2.3 mainly includes incremental training and feedback optimization, achieved through three sub-steps: data fusion, model update, and model formation. For example, the actual feature vector of the lakeside line, the simulation data vector (25, 6, 8, 0.35, 3.8...) generated in step S2.2, and a piece of historical feedback data (last year's operation on a similar #15 pole type, actually taking 32 minutes, and the operators reported that the space was narrow) are packaged into a new training sample, thereby achieving data fusion. The "label" of this training sample is jointly determined by the initial model prediction score and the expert's comprehensive judgment on the simulation risk.

[0067] In step S2.3, the feedback data on the actual operation results mainly includes the actual operation time, unexpected difficulties encountered, and the subjective ratings of the operators on the effectiveness of the optimization suggestions provided by the model. Incremental training updates the model parameters using online learning or mini-batch retraining. To achieve model updates, mini-batch gradient descent is used to mix the new sample set with some historical data for one round of incremental training on the initial XGBoost model model_v1.0.pkl. The training focuses on adjusting the feature weights related to "operation space" and "operational complexity". The optimized model model_v1.1.pkl is obtained, which is the current live-line operation adaptive evaluation model. This live-line operation adaptive evaluation model is more sensitive to the potential risks caused by "narrow operation space".

[0068] Input the complete field data of the lakeside line (acquired via S1 and weighted using a combined weighting method) into model_v1.1. Model output: Overall score: 71 points (the initial model v1.0 predicted 75 points, but the score is more conservative because it incorporates risk information from simulation and feedback).

[0069] Preliminary analysis: The decision tree inside the model shows that the deductions mainly stem from the "accessibility of the work platform" and "spatial risks exposed in the simulation" of pole #15.

[0070] Step S2.3 uses the "synthetic data" generated by the simulation in step S2.2 and the "effect data" fed back from the actual operation as new training samples to incrementally learn the initial model, continuously adjusting its parameters and achieving continuous model evolution. Furthermore, it connects evaluation with operational practice, feeding the results back to the model so that it performs better in the next evaluation, thus forming a learning loop. Step S2.3 endows the model with the ability to update. Without this step, the model would be static and one-off. Therefore, step S2.3 ensures that the system becomes smarter with use, adapting to changes in power grid equipment updates and operational technology development, serving as a dynamic maintenance mechanism for adaptive characteristics.

[0071] Overall, step S4 replaces the static calculation module in the traditional evaluation method with a dynamic intelligent agent. Through a cycle of learning from history (step S2.1) – verifying and exploring in the virtual environment (step S2.2) – optimizing in practice (step S2.3), the evaluation capability is self-itererated and enhanced.

[0072] Step S3 mainly includes outputting a comprehensive evaluation score and optimization suggestions, and outputting them in the form of a visual report, GIS (Geographic Information System) map annotation, and structured list. This step transforms the model's analysis results into decision support information that can be directly used by personnel in different roles such as operation and maintenance, design, and safety supervision, and is the final step in realizing value. For example, the decision support module calls the report template and automatically generates a "10kV Binhu Line Live-Line Operation Adaptability Evaluation Report". The first page of the report prominently displays a score of 71 points (yellow warning level) and includes a radar chart of scores for four levels of indicators, showing that the score for the "Operation-Friendly Layer Dimension" is significantly low. The following optimization suggestions are given: 1) Risk Tower List: "#15 tower (fishpond edge): The insulated bucket truck cannot be used, and there is insufficient working space for the middle phase during tower climbing operations. Simulation shows a risk of insufficient safety distance. Priority modification suggestion: In conjunction with planned maintenance, replace the single angle iron crossarm of #15 tower with an insulated crossarm or add a working platform."; 2) Tool Configuration Scheme: "For the complex shielding problem of tension towers such as #40 tower, it is recommended to purchase 2 sets of 'tension tower dedicated integrated shielding covers'. Simulation verification shows that it can reduce the complexity index by 30%."; 3) Dynamic Operation Warning: "In the current season, there are often gusts of wind in the afternoon, and the humidity is high. Warning: If the humidity is >80% and the wind speed is >5m / s, it is not recommended to carry out complex shielding operations."; 4) Recommended Operation Method: "For the middle phase connection of #15 tower, after digital twin verification, it is recommended to use an insulated platform with insulated gloves for operation, and strictly limit the bucket arm rotation path (see simulation path diagram attachment)." Furthermore, on the company's intranet-based intelligent operation and maintenance management platform, the Binhu Line is displayed as a yellow line on the map, with pole #15 flashing red. Clicking on it allows users to view a 3D simulation video playback and a risk heat map.

[0073] The optimization suggestions output in step S3 are more specific and feasible because they are derived from the simulation verification in step S2.2, and more realistic because they are derived from the real feedback in step S2.3, thus truly realizing a closed loop from evaluation to decision support. System Implementation Examples

[0074] like Figure 4As shown, embodiments of the present invention also disclose an adaptive evaluation system for 10kV overhead lines for live-line work, used to implement the method described above. In embodiments of the present invention, the adaptive evaluation system for 10kV overhead lines for live-line work includes: a data acquisition and feature construction module, an adaptive evaluation model engine, and a decision support and output module. The data acquisition and feature construction module is used to acquire field data of the target line and construct a multi-dimensional evaluation feature vector based on a preset evaluation dimension system. The adaptive evaluation model engine is used to store and run the live-line work adaptive evaluation model. The decision support and output module is used to output and display the comprehensive evaluation score and optimization suggestions. The construction and optimization of the live-line work adaptive evaluation model (hereinafter referred to as the evaluation model) are achieved through sub-modules such as an initial training unit, a digital twin simulation and enhancement unit, and an incremental learning and optimization unit. The initial training unit is used to train an initial machine learning model based on historical data. The digital twin simulation and enhancement unit is used to construct a digital twin of the target line, simulate the work process, and generate simulation data vectors. The incremental learning and optimization unit is used to integrate simulation data and actual work feedback to incrementally train and optimize the model.

[0075] Preferably, the digital twin simulation and data augmentation module also includes a configurable task library and a rule engine. The task library is used to define standard operating procedures, and the rule engine is used to determine operational compliance and risks in real time during the simulation process.

[0076] Preferably, the 10kV overhead line evaluation system for uninterrupted power supply operations of the present invention further includes a federated learning client module, embedded in the model self-learning optimization module, used to complete the federated learning process in collaboration with the central server after completing model training locally. Figure 3 As shown below, preferably, multiple local evaluation models deployed in different power supply areas (e.g., three local models A, B, and C at branch offices) are incrementally trained on local data, and only encrypted model parameter updates are uploaded to the central server for secure aggregation, forming a globally optimized next-generation evaluation model, which is then synchronized to each local system. Furthermore, this invention introduces model self-learning and federated learning for performance enhancement. Weekly, the system automatically collects feedback data from all completed tasks of the past week (from mobile form submissions), and the model self-learning module organizes a nightly incremental training. Monthly, system instances deployed in different city branches (acting as federated learning clients) encrypt and upload local model updates to the provincial company's central server. After secure aggregation (e.g., using the FedAvg algorithm), the server generates a global model version V2.0, which is then distributed to each client, achieving secure knowledge sharing and co-evolution.

[0077] Preferably, the decision support and visualization output module also includes an augmented reality interface for converting key optimization suggestions, risk warnings, or simulation-verified operational steps into guidance information that can be displayed on augmented reality glasses or mobile terminals.

[0078] Preferably, this system is deployed on a cloud platform. Each power supply station can access it via a web client. When an evaluation of the "Binhu Line" is required, the system automatically executes the above process. Furthermore, this system is also applicable to other lines. For example, when a user submits an evaluation task for the "10kV West Trunk Line," the data acquisition module will automatically retrieve the corresponding data for the "West Trunk Line," and processes such as weight calculation, digital twin simulation, and model inference will be automatically adapted and executed, ultimately outputting a dedicated evaluation report for the "West Trunk Line." This demonstrates the system's versatility and automation capabilities.

[0079] Figure 4 This paper demonstrates the overall architecture and data collaboration mechanism of the adaptive evaluation system for 10kV overhead lines for uninterrupted power supply operations. The architecture adopts a three-layer modular design, clearly demonstrating the complete technical chain of multi-source data fusion, intelligent model-driven operation, and multi-dimensional decision support. The left side is the data input layer, which aggregates three major data sources: static data sources (production management system, design drawings, historical records), dynamic data sources (meteorological sensors, real-time environmental monitoring status), and spatial data sources (UAV imagery, LiDAR point clouds, and image recognition results). These heterogeneous data are uniformly accessed, processed, and constructed into standardized multi-dimensional feature vectors through the data acquisition and feature construction module. The middle section is the core processing layer, with the adaptive evaluation model engine as the intelligent hub. This engine integrates an initial training unit, a digital twin simulation unit, an incremental learning optimization unit, and a federated learning client. It relies on line data and expert labels in the historical database for basic training, uses the digital twin platform for operation simulation and enhancement, and integrates actual operation feedback data (time consumption, difficulty, and effectiveness scores) for continuous optimization. The architecture also innovatively introduces a federated learning framework. Local model parameters are encrypted and then uploaded to the central server for secure aggregation, enabling cross-regional knowledge sharing and global model evolution while effectively protecting local data privacy. The right side represents the output and application layer, which transforms model analysis results into multi-faceted, actionable outcomes through decision support and output modules. These include generating quantitative evaluation reports (safety levels, risk lists, and remediation plans), creating spatially visualized GIS map annotations and risk heat maps, and outputting AR work instructions (real-time warnings and step-by-step guidance) for wearable devices, achieving a closed-loop process from intelligent analysis to on-site empowerment. Through interconnected external support systems (expert knowledge systems and federated learning servers) and internally and externally circulating data feedback streams, this architecture constructs an adaptive evaluation ecosystem that can deeply learn from historical experience, adapt to real-time changes on-site, and evolve collaboratively.

[0080] also, Figure 5 This invention showcases the functional layout and multimodal interactive design of the digital twin simulation interface in the adaptive evaluation system for 10kV overhead lines for uninterrupted power supply operations, demonstrating the deep integration of virtual simulation and physical reality. The digital twin simulation interface adopts a four-part structure: the main view area presents the tower structure, electrical clearances, and dynamic virtual operation process of the 10kV overhead line in a high-fidelity 3D scene, creating an immersive simulation environment; the control panel provides a human-computer interaction interface for selecting work tasks, configuring environmental parameters, and controlling the simulation process; the data display area monitors and visualizes key safety and efficiency indicators in real time, including potential intensity, minimum safe distance, virtual operation time, and dynamic risk level; the analysis function area focuses on in-depth data mining, recording risk events, quantifying operational complexity, and verifying the effectiveness of optimization schemes. The simulation core engine at the bottom of the interface reveals the computational foundation supporting the above functions, integrating 3D modeling, electric field calculation, physical collision detection, and rule judgment systems to ensure the accuracy and compliance of the simulation. The output interface on the right clearly defines the three value flows of simulation data: the generated simulation data vectors are fed back as augmented samples to the machine learning model for training and optimization; the verified work suggestions directly support on-site decision-making; and the visualized standard operating procedures can be converted into augmented reality work guidance through AR / VR devices, realizing a closed-loop chain from digital simulation to on-site empowerment.

[0081] It should be noted that in this paper, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply these relationships. There is no such actual relationship or order between entities or operations. Furthermore, the terms "including" and "package" do not apply. The word "comprise" or any other variation thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprises a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0082] In this embodiment of the invention, the term "and / or" describes the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The character " / " generally indicates that the preceding and following associated objects have an "or" relationship.

[0083] The various embodiments in this specification are described in a related manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0084] In particular, the device embodiment is basically similar to the method embodiment, so the description is relatively simple. For relevant details, please refer to the description of the method embodiment.

[0085] For ease of description, the above apparatus is described by dividing it into various functional units / modules. Of course, in implementing this invention, the functions of each unit / module can be implemented in one or more software and / or hardware.

[0086] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0087] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An adaptive evaluation method for 10kV overhead lines for live-line work, characterized in that, Includes the following steps: S1: Acquire the field data of the target line, and construct the field data into a multi-dimensional evaluation feature vector based on the preset evaluation dimension system for live-line work; wherein, the evaluation dimension system includes at least the tower structure layer dimension, the electrical clearance layer dimension, the work-friendly layer dimension, and the environmental adaptation layer dimension. S2: Input the multi-dimensional evaluation feature vector into the live-line operation adaptability evaluation model to obtain the comprehensive evaluation score and optimization suggestions for the target line; S3: Output the comprehensive evaluation score and optimization suggestions; The live-line work adaptability evaluation model is constructed in the following way: S21: Based on the multidimensional evaluation feature vectors of the line samples and the corresponding expert evaluation labels, the initial machine learning model is trained; S22: For the target line, construct its digital twin, and simulate the preset uninterrupted power supply operation task process in the digital twin to generate a simulation data vector containing operation efficiency, operation complexity and virtual risk indicators. S23: Using the simulation data vector and the operation effect data fed back after the actual operation on the target line, the initial machine learning model is incrementally trained and the parameters are optimized to form the live-line operation adaptability evaluation model.

2. The adaptive evaluation method for 10kV overhead lines according to claim 1, characterized in that, The characteristic parameters corresponding to the pole structure layer dimension include crossarm type, number of hanging points, and pole health; the characteristic parameters corresponding to the electrical clearance layer dimension include phase-to-phase distance, distance to ground, and safe working space margin index; the characteristic parameters corresponding to the work-friendliness layer dimension include hardware disassembly and assembly convenience index, work platform accessibility index, and standard shielding operation complexity index; the characteristic parameters corresponding to the environmental adaptability layer dimension include wind speed and wind direction influence coefficient, humidity and condensation influence coefficient, and regional pollution level coefficient.

3. The adaptive evaluation method for 10kV overhead lines according to claim 1, characterized in that, The acquisition of on-site data for the target route includes: obtaining static ledger parameters of the target route from the production management system, collecting dynamic environmental parameters of the target route in real time through sensors installed on poles and towers, and obtaining corridor status parameters of the target route through drone inspections and image recognition.

4. The adaptive evaluation method for 10kV overhead lines according to claim 1, characterized in that, In step S1, when constructing the field data into a multidimensional evaluation feature vector, a combined weighting method is used to determine the weights of each dimension and the underlying feature parameters. The combined weighting method integrates the subjective weights obtained based on the analytic hierarchy process and the objective weights obtained based on the entropy weighting method.

5. The adaptive evaluation method for 10kV overhead lines according to claim 1, characterized in that, In step S21, the initial machine learning model adopts one of gradient boosting decision tree, random forest or deep neural network model; the expert evaluation label is the line live-line operation safety level determined based on the multi-expert Delphi method.

6. The adaptive evaluation method for 10kV overhead lines according to claim 1, characterized in that, Step S22 specifically includes: Based on the geometric and electrical parameters of the target line, a digital twin of the target line is constructed in the digital twin platform; The digital twin incorporates parameterized virtual models of standard operating personnel, insulating tools, and shielding equipment. The virtual model is driven to execute a preset uninterrupted power supply operation task process, and the tool path, operation time, virtual distance parameters, and potential risk events determined by the built-in rule engine are recorded simultaneously. Based on the recorded data, the operation efficiency, operational complexity, and virtual risk indicators are calculated, and a simulation data vector is formed.

7. The adaptive evaluation method for 10kV overhead lines according to claim 6, characterized in that, Calculating virtual risk indicators includes: Based on the simplified electric field calculation module integrated in the digital twin platform, virtual risk indicators are calculated; The simplified electric field calculation module is used to simulate and estimate the changes in electric field intensity at preset key locations during the operation.

8. The adaptive evaluation method for 10kV overhead lines according to claim 1, characterized in that, The operational performance data fed back after the actual operation of the target line in step S23 includes the actual operation time, unexpected difficulties encountered, and the subjective rating of the effectiveness of the optimization suggestions provided by the model by the operators; the incremental training updates the model parameters by means of online learning or mini-batch retraining.

9. The adaptive evaluation method for 10kV overhead lines according to claim 1, characterized in that, The optimization suggestion refers to a suggestion generated by integrating the comprehensive evaluation score, digital twin simulation results, and model decision-making basis; The optimization recommendations include at least: a list of unfriendly towers or sections and their specific risk causes, simulation-verified hardware modification or tool configuration schemes, dynamic operational risk warnings based on current environmental parameters, and recommended preferred operational methods. Step S23 further includes a model co-evolution step: by deploying a federated learning framework, multiple local live-line work adaptive evaluation models deployed in different power supply areas are incrementally trained on local data, and only the encrypted live-line work adaptive evaluation model parameters are updated and uploaded to the central server for secure aggregation, forming a globally optimized new generation of live-line work adaptive evaluation model and synchronizing it to each local system.

10. An adaptive evaluation system for 10kV overhead lines for live-line work, used to implement the adaptive evaluation method for 10kV overhead lines as described in any one of claims 1-9, characterized in that, include: The data acquisition and feature construction module is used to acquire on-site data of the target route and construct multi-dimensional evaluation feature vectors based on a preset evaluation dimension system. An adaptive evaluation model engine is used to store and run the aforementioned uninterrupted power supply (UPS) adaptive evaluation model. The decision support and output module is used to output and display the comprehensive evaluation score and optimization suggestions. The construction and optimization of the evaluation model are achieved through the following sub-modules: Initial training unit, used to train an initial machine learning model based on historical data; The digital twin simulation and enhancement unit is used to construct a digital twin of the target line, simulate the operation process, and generate simulation data vectors. The incremental learning and optimization unit is used to integrate simulation data and actual operation feedback to perform incremental training and optimization of the model.

Images