A method for automatic fault detection and repair of self-healing intelligent power lines

Abstract

The present invention discloses a method for automatic fault detection and repair of a self-healing intelligent power line, comprising the following steps: using a K-Means clustering algorithm to perform cluster analysis on fault sensor data, determining fault-frequently occurring areas, and then, for line equipment parameters and electrical parameters in a local repair section, using a convolutional neural network model of heterogeneous data fusion to perform feature extraction and matching analysis on line layout images and parameter data, judging the compatibility of a local repair scheme with the original line planning, using a least squares support vector machine model to perform real-time data analysis during the local repair construction process, performing live debugging and comprehensive evaluation on the reconstructed local line to judge whether the reconstruction effect meets the construction standard of the self-healing intelligent power line, using a knowledge graph-driven long-short-term memory network model to predict and infer equipment health status and fault risk, automatically generating state maintenance and fault repair strategies, and realizing self-healing operation of the local line.

Description

Technical Field

[0001] The invention relates to the field of information technology, and in particular to a method for automatically detecting and repairing faults of a self-healing intelligent power line. Background Art

[0002] In areas with high incidence of faults, there is a technical contradiction between batch local area repair and line re-layout self-healing. On the one hand, batch repair can quickly restore faults in the area and reduce the impact time of faults on business; but on the other hand, large-scale local repair will cause changes in the line layout in the area, which will be incompatible with the original overall line planning and affect the stability and reliability of the network. At the same time, although line re-layout can achieve self-healing and improve the reliability and fault tolerance of the network, the re-layout process is complicated and needs to consider the connectivity and load balancing factors of the overall line. The implementation cycle is long and the business cannot be restored quickly. In addition, frequent line re-layout will also affect network performance. Increase network instability. In actual business scenarios, areas with high incidence of failures usually have large business volumes and dense users, and have high requirements for network quality and reliability. Therefore, when performing batch repairs and line re-layout in such areas, it is necessary to take into account both repair efficiency and network stability. One technical solution is to divide the local area into multiple small repair units, and use batch repairs in each repair unit to quickly restore business. At the same time, a self-healing mechanism is used between repair units to achieve local self-healing and improve network reliability by dynamically adjusting link weights and optimizing routing methods. However, this solution needs to solve the technical problems of repair unit division and self-healing algorithm to achieve a balance between repair efficiency and network stability.

[0003] The present invention proposes solutions to the above shortcomings: the fault-frequent area positioning adopts the K-Means clustering algorithm to perform cluster analysis on the fault sensor data; the fault-frequent area positioning, line parameter optimization and equipment transformation use the convolutional neural network model of heterogeneous data fusion to analyze the line layout image and parameter data, and judge the compatibility of the local repair plan with the original line planning; the line layout multi-objective optimization adopts the non-dominated sorting genetic algorithm to perform multi-objective optimization on the line layout to minimize the impact on the global line planning; the repair plan is dynamically adjusted during the local repair construction process, the operating condition data is collected through distributed sensors, and the transfer learning enhanced least squares support vector machine model is used for real-time analysis; the line reconstruction plan optimization optimizes the reconstruction plan for the key monitored line equipment and location; the line reconstruction effect evaluation carries out live debugging and comprehensive evaluation on the reconstructed local line according to the construction drawing, and the integrated learning optimized least squares support vector machine model is used to perform fusion analysis on the data; the local line self-healing operation establishes a supporting intelligent operation and maintenance system for the local line reconstruction area, adopts the key technology of wireless sensor network Internet of Things to collect line status data, and predicts and infers through the long short-term memory network model driven by knowledge graph. Summary of the invention

[0004] In order to solve the above-mentioned technical problems, the present invention provides a method for automatically detecting and repairing faults of a self-healing intelligent power line.

[0005] The technical solution of the present invention is implemented as follows: a method for automatic fault detection and repair of a self-healing intelligent power line, comprising:

[0006] According to the distributed sensor layout plan of the self-healing intelligent power line, the K-Means clustering algorithm is used to perform cluster analysis on the fault sensor data, obtain the spatial distribution characteristics of the fault-frequent areas, and determine the key line sections that need to be repaired locally;

[0007] For the line equipment parameters and electrical parameters in the local repair section, a convolutional neural network model with heterogeneous data fusion is used to extract features and perform matching analysis on the line layout image and parameter data to determine the compatibility of the local repair plan with the original line plan. If there is an incompatibility problem, line parameter optimization and equipment transformation are carried out.

[0008] According to the changes in line direction and length caused by local repairs, a non-dominated sorting genetic algorithm is used to perform multi-objective optimization of the line layout. By constructing a fitness function that considers terrain adaptability, electrical performance, and economy, the impact on global line planning is minimized while meeting local repair requirements, and an optimization plan for the line layout that takes both local and overall considerations into account is obtained.

[0009] During the local repair construction process, automatic fault detection is carried out simultaneously. Distributed sensors are used to collect line operation data. The data is analyzed in real time using a transfer learning-enhanced least squares support vector machine model to detect line anomalies and potential faults that occur during the repair process, dynamically adjust the repair plan, and obtain the line equipment and locations that need to be monitored.

[0010] Optimize the reconstruction plan for the equipment on the line for the key monitored line equipment and locations. Under the premise of meeting the physical and electrical constraints of the line, optimize the selection of conductor models, pole towers and insulator configurations, maximize the line's risk resistance, fault self-healing ability and intelligence level, and generate line reconstruction construction drawings;

[0011] According to the construction drawings, live debugging and comprehensive evaluation were carried out on the reconstructed local lines. The line voltage, current and temperature were collected, and the least squares support vector machine model optimized by ensemble learning was used to perform data fusion analysis. The evaluation accuracy was improved through cross-validation and grid search parameter optimization. The technical indicators before and after the line reconstruction were compared to determine whether the reconstruction effect met the construction standards of self-healing intelligent power lines.

[0012] Establish a supporting intelligent operation and maintenance system in the local line reconstruction area, collect line status data through key technologies of wireless sensor network and Internet of Things, use knowledge graph-driven long and short-term memory network models to predict and infer equipment health status and fault risks, automatically generate status maintenance and fault repair strategies, and coordinate the dispatch of drones and robots to achieve self-healing operation of local lines. Beneficial Effects

[0013] The present invention aims to solve the problem of rapid positioning, repair and reconstruction of power line faults, and improve the operation efficiency and risk resistance of the power grid. The unique business scenario problem is how to quickly and accurately locate the fault-prone areas after a power line fault occurs, and formulate effective local line repair and optimization strategies, while ensuring that the repair process does not affect the stable operation of the overall power grid;

[0014] The present invention provides a comprehensive technical solution to effectively solve the problems of smart power lines in fault detection, rapid repair, optimized reconstruction and intelligent operation and maintenance. The solution can not only quickly locate the fault area, but also achieve local repair and optimization without affecting the stability of the overall power grid. Through dynamic adjustment and intelligent operation and maintenance, the reliability and intelligence level of the power system can be further improved, and the continuity and safety of power supply can be guaranteed. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A structural block diagram of a method for automatic fault detection and repair of a self-healing intelligent power line in an embodiment of the present invention;

[0016] Figure 2 The present invention is a flowchart of a method for automatically detecting and repairing faults of a self-healing intelligent power line according to an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0018] The preferred implementation methods of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these implementation methods are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0019] It should be noted that when an element is considered to be "connected" to another element, it can be directly connected to the other element, or connected to the other element through an intermediate element. In addition, the "connection" in the following embodiments should be understood as "electrical connection", "communication connection", etc. if there is transmission of electrical signals or data between the connected objects.

[0020] When used herein, the singular forms "a", "an", and "said / the" may also include plural forms, unless the context clearly indicates otherwise. It should also be understood that the terms "include / comprise" or "have" and the like specify the presence of stated features, wholes, steps, operations, components, parts, or combinations thereof, but do not exclude the possibility of the presence or addition of one or more other features, wholes, steps, operations, components, parts, or combinations thereof. At the same time, the term "and / or" used in this specification includes any and all combinations of the relevant listed items.

[0021] See also Figure 1-2 As shown, a method for automatic fault detection and repair of a self-healing intelligent power line includes:

[0022] S101. According to the distributed sensor layout plan of the self-healing intelligent power line, the K-Means clustering algorithm is used to perform cluster analysis on the fault sensor data, obtain the spatial distribution characteristics of the fault-frequently occurring areas, and determine the key line sections that need to be repaired locally;

[0023] S102. For the line equipment parameters and electrical parameters in the local repair section, a convolutional neural network model with heterogeneous data fusion is used to perform feature extraction and matching analysis on the line layout image and parameter data, and the compatibility of the local repair plan with the original line plan is determined. If there is an incompatibility problem, line parameter optimization and equipment transformation are performed;

[0024] S103. Based on the changes in line direction and length caused by local repair, a non-dominated sorting genetic algorithm is used to perform multi-objective optimization on the line layout. By constructing a fitness function that considers terrain adaptability, electrical performance, and economy, the impact on global line planning is minimized while meeting local repair requirements, and an optimization plan for the line layout that takes both local and overall considerations into account is obtained;

[0025] S104. During the local repair construction process, automatic fault detection is carried out simultaneously. The line operation condition data is collected through distributed sensors. The data is analyzed in real time using the least squares support vector machine model enhanced by transfer learning. Line anomalies and potential faults that occur during the repair process are discovered, and the repair plan is adjusted dynamically to obtain the line equipment and locations that need to be monitored;

[0026] S105. Optimize the reconstruction plan for the equipment on the line for the key monitored line equipment and locations. Under the premise of meeting the physical and electrical constraints of the line, optimize the selection of conductor models, pole towers and insulator configurations, maximize the risk resistance, fault self-healing ability and intelligent level of the line, and generate the line reconstruction construction drawings;

[0027] S106. According to the construction drawings, conduct live commissioning and comprehensive evaluation of the reconstructed local lines, collect line voltage, current and temperature, use the least squares support vector machine model optimized by ensemble learning to perform data fusion analysis, improve the evaluation accuracy through cross-validation and grid search parameter optimization, obtain the comparison of various technical indicators before and after the line reconstruction, and judge whether the reconstruction effect meets the construction standards of self-healing intelligent power lines;

[0028] S107. Establish a supporting intelligent operation and maintenance system for the local line reconstruction area, collect line status data through key technologies of wireless sensor network Internet of Things, use knowledge graph-driven long-short term memory network models to predict and infer equipment health status and fault risks, automatically generate status maintenance and fault repair strategies, and coordinate the dispatch of drones and robots to achieve self-healing operation of local lines.

[0029] In S101, according to the distributed sensor layout plan of the self-healing intelligent power line, a K-Means clustering algorithm is used to perform cluster analysis on the fault sensor data to obtain the spatial distribution characteristics of the fault-frequently occurring areas and determine the key line sections that need to be locally repaired.

[0030] Specifically, according to the distributed sensor layout scheme of the self-healing intelligent power line, the K-Means clustering algorithm is used to pre-process the fault sensor data, and a high-quality data foundation is obtained by removing outliers and noise data; the pre-processed data is used to extract and select features, obtain key attributes that can reflect the fault characteristics, and construct a fault feature vector; according to the characteristics of the fault data and business needs, the number of clusters k is preset, and the K-Means clustering algorithm is used to perform cluster analysis on the fault feature vector, and the optimal clustering result is obtained by iteratively optimizing the cluster center and sample allocation; based on the clustering results, the spatial distribution characteristics of the fault-frequent area are extracted, and the spatial distribution heat map of the fault-frequent area is generated by using spatial interpolation and visualization technology; on the basis of the identification of the fault-frequent area, combined with the topological structure, operating parameters, and load distribution factors of the power line, the line section that needs to be focused on local repair is determined through multi-attribute decision-making and optimization algorithms; according to the key line section, combined with historical fault data, a machine learning algorithm is used to automatically generate a local repair plan, optimize the priority of the repair task, the scheduling of repair resources, and the process parameters of the repair process, and improve the intelligence level of line repair.

[0031] In one embodiment, according to the distributed sensor layout scheme of the self-healing intelligent power line, the K-Means clustering algorithm is used to pre-process the fault sensor data. By removing outliers and noise data, the improvement of data quality and reliability is evaluated based on the statistical characteristics and distribution laws of the data, providing a high-quality data basis for subsequent analysis. The pre-processed data is used for feature extraction and selection to obtain key attributes that can reflect the fault characteristics, including fault type, fault location, and fault duration. A fault feature vector is constructed to provide effective input for cluster analysis. According to the characteristics of the fault data and business needs, the number of clusters k is preset, and the K-Means clustering algorithm is used to perform cluster analysis on the fault feature vector. By iteratively optimizing the cluster center and sample allocation, the optimal clustering result is obtained, and the fault-frequent area is accurately identified. Based on the clustering results, the spatial distribution characteristics of fault-prone areas are extracted, including the geographical coordinates of the fault points and the density of the fault points. The spatial distribution heat map of fault-prone areas is generated by using spatial interpolation and visualization technology to intuitively present the high-incidence areas of faults. On the basis of identifying fault-prone areas, the topological structure, operating parameters, and load distribution factors of the power lines are comprehensively considered. Through multi-attribute decision-making and optimization algorithms, including hierarchical analysis method and fuzzy comprehensive evaluation, the line sections that need to be focused on for local repair are determined to improve the pertinence and effectiveness of fault handling. According to the determined key line sections, combined with historical fault data, machine learning algorithms, including decision trees and random forests, are used to automatically generate local repair plans, optimize the priority of repair tasks, the scheduling of repair resources, and the process parameters of the repair process, and improve the intelligent level of line repair.

[0032] When preprocessing the fault sensor data, the 3σ principle is used to remove outliers from the data, that is, data beyond the mean ±3 times the standard deviation is regarded as outliers and removed. At the same time, wavelet transform is used to reduce the noise of the data, and the wavelet basis function (db4) and the number of decomposition layers (5 layers) are selected to perform multi-scale decomposition on the data to remove high-frequency noise components. The peak signal-to-noise ratio (PSNR) and root mean square error (RMSE) of the data are calculated to evaluate the improvement effect of data quality. The PSNR is improved by 5dB and the RMSE is reduced by 10%, indicating that data preprocessing effectively improves data quality. When selecting features, the recursive feature elimination (RFE) algorithm is used. By recursively training the model and evaluating the importance of the features, the Top-10 features that contribute the most to fault identification are gradually selected to construct the optimal feature subset, including selecting the key features of fault current, fault voltage, and fault duration through RFE to form a fault feature vector, which provides high-quality input for clustering analysis. When determining the number of clusters k, the elbow method (ElbowMethod) is used to evaluate the clustering results under different k values. , select the k value at the inflection point where the clustering effect is more significantly improved, including the elbow method to determine that when k=5, the clustering effect is optimal. When generating the distribution map of the fault-frequent area, Kriging Interpolation is used to spatially interpolate the fault points. According to the spatial autocorrelation of the fault points, the fault probability of the unknown position is estimated. At the same time, the kernel density estimation (KDE) method is used to smooth the spatial distribution of the fault points to generate a continuous heat map. In the determination of key line sections, the fuzzy analytic hierarchy process (FAHP) is used to construct an evaluation system including line importance and fault risk indicators, and the indicator weights are assigned. The priority scores of each line section are comprehensively calculated, and the sections with the top-3 scores are selected as the key repair objects. When generating local repair plans, the decision tree algorithm is used to train the optimization model based on historical repair data, and automatically generate repair decision rules, including when the fault type is short circuit and the fault current is greater than 1kA, the live working method is preferred for emergency repair, the amount of scheduled repair resources is 1.5 times the fault span, and the repair time is controlled within 2 hours.

[0033] In S102, for the line equipment parameters and electrical parameters in the local repair section, a convolutional neural network model with heterogeneous data fusion is used to perform feature extraction and matching analysis on the line layout image and parameter data, to determine the compatibility of the local repair plan with the original line plan, and if there is an incompatibility problem, line parameter optimization and equipment modification are performed.

[0034] Specifically, the line equipment parameters and electrical parameters in the local repair section are obtained, and a line parameter database is constructed; wherein the line equipment parameters and electrical parameters include line length, conductor model, tower type, insulator parameters and transformer capacity, and the line layout image is preprocessed by image processing technology to extract key information of line direction and equipment location in the image, and obtain a digitized feature vector, and the line parameter database is matched and associated with the feature vector, and a fused line data representation is constructed through database query and image analysis technology; wherein the line data representation includes equipment parameters, electrical parameters and spatial layout information, and the line data representation is input into a pre-constructed convolutional neural network model, and the key features of the line are extracted through the convolution layer and the pooling layer. , and realize nonlinear combination and mapping of features through the fully connected layer to obtain the compatibility evaluation results of the line layout and parameters. If the compatibility evaluation result is lower than the preset threshold, it is judged that there is an incompatibility problem between the local repair plan and the original line plan. The Morris sensitivity analysis method is used to determine the key parameters whose impact on compatibility is greater than the preset threshold. According to the line specifications and safety constraints, candidate solutions for line parameter optimization and equipment transformation are generated; wherein the candidate solutions include parameter adjustment range and equipment replacement list, a multi-objective optimization model is constructed, and the NSGA-II algorithm is used for multi-objective optimization solution to obtain the Pareto optimal solution set, and the optimal line parameter combination and equipment transformation plan are selected from the Pareto optimal solution set to generate an implementation plan and process documents.

[0035] In one embodiment, the line equipment parameters and electrical parameters in the local repair section are obtained, including line length, conductor model, tower type, insulator parameters, and transformer capacity. A line parameter database is constructed, and a relational database (MySQL) is used for storage. A reasonable table structure and index are set to ensure data integrity and consistency. The line layout image is preprocessed through image processing, including image denoising, enhancement, and correction to improve image quality. At the same time, key information such as line direction and equipment location in the image is extracted and converted into a digital feature vector. Heterogeneous data fusion is used to match and associate the line parameter database with the image feature vector. Based on spatial coordinates and equipment The type key field is used to achieve accurate correspondence between parameter data and image features through spatial indexing and similarity calculation, and the fused line data representation is constructed, including equipment parameters, electrical parameters and spatial layout information, providing rich feature input for the training of the convolutional neural network model. The convolutional neural network model is constructed and trained. The pooling method is maximum pooling. The fused line data representation is used as input. The key features of the line are extracted through the convolution layer and the pooling layer, and the nonlinear combination and mapping of the features are realized through the fully connected layer, so as to judge the compatibility of the line layout and parameters. Based on the historical line planning data, a high-quality training data set is constructed, and the model training is carried out by cross-validation. The line layout and parameter data of the local repair plan are re-preprocessed and feature extracted to form an input format consistent with the training data. The extracted features are input into the trained convolutional neural network model to obtain the compatibility evaluation results of the local repair plan and the original line plan. The compatibility threshold is set. If the compatibility score is lower than the threshold, it is judged that there is an incompatibility problem and further optimization is required. For the identified incompatibility problems, the Morris sensitivity analysis method is used to multiple perturb the line parameters and evaluate the compatibility changes to determine the key parameters that have a greater impact on compatibility. The influence coefficient of each parameter is quantified through the logistic regression model, combined with line specifications and safety Constraints, generate candidate solutions for line parameter optimization and equipment transformation, form parameter adjustment range and equipment replacement list, comprehensively consider the technical feasibility, economy and impact on line reliability of candidate solutions, build a multi-objective optimization model, the objective function includes compatibility score, investment cost and line reliability index, constraints include equipment selection specifications, line parameter value range, use NSGA-II algorithm for multi-objective optimization solution, obtain Pareto optimal solution set, select the optimal line parameter combination and equipment transformation plan according to actual needs and preferences, generate detailed implementation plan and process documents, and feed them back to the local repair plan to guide the refined repair and transformation of the line;

[0036] When constructing the line parameter database, the MySQL relational database is used, which includes 10 related tables: line basic information table, tower information table, conductor information table, and insulator information table. B+ tree index is established for the key fields of line ID, tower number, and conductor model to improve data query efficiency. For the line layout image, the Canny operator is used for edge detection to extract the line direction contour. At the same time, the Hough transform is used to detect the tower position, and the conductor direction is extracted through the Hough linear transform to form a 32-dimensional image feature vector. When merging heterogeneous data, the line ID is used as the primary key, and the tower number is used as the primary key. The tower number is the foreign key, and the spatial indexing technology is used to spatially match the image feature vector and parameter data. The feature similarity is calculated by Euclidean distance, and the distance threshold is set to 0.1 to complete the data association. The convolutional neural network model uses 5 convolutional layers and 3 fully connected layers. The convolution kernel sizes are 3x3 and 5x5 respectively, and the pooling kernel size is 2x2. The ReLU activation function and Dropout regularization are used. The training data set contains 10,000 historical route planning records, and the positive and negative ratio of samples is 1:1. The Adam optimizer is used, the learning rate is 0.001, and the batch size is 128. , trained for 200 epochs, the model accuracy reached more than 95%. When performing Morris sensitivity analysis, each line parameter was perturbed 10 times, and the perturbation step was 1 / 10 of the parameter range. Based on the change range of the compatibility score output by the network model, the basic influence value and interactive influence value of the parameter were calculated, and 8 key parameters were screened out. The influence coefficient of each parameter was fitted by the logistic regression model. The parameters with an absolute value of the coefficient greater than 0.3 were considered to be the dominant factors of compatibility. In multi-objective optimization, the compatibility score target was set to be greater than 0.95, and the investment cost target was set to not exceed The line reliability target is set to reduce the failure rate by 10% and the NSGA-II algorithm is used for 500 iterative searches with a crossover probability of 0.8, a mutation probability of 0.1 and a population size of 100. The non-dominated solutions on the Pareto frontier are selected as candidate optimization schemes. After technical and economic evaluation, the scheme with the highest compatibility score and the lowest investment cost is selected as the final line optimization and transformation scheme. A detailed implementation plan is formed, including an equipment replacement list, construction process documents and optimization effect estimates, which is fed back to the local repair strategy to guide the specific repair construction.

[0037] S103, based on the line direction and length changes caused by local repairs, a non-dominated sorting genetic algorithm is used to perform multi-objective optimization on the line layout, and by constructing a fitness function that considers terrain adaptability, electrical performance, and economy, the impact on global line planning is minimized while meeting the local repair requirements, thereby obtaining a line layout optimization solution that takes both local and overall considerations into account.

[0038] Specifically, the line coordinate point sequence and line length difference before and after local repair are obtained, and a digital model of the line layout before and after repair is constructed; according to the digital model, the resistance, reactance and susceptance electrical parameters of the repaired line are calculated, and the terrain correction coefficient is introduced to correct the electrical parameters; the entropy weight method is used to determine the weight of the electrical performance evaluation index, and the weighted sum of each index value is normalized to obtain a comprehensive evaluation value of the electrical performance; the elevation, slope and geological key terrain parameters of the repair area are extracted to construct a terrain adaptability evaluation index system including elevation difference, slope and geological conditions; the hierarchical analysis method is used to determine the weight of the terrain adaptability evaluation index, and the score value of each index is obtained through expert scoring, and the weighted sum is obtained to obtain the comprehensive evaluation value of terrain adaptability; the material cost and construction cost of the repair plan are calculated, and the economic cost is processed by the linear normalization method to obtain the economic index; the terrain adaptability score, electrical performance evaluation value and economic index are used as optimization variables, and a multi-objective optimization model is constructed in combination with the line direction restriction constraint conditions; the multi-objective optimization model is solved by a non-dominated sorting genetic algorithm to obtain the line layout optimization plan with the best comprehensive performance.

[0039] In one embodiment, the line direction and length change data caused by local repair are obtained, including the line coordinate point sequence and line length difference before and after the repair, and a digital model of the line layout before and after the repair is constructed to provide basic data support for subsequent optimization. The elevation, slope, and key geological parameters are extracted according to the terrain characteristics of the repair area, and the terrain complexity index is introduced to construct a terrain adaptability evaluation index system. The fuzzy comprehensive evaluation method is used to quantitatively score the terrain adaptability of the repair scheme. By establishing a membership function, the membership of each evaluation index is calculated, and then a weighted sum is performed to obtain a comprehensive evaluation value of terrain adaptability as one of the optimization goals. The line layout is designed to calculate the electrical parameters of the line, including line resistance, reactance, susceptance, and voltage loss. The mapping relationship between terrain factors and electrical parameters is established. The terrain correction coefficient is introduced to correct the electrical parameters and construct an electrical performance evaluation model. The weighted summation method is used to normalize the various indicators, and the weights are set reasonably to form a comprehensive evaluation value of the electrical performance. As another optimization goal, from the aspects of material cost and construction cost, an economic evaluation model is constructed according to the characteristics of local repair. The material consumption and construction process in the repair area are considered, and the material cost and construction cost are calculated to obtain the direct economic cost of the repair plan. The economic cost is indexed to form an economic evaluation model. The economic optimization goal is achieved. The terrain adaptability score, electrical performance evaluation value, and economic index are used as optimization variables. Combined with the constraints of local repair, including line direction restrictions and tower position fine-tuning range, a multi-objective optimization model is constructed. The weighted summation method is used to form a comprehensive performance evaluation function as the fitness function. The weight coefficient reflects the relative importance of each goal. The non-dominated sorting genetic algorithm is used to perform multi-objective optimization on the line layout. Based on the fitness function, the Pareto optimal solution set is iteratively searched through selection, crossover, and mutation operations. The population size, crossover probability, mutation probability, and number of iterations are set to achieve the convergence rate of the fitness function value. The algorithm uses the degree as the termination condition, takes into account the integrity and consistency of the global line layout while meeting the local repair requirements, obtains the line layout optimization plan with the best comprehensive performance, conducts electrical simulation and economic evaluation on the line layout optimization plan, verifies its feasibility and superiority, generates detailed line plan layout, tower coordinate table, conductor parameter table engineering design files, meets the requirements of engineering drawing standards and specifications, and simultaneously updates the global line planning. The line layout and electrical parameter information of the local repair plan are compared and checked with the global plan to ensure seamless connection and coordination, and form a line layout optimization plan that takes into account local characteristics and overall coordination, and instructs the implementation of the project;

[0040] When obtaining the line direction and length change data before and after local repair, the GIS spatial database is used to extract the line coordinate points within the repair area, and the difference in line length before and after the repair is calculated. If the difference exceeds 5%, it is marked as a major change. For the terrain of the repair area, the digital elevation model (DEM) is used to extract elevation information and calculate the average slope. The area with a slope greater than 15 degrees is marked as complex terrain. In the fuzzy comprehensive evaluation, three factors, elevation, slope, and geology, are selected as evaluation indicators, and a fuzzy judgment matrix is ​​constructed. The trigonometric function method is used to determine the membership function, and the indicator weight is calculated by the hierarchical analysis method. Finally, the terrain adaptability score is obtained, and the score range is [0,100]. When calculating the line electrical parameters, the line conductor parameters and tower coordinate data are used to calculate the flow through the Newton-Raphson method to obtain the voltage and power distribution. If the voltage qualification rate is less than 95%, it is judged as poor electrical performance. When constructing the economic evaluation model, the material consumption and construction hours are extracted from the material list and construction plan, and multiplied by the unit price. The direct cost is obtained by grid, and then the total cost is calculated by considering 10% management fee and 5% tax. When constructing the fitness function, the terrain adaptability weight is set to 0.3, the electrical performance weight is set to 0.5, and the economic weight is set to 0.2. The comprehensive performance evaluation value is obtained by weighted summation. In the genetic algorithm optimization, the tournament selection operator, uniform crossover operator and Gaussian mutation operator are used, and the elite retention strategy is set to select the 10% individuals with the highest fitness value to directly enter the next generation. When generating the engineering design file, AutoCAD is used to draw the line plan layout, and the pole tower serial number, coordinate position and rotation angle are marked. The conductor model, length, tension and sag are also marked in the figure to meet the power engineering drawing standards. In the docking of the line layout optimization plan and the global planning, the line direction deviation and electrical parameter difference of the two are minimized through iterative optimization. When the deviation is less than 1%, seamless docking is achieved. The final output line layout optimization plan includes line layout, electrical parameters, and cost estimation content, providing detailed technical reference for engineering implementation.

[0041] The S104, during the local repair construction process, synchronously carry out automatic fault detection, collect line operation condition data through distributed sensors, use the transfer learning enhanced least squares support vector machine model to perform real-time analysis on the data, discover line anomalies and potential faults that occur during the repair process, dynamically adjust the repair plan, and obtain the line equipment and locations that need to be monitored.

[0042] Specifically, according to the network topology and data transmission requirements, a key node centrality analysis method based on graph theory is used to optimize the deployment location and number of sensors, wherein the centrality analysis method includes degree centrality and betweenness centrality; the collected line operation condition data is preprocessed, including cleaning, normalization and feature selection, and the statistical characteristics and frequency domain characteristics of the current, voltage and temperature time series data are extracted to construct feature vectors; an instance-based transfer learning method is used to enhance the generalization ability of the fault diagnosis model by weightedly fusing samples similar to the local repair construction data in the training set in the historical data; the local repair construction data collected in real time is used to fine-tune and incrementally train the fault diagnosis model, and the gradient penalty strategy is used to align the gradients of the source domain and the target domain to improve the model's adaptability and recognition of unknown fault types. The preprocessed feature vector is input into the fault diagnosis model for real-time online reasoning, the line operation status is classified and anomaly detection is performed, and the fuzzy reasoning method is used to distinguish the fault type in combination with the constructed fuzzy diagnosis rule base; if the fault detection result shows that the line status is abnormal and the abnormal index exceeds the preset threshold, the dynamic adjustment of the repair plan is triggered, and the maximization of repair efficiency and the minimization of resource consumption are taken as the optimization goals, a multi-objective optimization model is constructed, and the non-dominated sorting genetic algorithm is used for solving the problem to generate the Pareto optimal solution set; according to the fault detection results and the adjustment of the repair plan, the hierarchical analysis method is used to construct a risk assessment index system, and the risk level of each line equipment and location is calculated through fuzzy comprehensive evaluation. For high-risk areas, the sampling frequency and data upload cycle of the sensor network are dynamically configured to improve the spatiotemporal granularity of monitoring.

[0043] In one embodiment, a distributed sensor network is deployed, and current, voltage, and temperature sensors of various types are installed on key line nodes and equipment in the local repair construction area to collect line operation data in real time. The deployment location and quantity of sensors are optimized according to the network topology and data transmission requirements. A key node degree analysis method based on graph theory is used to select nodes with high degree centrality as sensor deployment points. Data is transmitted to the edge computing gateway through a wireless communication module, and an adaptive routing protocol is used to dynamically optimize the data transmission path to improve the robustness and reliability of the network. Local data processing and storage are performed at the gateway node to construct a transfer learning enhanced least squares support vector machine. The model is pre-trained using historical line operation data and known fault cases, and the model hyperparameters are optimized using grid search and cross-validation methods, including kernel function type and regularization coefficient, to balance the generalization ability of the model and the risk of overfitting. The model is then fine-tuned and incrementally trained using real-time data collected during local repair construction, and the model parameters are updated using the gradient descent algorithm to improve the model's adaptability and recognition accuracy to unknown fault types. The line operation condition data collected in real time is pre-processed, and the outlier detection algorithm (boxplot method) and missing value processing algorithm (kNN interpolation) are used for data cleaning to improve data quality. The Pearson correlation coefficient method and chi-square The inspection method selects features with high correlation with fault diagnosis, constructs multi-dimensional feature vectors, uses the moving average method to smooth the time series data, eliminates noise and fluctuations, and uses Fourier transform to extract frequency domain features, including harmonic components and spectral entropy, to enrich feature representation. The preprocessed feature vector is input into the least squares support vector machine model enhanced by transfer learning. Through real-time online reasoning, the line operation status is classified and anomalies are detected. Based on the statistical process control (SPC) theory, the upper and lower control limits of the feature quantity are calculated, and the control chart is constructed to realize fault warning and alarm. Combined with fuzzy diagnosis rules, the fault type is inferred and judged, and a diagnosis report and disposal suggestions are generated. According to the severity of the fault and the scope of impact, the fuzzy comprehensive evaluation method is used to determine the priority of the fault alarm, dynamically adjust the alarm upgrade strategy, conduct a comprehensive analysis of the detected line anomalies and fault hazards, and evaluate the impact on the progress and quality of the repair construction. If the impact exceeds the preset threshold, the dynamic adjustment mechanism of the repair plan is triggered. Through the multi-objective optimization model, with the construction progress, cost, and quality as constraints, and the minimization of fault risk as the objective function, the optimal repair plan adjustment strategy is solved. Using the online learning algorithm, according to the fault detection results and the plan execution feedback, the optimization model parameters are dynamically updated to improve the real-time and accuracy of the decision-making. According to the fault detection results and the repair plan adjustment situation,Dynamically configure the sampling frequency and data upload cycle of the sensor network, and use risk assessment algorithms to calculate the risk level of each line equipment and location based on the fault detection results and repair plan adjustments, and obtain a list of line equipment and locations that need to be monitored. For the line equipment and locations that need to be monitored, improve the time and space granularity of data collection to achieve accurate monitoring and early warning, integrate the automatic fault detection function in the local repair construction management system, develop a visual human-computer interaction interface, and realize real-time display and query of line operation status and fault alarm information;

[0044] Install current, voltage, and temperature sensors of various types on key line nodes and equipment in the local repair construction area, including current sensors on the low-voltage side of the transformer, with a range of 0-1000A and a precision of Class 5; voltage sensors on the busbar, with a range of 0-10kV and a precision of Class 2; and temperature sensors on the transformer oil tank, with a range of -20℃~150℃ and a precision of ±1℃. According to the network topology and data transmission requirements, the Kruskal algorithm is used to solve the minimum spanning tree, and nodes with a degree centrality greater than 5 are selected as sensor deployment points. On average, 2-3 sensors are deployed on each key node. The data is transmitted to the edge computing gateway through the Zigbee wireless communication module. The Ad-hoc On-demand Distance Vector routing protocol is used, with an average transmission delay of less than 100ms and a packet loss rate of less than 1%. A transfer learning-enhanced least squares support vector machine model is constructed, and 10,000 historical line operation data and 100 known fault cases are used for pre-training. The 5-fold cross-validation method is used to optimize the model hyperparameters, and the RBF kernel function is finally selected with a regularization coefficient of 0.1. The accuracy of the training set reached more than 95%. The outlier detection algorithm (box plot method) was used to clean the real-time data, and the outliers that were more than 5 times the distance from the upper and lower quartiles were removed; the Pearson correlation coefficient method was used to select 10 features with a correlation greater than 6 with fault diagnosis to construct a feature vector; the moving average method was used to smooth the time series data with a window size of 10; the fast Fourier transform was used to extract frequency domain features, and the first 5 harmonic components and spectral entropy were selected as features. Based on the statistical process control theory, the upper and lower control limits (±3σ) of the feature quantity were calculated, and the I-MR control chart was constructed to realize fault warning and alarm. Combined with fuzzy diagnosis rules (IF current>2IeAND oil temperature>90℃THEN fault type=transformer overload), the maximum membership method was used to infer the fault type, generate diagnostic reports and disposal suggestions, and the fuzzy comprehensive evaluation method was used to determine the priority of the fault alarm. The weight vector is [3,5,2], which represents the fault type, respectively. The severity of the fault, the scope of impact, and the importance of duration are divided into three levels: high, medium, and low. If the fault risk exceeds 7, the dynamic adjustment mechanism of the repair plan is triggered. The genetic algorithm is used to solve the multi-objective optimization model with a population size of 50, a crossover probability of 8, a mutation probability of 1, and 100 iterations to solve the optimal repair plan adjustment strategy. The sensor sampling frequency is dynamically configured according to the fault detection results, including a sampling frequency of 1 time / minute before the fault and 1 time / 5 seconds after the fault. The fuzzy hierarchical analysis method is used to calculate the risk level of each line equipment and location, and a three-level evaluation index system is constructed, including 9 secondary indicators such as equipment importance, fault frequency, and fault impact. The judgment matrix is ​​constructed by pairwise comparison, the weight vector is calculated and consistency test is performed, and the combined weight of each indicator is obtained. The risk level is comprehensively calculated and divided into three levels: high risk (above 7), medium risk (4-7), and low risk (below 4), and a key monitoring list is obtained. .

[0045] S105, for the line equipment and locations that are monitored in key areas, optimize the reconstruction plan for the equipment on the line, optimize the conductor model selection, pole tower and insulator configuration under the premise of meeting the physical and electrical constraints of the line, maximize the line's risk resistance, fault self-healing ability and intelligence level, and generate a line reconstruction construction drawing.

[0046] Specifically, a list of key monitored line equipment and locations is obtained, the list including equipment type, specification parameters, and installation location information; weak links and key sections in the line are identified through fault data analysis and risk assessment methods; a three-dimensional geographic information model of the line is constructed by using Kriging interpolation and inverse distance weighted interpolation spatial interpolation methods, combined with topographic measurement data of the line corridor; a fitness function including line reliability and economic cost is constructed for key monitored line sections; a genetic algorithm and a particle swarm algorithm are used for intelligent optimization algorithms to search for the optimal combination of conductor model, pole tower and insulator configuration, which optimal combination satisfies the physical and electrical constraints of the line; stress and strain analysis of the pole tower and conductor in the candidate reconstruction scheme is performed through finite element analysis software, combined with the actual material parameters and load conditions of the line, to quantitatively evaluate the improvement effect of the risk resistance and fault self-healing ability of the reconstructed line; an indicator system including safety, reliability, economy, and environmental protection criteria layers is established, the indicator weights are determined by using the hierarchical analysis method, the comprehensive score and ranking of each scheme is calculated by using the fuzzy comprehensive evaluation method, and the scheme with the best comprehensive performance is selected as the final line reconstruction scheme.

[0047] In one embodiment, a list of key monitored line equipment and locations is obtained, including equipment type, specification parameters, and installation location information. Through graphic topology analysis and key node evaluation, the equipment and line sections that need to be optimized are determined to form a target range for reconstruction optimization. The topography and meteorological environment external condition data of the line are collected. Through spatial interpolation and digital elevation models, a three-dimensional geographic information model of the line is constructed. At the same time, the voltage level of the line, the material of the conductor, and the internal constraints of the tower structure are obtained, and a line physical and electrical parameter library is established. For the key monitored line sections, a multi-objective optimization algorithm is used to improve the line's risk resistance, fault self-healing ability, and intelligence level as the optimization goals. A fitness function including line reliability and economic cost is constructed to meet the physical and electrical constraints of the line as the restriction conditions. The population size and crossover mutation probability algorithm parameters are reasonably set. Genetic algorithm and particle swarm algorithm are used to search for the optimal combination of conductor model, tower and insulator configuration to obtain the line reconstruction. A set of candidate schemes, a comprehensive evaluation and comparison of candidate reconstruction schemes, an index system including safety, reliability, economy, and environmental protection criteria layers, a 1-9 scale method to determine the index weights, and a membership function, including triangle and trapezoidal functions, is selected. The comprehensive scores and rankings of each scheme are calculated using the analytic hierarchy process and the fuzzy comprehensive evaluation method. Through the finite element analysis software ANSYS and ABAQUS, combined with the actual material parameters and load conditions of the line, stress and strain analysis of the towers and conductors is performed, fatigue life and ultimate strength evaluation is performed, and the improvement effect of the risk resistance and fault self-healing ability of the reconstructed line is quantitatively evaluated. The safety and reliability of the reconstruction scheme is verified, and the scheme with the best comprehensive performance is selected as the final line reconstruction scheme. According to the selected line reconstruction scheme, computer-aided design (CAD) and building information modeling (BIM) technology are used to generate detailed line reconstruction construction drawings, including line plan layout drawings, tower structure drawings, conductor layout drawings, and insulator installation drawings;

[0048] When determining the line sections to be optimized, the Kruskal algorithm is used to perform a minimum spanning tree analysis on the line topology to determine the key line nodes connecting important loads. The Dijkstra algorithm is then used to calculate the electrical distances of the key nodes, and the nodes in the top 10% of the electrical distances are selected as key monitoring objects. When constructing a three-dimensional geographic information model of the line, the Kriging interpolation method is used to spatially interpolate the terrain elevation to generate a 10-meter resolution digital elevation model (DEM). At the same time, a 1:50,000 scale geographic base map is superimposed to extract vegetation and water system environmental elements in the line corridor. In multi-objective optimization, the line reliability fitness function is based on a fault probability model, including a state enumeration method to calculate the expected power loss of the line. The economic cost fitness function takes into account the full life cycle costs of equipment procurement, construction and installation, and operation and maintenance, and multiple fitness functions are combined in a weighted summation manner. In terms of the parameter setting of the optimization algorithm, the population size of the genetic algorithm is twice the number of line sections, the crossover probability is 0.8, and the mutation probability is 0.1; the population size of the particle swarm algorithm is 50, the learning factor is 2, and the inertia weight is 0.7. In the evaluation and selection of the schemes, the fuzzy AHP method is used, and 5 experts in the field are invited to compare the schemes in pairs, build a 10th-order judgment matrix, and perform consistency tests to calculate the weight vectors of the criterion layer indicators. In the finite element analysis, ANSYS Mechanical software is used to establish a geometric solid model of the tower, divide it into tetrahedral grids, and the grid size is 0.1 meters. The load conditions consider wind load, ice load and self-weight. The material properties are selected as Q345 steel, with a yield strength of 345MPa, a Young's modulus of 200GPa, and a Poisson's ratio of 0.3. The stress cloud map and deformation cloud map are used to intuitively analyze the force of the tower.

[0049] S106, according to the construction drawings, carries out live debugging and comprehensive evaluation of the reconstructed local lines, collects line voltage, current and temperature, uses the least squares support vector machine model optimized by ensemble learning to perform fusion analysis on the data, improves the evaluation accuracy through cross-validation and grid search parameter optimization, obtains comparison of various technical indicators before and after line reconstruction, and determines whether the reconstruction effect meets the construction standards of self-healing intelligent power lines.

[0050] Specifically, according to the line reconstruction construction drawing, the target section and test items for live debugging of the local line are determined, and the test items include key indicators of conductor temperature, line voltage, and current harmonics; online monitoring devices and portable test instruments are used to collect real-time voltage, current, and temperature parameters of the reconstructed local line, where voltage transformers and current transformers are installed at key measuring points of the line respectively; for the massive amount of monitoring data collected, a standardized data transmission, storage, and management process is established, and a highly reliable communication protocol is used to ensure the real-time and reliability of data transmission; before analyzing and processing the data, the data is preprocessed, including data cleaning, data normalization, and data dimensionality reduction operations to remove noise, anomalies, and redundant information in the data; for For the same type of monitoring data, corresponding data cleaning and feature selection algorithms are used, including the median filtering algorithm to remove spike noise in the current data and the wavelet transform algorithm to reduce the noise of the temperature data; combined with the support vector machine model, it has good small sample learning ability and nonlinear classification ability, which is suitable for processing high-dimensional complex data. A comprehensive evaluation is conducted on the line monitoring data before and after reconstruction. According to the construction requirements of self-healing intelligent power lines, a comprehensive evaluation index system including power supply reliability, voltage qualification rate, line loss rate and failure rate is constructed; a complete line reconstruction effect evaluation report is formed, which mainly includes the basic situation of the line, reconstruction measures, evaluation indicators, data analysis, comprehensive evaluation, conclusion and suggestion chapters, and is supplemented by rich data charts and visual displays.

[0051] In one embodiment, according to the line reconstruction construction drawing, the target section and test items for live debugging of the local line are determined, including the key indicators of conductor temperature, line voltage, and current harmonics. A detailed live testing plan and operating procedures are formulated, and the configuration of test instruments, personnel division of labor, and specific requirements for safety protection are clarified to ensure the standardization and reliability of the debugging process. Online monitoring devices and portable test instruments are used to collect voltage, current, and temperature parameters of the reconstructed local line in real time. Voltage transformers and current transformers are installed at key measuring points of the line respectively. The selection of measuring points needs to comprehensively consider the topological structure, load distribution, and fault risk factors of the line. The number and distribution of measuring points are determined by sensitivity analysis and optimization configuration algorithms. The temperature measuring points are mainly arranged at the conductors. For the hot spots of hardware, optical fiber temperature measuring devices are used for distributed sensing. For the massive amount of monitoring data collected, it is necessary to establish standardized data transmission, storage, and management processes, use highly reliable communication protocols and encryption algorithms to ensure the integrity and security of the data, and select distributed storage architectures and time series databases to improve data storage and retrieval efficiency. For different types of monitoring data, corresponding data cleaning and feature selection algorithms are used to remove noise and redundancy and extract key features. The median filter algorithm is used to remove spike noise in the current data, and the wavelet transform algorithm is used to reduce the noise of the temperature data. The correlation coefficient method is used to select the voltage feature quantity that is closely related to the line status evaluation. On this basis, a fusion feature matrix of multi-source heterogeneous data is constructed as The input of the model is to build an integrated learning optimized least squares support vector machine model, and to perform fusion modeling on multi-source monitoring data. Considering the limited processing ability of the support vector machine model for time series data, the LSTM deep learning model is first used to extract and encode the time series features, and then the encoded feature vector is input into the support vector machine model for classification prediction. At the level of integrated learning, the random forest algorithm is selected, and multiple decision trees are constructed through Bootstrap sampling and feature random selection. The output is integrated through a voting mechanism to improve the generalization performance and robustness of the model. Cross-validation and grid search techniques are used to optimize the key parameters of the support vector machine model and the random forest model, and the accuracy and reliability of the line evaluation after reconstruction are improved. The integrated learning model is used to comprehensively evaluate the line monitoring data before and after reconstruction. According to the preset knowledge base, a comprehensive evaluation index system including power supply reliability, voltage qualification rate, line loss rate and failure rate is constructed. The weight of each index is calculated by analytic hierarchy process, a fuzzy evaluation matrix is ​​constructed, and the Gaussian function membership function is selected to quantitatively judge the line reconstruction effect. The comprehensive evaluation results before and after the line reconstruction are compared with the construction standards of self-healing intelligent power lines to judge whether the reconstruction effect meets the construction standards of self-healing intelligent power lines, and a reconstruction effect evaluation report is formed. The report mainly includes the basic situation of the line, reconstruction measures, evaluation indicators, data analysis, comprehensive evaluation, conclusion and suggestion chapters, and is supplemented by rich data charts and visual displays.Use an automated report generation engine to improve compilation efficiency and standardization;

[0052] When formulating the line live commissioning plan, we focus on selecting high-voltage lines of 10kV and above for testing. We set a voltage measurement point every 1km and a current measurement point every 500m. The temperature measurement points are selected at the key locations where the wire connection and branch box temperature are prone to rise. The test lasts for 48 hours to cover the load changes around the clock. The ZigBee wireless communication protocol is used to achieve real-time data transmission. The data frame encryption algorithm uses AES-128. The communication distance can reach 1km, the transmission rate is 250kbps, and the monitoring data is stored in Hadoop distributed storage. Framework, HBase column family database, can be horizontally expanded, support TB-level massive time series data storage, when cleaning, the Grubbs algorithm and 3σ criterion are combined to detect and remove abnormal values ​​of current data, wavelet transform uses db4 wavelet basis function, 4-layer decomposition, denoise temperature data, when selecting features, the Pearson correlation coefficient method is used to select voltage features with a correlation greater than 0.6 with the fault label, including 2nd, 3rd, and 5th harmonic content, the cleaned data is serialized according to time, and summarized into a record every 30 minutes, temperature, current, and electricity The pressure constitutes a 60-dimensional feature vector, and the bidirectional LSTM model is used to extract the time series features. The number of hidden layer neurons is 128, the learning rate is 0.001, the batch size is 64, and the training is 100 epochs. The number of decision trees in the random forest model is set to 100, and the integration adopts the Bagging algorithm. The sample sampling ratio is 0.8. The fault prediction accuracy of the final integrated model on the test set reaches more than 95%. The line indicators are quantified into three levels: normal, abnormal, and critical, corresponding to the comprehensive evaluation scores of [0.6,1], [0.3,0.6], [0 ,0.3] interval, a three-level indicator system is established, and the 1-9 scaling method is used to construct the judgment matrix when calculating the weight. The consistency ratio is less than 0.1, and the trapezoidal membership function is used. The trapezoidal coordinates of the failure rate indicator are (10,10,20,40), that is, when the failure rate is less than 10, the membership is 0, and when it is greater than 40, the membership is 1. When generating an evaluation report, Python's automatic report generation library xlsxwriter is used to set a unified cover, directory, header and footer format, automatically fill in the indicator value, generate radar charts and bar charts, and form a standardized and beautiful evaluation report.

[0053] S107, establishing a supporting intelligent operation and maintenance system for the local line reconstruction area, collecting line status data through key technologies of wireless sensor network Internet of Things, using knowledge graph-driven long short-term memory network models to predict and infer equipment health status and fault risks, automatically generating status maintenance and fault repair strategies, and coordinating the dispatch of drones and robots to achieve self-healing operation of local lines.

[0054] Specifically, the network topology and node deployment plan of the wireless sensor network are planned according to the geographical environment, equipment distribution, and grid structure characteristics of the local line reconstruction area; the IoT communication technologies suitable for different scenarios are selected, including Zigbee, LoRa, and NB-IoT, in response to the needs of heterogeneous node management, energy balance optimization, and secure and reliable communication; a preset number of wireless sensors of different types are installed on key equipment to collect equipment operating conditions and environmental status data; the wireless sensors include micrometeorological sensors, conductor sag measurement sensors, and image vibration sensors; a data processing architecture combining edge computing and cloud computing is adopted for massive heterogeneous sensor data; machine learning algorithms are used on gateway nodes to extract features and make preliminary judgments on time series data, and the preprocessed data is uploaded to the distributed storage system of the cloud platform; a distribution network knowledge graph covering local lines is constructed, using an ontology construction method, Define the core concepts and relationships of equipment, components, materials, defects, and maintenance; map multi-source heterogeneous data to the knowledge graph through knowledge extraction, ontology matching, and entity alignment technology; build a long-short-term memory network prediction model for distribution network equipment driven by the knowledge graph; use multi-dimensional time series sensor data and semantic features in the knowledge graph as model inputs, and model the long-term dependency of equipment status through a gated recurrent unit network structure; use the trained prediction model to conduct health status assessment and fault risk analysis on the distribution equipment of local lines; generate preventive status maintenance strategies for equipment with a health index below the threshold; initiate a fault repair emergency plan for equipment with a fault probability exceeding the threshold; send the generated equipment operation and maintenance strategies and repair tasks to the drone and robot collaborative operation platform to achieve self-healing operation of local lines; the drone is responsible for daily inspections, and the robot is responsible for simple maintenance operations.

[0055] In one embodiment, a supporting intelligent operation and maintenance system for the local line reconstruction area is established, the geographical environment, equipment distribution, and grid structure characteristics of the local line reconstruction area are analyzed, the network topology and node deployment plan of the wireless sensor network are planned, and the heterogeneous node management, energy balance optimization, and secure and reliable communication requirements are comprehensively considered. Zigbee, LoRa, and NB-IoT Internet of Things communication technologies are selected, and energy collection and data encryption technologies are combined to establish a highly reliable, low-power, and easily expandable multi-hop self-organizing network. Various types of wireless sensors, including micro-meteorological sensors, conductor sag measurement sensors, and image vibration sensors, are installed on the line towers, switch cabinets, and distribution transformers in the reconstruction area. Key equipment collects equipment operating conditions and environmental status in real time. Data, combined with the actual characteristics of the line and data features, adopts a multi-sensor data fusion algorithm, sets asynchronous correction, joint estimation, and weighted average fusion strategies at the data layer, feature layer, and decision layer respectively to improve the accuracy, reliability, and real-time performance of the data. For massive heterogeneous sensor data, a data processing architecture combining edge computing and cloud computing is adopted. The long short-term memory network (LSTM) algorithm is used on the gateway node to perform feature extraction, anomaly detection, and preliminary judgment preprocessing on the time series data, reduce the computing pressure on the cloud, and improve the response speed. The preprocessed data is uploaded to the distributed storage system of the cloud platform. The time series database (TSDB) and NoSQL database are used to support high concurrency, low latency data reading, writing, and query, and construct Build a distribution network knowledge graph covering local lines, follow the smart grid common information model (CIM) standard, adopt an ontology construction method, define the core concepts and relationships of equipment, components, materials, defects, and maintenance, and form a line professional field ontology with clear semantics and standardized structure. Through knowledge extraction, ontology matching, and entity alignment technology, map sensor data, historical ledgers, operation and maintenance plans, and multi-source heterogeneous data to the knowledge graph to achieve data semantic representation and association analysis, and study the knowledge graph embedding representation learning method to achieve low-dimensional dense vector representation of knowledge, improve the flexibility of knowledge expression and computational efficiency, and drive the knowledge graph to build a long-term and short-term memory network prediction model for distribution network equipment, and integrate multi-dimensional time series sensor data and knowledge The semantic features in the graph are used as model input. The long-term dependency of equipment status is modeled through the gated recurrent unit network structure. The attention mechanism is used to focus on key influencing factors to improve the accuracy of status prediction and fault diagnosis. Transfer learning and incremental learning methods are introduced. The historical operating data of similar equipment and expert diagnosis cases are used to realize cross-domain migration and online evolution of model knowledge, improve the interpretability, robustness and adaptability of the model, and optimize model parameters in combination with digital twin simulation. Through virtual-reality mapping and interaction, the model structure and weights are dynamically adjusted to improve the accuracy and convergence speed of the prediction model. The trained prediction model is used to conduct health status assessment and fault risk analysis on the distribution equipment of the local line. For different types of equipment,Through a large number of simulation experiments, the health index and fault probability thresholds are reasonably set. When the predicted value exceeds the threshold range, the alarm and decision are triggered in time. For equipment with a health index below the threshold, a preventive status maintenance strategy is generated, and targeted inspection, cleaning, debugging and maintenance measures are formulated. For equipment with a fault probability exceeding the threshold, a fault repair emergency plan is initiated. Taking into account multiple factors such as defect type, severity, load importance, and weather conditions, heuristic search and multi-objective optimization algorithms are used to generate a dynamic and coordinated repair scheduling strategy. The generated equipment operation and maintenance strategy and repair tasks are sent to the drone and robot collaborative operation platform to achieve self-healing operation of local lines. Complying with the IEC61850 smart substation standard, human-machine interaction and operation control functions are built. Drones are responsible for daily inspections such as live detection, partial discharge measurement, and external force damage identification. Robots are responsible for vegetation clearing near poles and towers and simple maintenance operations such as insulator replacement. Multi-sensor fusion is used to achieve high-precision environmental modeling and obstacle detection. Trajectory optimization algorithms are used to achieve autonomous robot positioning and navigation. Artificial intelligence visual algorithms are used to achieve fine operation control. 5G communication networks are used to transmit drone aerial images and robot operation information in real time. VR / AR systems are used to achieve remote monitoring and human-machine collaboration to ensure the safety, efficiency, and economy of line operation and maintenance operations.

[0056] In the deployment of wireless sensor network, Zigbee technology is adopted, with a networking range of 1km, supporting 255 nodes, a transmission rate of 250kbps, and an operating frequency band of 2.4GHz. A distributed cluster head rotation algorithm is adopted, with the remaining energy of the node as the weight, and cluster heads are elected in turn to balance the energy consumption of nodes. Knowles masking and channel blacklist technology are used to detect malicious nodes, isolate threat traffic, and analyze line fault statistics. With the span rate and breakdown probability as weights, temperature and vibration sensors are deployed at key towers and branch boxes. The sensor sampling frequency is 1kHz, the accuracy is 1%, and the integration The Dempster-Shafer (DS) evidence theory is used to establish the basic probability assignment function of temperature, vibration and meteorology. The Dempster combination rule is used to calculate the fused fault probability. The edge gateway uses an LSTM network with 128 neurons in the hidden layer, a learning rate of 0.01, and a sliding window every 30 minutes to extract the mean, variance and peak 20-dimensional feature vectors. The knowledge graph contains 23 categories of ontologies, 145 concepts and 376 relationships. The ontology is built based on Protege, and the D2RQ mapping relational database is used. Jena Reasoner reasoning implicit knowledge, knowledge embedding using TransR model, embedding dimension 100, L2 regularization coefficient 0.01, negative sampling rate 0.001, batch size 1000, iteration 100 times, LSTM prediction model input dimension 180, including knowledge graph embedding vector, using Dropout rate 0.5, L1 regularization coefficient 0.001, iteration 500 times, during transfer learning, freeze the first 3 layers of LSTM parameters, fine-tune based on historical data of similar lines, digital twin simulation builds a three-dimensional scene model of the substation, receives real-time alarm events, dynamically renders the health status, and controls virtual sensors to generate simulation data to enhance the robustness of the prediction model. The drone is equipped with visible light, infrared, and ultraviolet The external camera has a spectral range of 0.43-12μm and a spatial resolution better than 5cm. The thermal imager has a sensitivity of 40mK. The robot has a 6-DOF mechanical arm with an arm span of 1.5m and a repeatability accuracy of ±0.1mm. It uses laser SLAM and visual SLAM fusion positioning with a point cloud density of 1.5mm. The operation and maintenance scheduling uses an improved genetic algorithm, which introduces task priority, time window, and position distance constraints. The crossover probability is 0.75, the mutation probability is 0.1, and the iteration is 500 times. The average fitness is improved by 30%. During operation control, the drone autonomously avoids obstacles, the robot models the environment in real time, and performs fine operations on tower bolts and insulator defects. Remote monitoring uses a VR system to render panoramic videos, support voice interaction and gesture control, and the operator feels like he is on the scene.

[0057] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0058] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and rules of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for automatic fault detection and repair of a self-healing intelligent power line, characterized in that: S101. According to the distributed sensor layout plan of the self-healing intelligent power line, the K-Means clustering algorithm is used to perform cluster analysis on the fault sensor data, obtain the spatial distribution characteristics of the fault-frequently occurring areas, and determine the key line sections that need to be repaired locally; S102. For the line equipment parameters and electrical parameters in the local repair section, a convolutional neural network model with heterogeneous data fusion is used to perform feature extraction and matching analysis on the line layout image and parameter data, and the compatibility of the local repair plan with the original line plan is determined. If there is an incompatibility problem, line parameter optimization and equipment transformation are performed; S103. Based on the changes in the line direction and length caused by the local repair, a non-dominated sorting genetic algorithm is used to perform multi-objective optimization on the line layout. By constructing fitness functions of terrain adaptability, electrical performance, and economy, the impact on the global line planning is minimized while meeting the local repair requirements, and an optimization plan for the line layout that takes both local and overall considerations into account is obtained; This step specifically includes: when obtaining the line direction and length change data before and after the local repair, the GIS spatial database is used to extract the line coordinate points within the repair area, and the difference in line length before and after the repair is calculated. If the difference exceeds 5%, it is marked as a major change. For the terrain of the repair area, the digital elevation model DEM is used to extract elevation information, and the average slope is calculated. The area with a slope greater than 15 degrees is marked as complex terrain. In the fuzzy comprehensive evaluation, three factors, namely elevation, slope, and geology, are selected as evaluation indicators, and a fuzzy judgment matrix is ​​constructed. The membership function is determined by the trigonometric function method, and the indicator weight is calculated by the hierarchical analysis method to obtain the terrain adaptability score, with a score range of [0,100]. When calculating the line electrical parameters, the line conductor parameters and the tower coordinate data are used to calculate the power flow through the Newton-Raphson method to obtain the voltage and power distribution. If the voltage qualification rate is lower than 95%, it is judged as poor electrical performance. S104. During the local repair construction process, automatic fault detection is carried out simultaneously. The line operation condition data is collected through distributed sensors. The data is analyzed in real time using the least squares support vector machine model enhanced by transfer learning. Line anomalies and potential faults that occur during the repair process are discovered, and the repair plan is adjusted dynamically to obtain the line equipment and locations that need to be monitored; This step specifically includes: installing current, voltage, and temperature sensors on key line nodes and equipment in the local repair construction area, including installing a current sensor on the low-voltage side of the transformer, with a range of 0-1000A and a precision of level 5; installing a voltage sensor on the busbar, with a range of 0-10kV and a precision of level 2; installing a temperature sensor on the transformer oil tank, with a range of -20℃~150℃ and a precision of ±1℃. According to the network topology and data transmission requirements, the Kruskal algorithm is used to solve the minimum spanning tree, and nodes with a degree centrality greater than 5 are selected as sensor deployment points. On average, 3 sensors are deployed at each key node, and the data is transmitted to the edge computing gateway through the Zigbee wireless communication module. Ad-hoc On-demandDistanceVector routing protocol, transmission delay is less than 100ms, packet loss rate is less than 1%, and a least squares support vector machine model enhanced by transfer learning is constructed. 10,000 historical line operation data and 100 known fault cases are used for pre-training. The model hyperparameters are optimized by 5-fold cross-validation method, and the RBF kernel function is finally selected with a regularization coefficient of 0.

1. The outlier detection algorithm is the box plot method to clean the real-time data and remove outliers that are more than 5 times the distance between the upper and lower quartiles; the Pearson correlation coefficient method is used to select 10 features with a correlation with fault diagnosis greater than 6 to construct a feature vector; the moving average method is used to smooth the time series data with a window size of 10 Fast Fourier transform is used to extract frequency domain features, and the first five harmonic components and spectral entropy are selected as features. Based on statistical process control theory, the upper and lower limits of the characteristic quantity are calculated as ±3σ, and the I-MR control chart is constructed. Combined with the fuzzy diagnosis rule, IF current>2IeAND oil temperature>90℃THEN fault type=transformer overload, the maximum membership method is used to infer the fault type, generate a diagnostic report and disposal suggestions, and the fuzzy comprehensive evaluation method is used to determine the priority of the fault alarm. The weight vector is [3,5,2], which respectively represents the importance of the fault severity, impact range and duration. The priority is divided into high, medium and low levels. If the fault risk exceeds 7, the dynamic adjustment mechanism of the repair plan is triggered, and the genetic algorithm is used to solve A multi-objective optimization model with a population size of 50, a crossover probability of 8, a mutation probability of 1, and 100 iterations was used to solve the optimal repair solution adjustment strategy. The sensor sampling frequency was dynamically configured according to the fault detection results, including a sampling frequency of 1 time / minute before the fault and 1 time / 5 seconds after the fault. The fuzzy hierarchical analysis method was used to calculate the risk level of each line equipment and location, and a three-layer evaluation index system was constructed, including 9 secondary indicators such as equipment importance, fault frequency, and fault impact. A judgment matrix was constructed through pairwise comparison, and the weight vector was calculated and consistency tested to obtain the combined weight of each indicator. The risk level was comprehensively calculated and divided into three levels: high risk (7 or above), medium risk (4-7), and low risk (4 or below), and a key monitoring list was obtained. S105. Optimize the reconstruction plan for the equipment on the line for the key monitored line equipment and locations, optimize the conductor model selection, tower and insulator configuration, and generate the line reconstruction construction drawing on the premise of meeting the physical and electrical constraints of the line; This step specifically includes: when determining the line section to be optimized, use the Kruskal algorithm to perform a minimum spanning tree analysis on the line topology structure, determine the key line nodes connecting important loads, and then use the Dijkstra algorithm to calculate the electrical distance of the key nodes, and select the nodes with the top 10% of the electrical distance as key monitoring objects. When constructing the three-dimensional geographic information model of the line, use the Kriging interpolation method to perform spatial interpolation of the terrain elevation to generate a 10-meter resolution digital elevation model DEM, and at the same time superimpose a 1:50,000 scale geographic base map to extract the vegetation and water environment elements of the line corridor. In multi-objective optimization, the line reliability fitness function is based on the fault probability model, including the state enumeration method to calculate the expected power loss of the line. The economic cost fitness function takes into account the equipment procurement, construction and installation, and operation and maintenance life cycle costs, and adopts a weighted summation method. The fitness function is combined in this way. In terms of the optimization algorithm parameter setting, the population size of the genetic algorithm is twice the number of line sections, the crossover probability is 0.8, and the mutation probability is 0.1; the population size of the particle swarm algorithm is 50, the learning factor is 2, and the inertia weight is 0.

7. In the scheme evaluation and comparison, the fuzzy AHP method is used to make pairwise comparisons, build a 10th-order judgment matrix, and perform consistency tests to calculate the weight vector of the criterion layer indicators. In the finite element analysis, ANSYS Mechanical software is used to establish a geometric solid model of the tower, divide it into tetrahedral grids, and the grid size is 0.1 meters. The load conditions consider wind load, ice load and dead weight. The material properties are selected as Q345 steel, with a yield strength of 345MPa, a Young's modulus of 200GPa, and a Poisson's ratio of 0.

3. The stress cloud map and deformation cloud map are used to intuitively analyze the force of the tower. S106. According to the construction drawings, conduct live commissioning and comprehensive evaluation of the reconstructed local lines, collect line voltage, current and temperature, use the least squares support vector machine model optimized by ensemble learning to perform data fusion analysis, and obtain the comparison of various technical indicators before and after line reconstruction through cross-validation and grid search parameter optimization to determine whether the reconstruction effect meets the construction standards of self-healing intelligent power lines; S107. Establish a supporting intelligent operation and maintenance system for the local line reconstruction area, collect line status data through key technologies of wireless sensor network Internet of Things, use knowledge graph-driven long-short term memory network models to predict and infer equipment health status and fault risks, automatically generate status maintenance and fault repair strategies, and coordinate the dispatch of drones and robots.

2. The method for automatic fault detection and repair of a self-healing intelligent power line according to claim 1, characterized in that: When performing fault sensor data preprocessing in step S101, the 3σ principle is used to remove outliers from the data, that is, data that exceeds the range of ±3 times the standard deviation of the mean is regarded as outliers and is removed. At the same time, wavelet transform is used to reduce noise on the data, and the wavelet basis function is selected as db4 and the number of decomposition layers is 5. The data is decomposed at multiple scales, and the peak signal-to-noise ratio PSNR and root mean square error RMSE of the data are calculated to evaluate the data quality.

3. The method for automatic fault detection and repair of a self-healing intelligent power line according to claim 1, characterized in that: In the step S102, when constructing the line parameter database, the MySQL relational database is used, which includes 10 associated tables, namely, the line basic information table, the tower information table, the conductor information table, and the insulator information table. A B+ tree index is established for the key fields of the line ID, the tower number, and the conductor model. For the line layout image, the Canny operator is used for edge detection to extract the line direction contour. At the same time, the Hough transform is used to detect the tower position, and the conductor direction is extracted by the Hough linear transform to form a 32-dimensional image feature vector. When heterogeneous data is fused, the line ID is used as the primary key and the tower number is used as the foreign key. The spatial indexing technology is used to spatially match the image feature vector and the parameter data. The feature similarity is calculated by the Euclidean distance, and the distance threshold is set to 0.1 to complete the data association. The convolutional neural network model uses 5 convolutional layers and 3 fully connected layers. The convolution kernel sizes are 3x3 and 5x5 respectively, and the pooling kernel size is 2x2. The ReLU activation function and Dropout regularization are used. The training data set contains 10,000 historical line planning records, with a positive and negative sample ratio of 1:1, the Adam optimizer was used, the learning rate was 0.001, the batch size was 128, and the training was 200 epochs. When performing Morris sensitivity analysis, each line parameter was perturbed 10 times, and the perturbation step was 1 / 10 of the parameter range. Based on the change amplitude of the compatibility score output by the network model, the basic influence value and interactive influence value of the parameter were calculated, and 8 key parameters were screened out. The influence coefficient of each parameter was fitted by a logistic regression model. Parameters with an absolute value of the coefficient greater than 0.3 were considered to be the dominant factors of compatibility. In multi-objective optimization, the compatibility score target was set to be greater than 0.95, the investment cost target was set to no more than 120% of the original plan, and the line reliability target was set to reduce the failure rate by 10%. The NSGA-II algorithm was used for 500 iterative searches, with a crossover probability of 0.8, a mutation probability of 0.1, and a population size of 100. The non-dominated solution on the Pareto frontier was selected as the candidate optimization plan.

4. The method for automatic fault detection and repair of a self-healing intelligent power line according to claim 1, characterized in that: When formulating the line live debugging plan, the S106 step selects high-voltage lines of 10kV and above for testing, sets a voltage measuring point every 1km, sets a current measuring point every 500m, and selects temperature measuring points at the wire connection and key parts of the branch box where the temperature is prone to rise. The test lasts for 48 hours, covering the load changes around the clock. The ZigBee wireless communication protocol is used, the data frame encryption algorithm uses AES-128, the communication distance is 1km, the transmission rate is 250kbps, and the monitoring data uses the Hadoop distributed storage framework ,HBase column family database, horizontal expansion, supports TB-level massive time series data storage. When cleaning, the Grubbs algorithm and the 3σ criterion are combined to detect and remove abnormal values ​​of current data. The wavelet transform uses the db4 wavelet basis function and 4-layer decomposition to denoise the temperature data. When selecting features, the Pearson correlation coefficient method is used to select voltage features with a correlation greater than 0.6 with the fault label, including the 2nd, 3rd, and 5th harmonic content. The cleaned data is serialized according to time and summarized into a record every 30 minutes, including temperature, current, and electricity. The pressure constitutes a 60-dimensional feature vector, and the bidirectional LSTM model is used to extract the time series features. The number of hidden layer neurons is 128, the learning rate is 0.001, the batch size is 64, and the training is 100 epochs. The number of decision trees in the random forest model is set to 100, and the integration adopts the Bagging algorithm. The sample sampling ratio is 0.

8. The fault prediction accuracy of the final integrated model on the test set reaches more than 95%. The line indicators are quantified into three levels: normal, abnormal, and critical, corresponding to the comprehensive evaluation scores of [0.6,1], [0.3,0.6] respectively. , [0,0.3] interval, establish a three-level indicator system, use the 1-9 scaling method to construct a judgment matrix when calculating the weights, the consistency ratio is less than 0.1, and the trapezoidal membership function is used. The trapezoidal coordinates of the failure rate indicator are 10, 10, 20, 40, that is, when the failure rate is less than 10, the membership is 0, and when it is greater than 40, the membership is 1. When generating an evaluation report, use Python's automated report generation library xlsxwriter to set a unified cover, directory, header and footer format, automatically fill in the indicator value, generate radar charts and bar charts, and form an evaluation report.

5. The method for automatic fault detection and repair of a self-healing intelligent power line according to claim 1, characterized in that : In the wireless sensor network deployment, the step S107 adopts Zigbee technology, with a networking range of 1km, supports 255 nodes, a transmission rate of 250kbps, a working frequency band of 2.4GHz, and adopts a distributed cluster head rotation algorithm. The node residual energy is used as the weight to elect cluster heads in turn to balance node energy consumption. Knowles masking and channel blacklist technology are used to detect malicious nodes, isolate threat traffic, analyze line fault statistics, and use the span rate and breakdown probability as weights. Temperature and vibration sensors are deployed at key towers and branch boxes. The sensor sampling frequency is 1kHz. The accuracy is 1%. The Dempster-Shafer evidence theory is used in fusion to establish the basic probability assignment function of temperature, vibration and meteorology. The Dempster combination rule is used to calculate the fused fault probability. The edge gateway uses the LSTM network with 128 neurons in the hidden layer, a learning rate of 0.01, and a sliding window every 30 minutes to extract the mean, variance and peak 20-dimensional feature vectors. The knowledge graph contains 23 categories of ontologies, 145 concepts and 376 relationships. The ontology is built based on Protege, and the D2RQ mapping relational database is used. Jena Reasoner reasoning implicit knowledge, knowledge embedding using TransR model, embedding dimension 100, L2 regularization coefficient 0.01, negative sampling rate 0.001, batch size 1000, iteration 100 times, LSTM prediction model input dimension 180, including knowledge graph embedding vector, using Dropout rate 0.5, L1 regularization coefficient 0.001, iteration 500 times, during transfer learning, freeze the first 3 layers of LSTM parameters, fine-tune based on historical data of similar lines, digital twin simulation builds a 3D scene model of the substation, receives real-time alarm events, dynamically renders the health status, and controls virtual sensors to generate simulations Data, the drone is equipped with visible light, infrared, and ultraviolet cameras with a spectral range of 0.43-12μm, a spatial resolution of 5cm, a thermal imager sensitivity of 40mK, a robot with 6 degrees of freedom mechanical arm, an armspan of 1.5m, and a repeatability accuracy of ±0.1mm. It uses laser SLAM and visual SLAM fusion positioning, and the point cloud density is 1.5mm. The operation and maintenance scheduling uses an improved genetic algorithm, introducing task priority, time window, and position distance constraints, with a crossover probability of 0.75, a mutation probability of 0.1, and 500 iterations. During operation control, the drone autonomously avoids obstacles, the robot models the environment in real time, operates on the defects of tower bolts and insulators, and remote monitoring uses a VR system.

Images