An ultrahigh-voltage substation inspection path planning method and system

Abstract

The application relates to the technical field of substation inspection path planning, and discloses a super-high-voltage substation inspection path planning method and system. The method collects multi-dimensional monitoring data such as equipment operation state data, environmental parameter data and equipment historical maintenance data of a super-high-voltage substation, dynamically evaluates the data, and generates equipment operation health degree, environmental influence factor value and equipment maintenance urgency. A three-dimensional evaluation space is constructed based on the three key indexes, and adaptive grid division is carried out in the three-dimensional evaluation space. Then, the density clustering algorithm is used to identify the optimal inspection interval in the divided grid, an inspection path sequence is generated according to the optimal inspection interval, a path optimization algorithm is used to carry out smoothing processing on the inspection path sequence, and finally, an inspection path scheme is output. The application comprehensively considers various factors, effectively improves the scientificity and efficiency of substation inspection path planning, can timely find potential problems of equipment, and guarantees the stable operation of the substation.

Description

Technical Field

[0001] This invention relates to the field of substation inspection route planning technology, specifically to a method and system for planning inspection routes in ultra-high voltage substations. Background Technology

[0002] As a critical hub in the power grid, substations undertake important tasks such as voltage conversion, power distribution, and control. The stable operation of their equipment directly affects the safety and reliability of the entire power grid. A substation equipment failure can trigger a large-scale power outage, severely impacting industrial production and causing huge economic losses, while also greatly inconveniencing residents and even disrupting social order. Therefore, ensuring the reliable operation of substation equipment is paramount, and substation inspection is an effective means of identifying and eliminating potential equipment faults.

[0003] Regular inspections allow for real-time monitoring of equipment operating status and parameters, enabling timely detection of potential problems and prevention of malfunctions. For example, electrical equipment may malfunction due to overheating or poor contact during prolonged operation. Regular inspections, utilizing specialized equipment and methods, can effectively detect these issues, thereby preventing power outages caused by equipment failures and ensuring the safe and stable operation of the power system.

[0004] With the vigorous construction and rapid development of ultra-high voltage AC and DC power projects, the scale of substations is constantly expanding, and the number of equipment is increasing daily. Traditional substation inspection methods are no longer suitable for the characteristics of large areas and numerous devices in ultra-high voltage substations. Traditional manual inspection methods have many drawbacks, one of which is low efficiency. Manual inspection requires inspectors to check each piece of equipment one by one, which consumes a lot of time and energy. Especially in large substations, where equipment is widely distributed and inspection routes are complex, the inspection cycle is long, making it difficult to meet the need for timely detection of equipment faults.

[0005] Manual inspections are costly in terms of manpower. Substation inspections require specialized personnel, and as the number and scale of substations increase, the number of required personnel also increases, undoubtedly raising the labor costs for power companies. Furthermore, manual inspections are susceptible to environmental factors. In extreme weather conditions such as high temperatures, cold, and rain, the working conditions for inspectors are harsh, affecting not only efficiency and quality but also potentially threatening their personal safety. In addition, manual inspections are subject to subjectivity and uncertainty, easily leading to missed or incorrect inspections, making it difficult to guarantee the accuracy and reliability of the results. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for planning inspection routes in ultra-high voltage substations, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a method for planning inspection routes in ultra-high voltage substations, the method comprising:

[0008] Collect multi-dimensional monitoring data of ultra-high voltage substations, including equipment operating status data, environmental parameter data, and equipment historical maintenance data;

[0009] The multi-dimensional monitoring data is dynamically evaluated to generate equipment operational health, environmental impact factor values, and equipment maintenance urgency.

[0010] A three-dimensional evaluation space is constructed based on the equipment's operational health, environmental impact factor values, and equipment maintenance urgency, and adaptive meshing is performed in the three-dimensional evaluation space.

[0011] The optimal inspection interval is identified in the divided grid using a density clustering algorithm, and an inspection path sequence is generated based on the optimal inspection interval.

[0012] The inspection path sequence is smoothed using a path optimization algorithm to output the final inspection path scheme.

[0013] Preferably, the dynamic evaluation of the multi-dimensional monitoring data includes:

[0014] The sliding window analysis method is used to process the equipment operating status data, and the coefficient of variation of the data characteristics within each time window is calculated as the equipment operating health.

[0015] The fluctuation patterns of environmental parameter data are analyzed using the entropy method to quantify the values ​​of environmental influencing factors.

[0016] The time decay model is used to process historical maintenance data of the equipment, and different weights are assigned according to the time of maintenance to calculate the urgency of equipment maintenance.

[0017] Preferably, the construction of the three-dimensional evaluation space includes:

[0018] A spatial coordinate system is established with equipment operating health as the X-axis, environmental influencing factors as the Y-axis, and equipment maintenance urgency as the Z-axis.

[0019] Historical monitoring data samples are projected onto the spatial coordinate system to form a data point cloud distribution;

[0020] Based on the density characteristics of the data point cloud, an adaptive spatial grid division granularity is set.

[0021] Preferably, the adaptive mesh partitioning includes:

[0022] Calculate the variance of the point cloud distribution on each coordinate axis, and determine the density of the grid division based on the magnitude of the variance;

[0023] Fine-grained grids are used in areas with dense data points, while coarse-grained grids are used in areas with sparse data points.

[0024] Assign a unique identifier to each grid cell and count the number of data points contained in each grid cell.

[0025] Preferably, identifying the optimal inspection interval in the divided grid using a density clustering algorithm includes:

[0026] Region growing begins with the grid cell with the highest data point density;

[0027] Set a density threshold and merge adjacent grid cells whose density reaches the threshold into a cluster;

[0028] Calculate the centroid coordinates of each cluster and determine the grid cell closest to all centroids as the core grid cell.

[0029] Preferably, the generation of the inspection path sequence includes:

[0030] Extract historical inspection interval data corresponding to the core grid unit as a candidate interval set;

[0031] Time series analysis was performed on the candidate interval set, and the interval value with the highest frequency of occurrence was used as the baseline interval.

[0032] Based on the baseline interval and equipment distribution topology, determine the starting point and the sequence of points along the inspection path.

[0033] Preferably, the smoothing process of the inspection path sequence using a path optimization algorithm includes:

[0034] Establish a directed graph model of the inspection path, where nodes represent inspection points and edges represent inspection path segments.

[0035] The shortest Hamiltonian path in a directed graph is found using dynamic programming. The shortest path is then curve-fitted to eliminate sharp angles in the path.

[0036] Curve fitting is performed on the shortest path obtained, and acute angle turns in the path are eliminated, including:

[0037] The discrete path points are made continuous using cubic spline interpolation.

[0038] Check whether the fitted path meets the mechanical motion constraints of the inspection equipment.

[0039] Local adjustments are made to path segments that do not meet the constraints to ensure the executability of the path.

[0040] Preferably, the establishment of the directed graph model includes:

[0041] Calculate the Euclidean distance between each inspection point based on the geographical coordinates of the equipment.

[0042] Considering the differences in the importance of devices, different access priority weights are set for different device nodes. Combining distance and priority weights, the weight matrix of each edge in the directed graph is calculated.

[0043] Preferably, the method of using dynamic programming to find the shortest Hamiltonian path in a directed graph includes:

[0044] An improved ant colony algorithm is used for path search in a directed graph;

[0045] Set pheromone evaporation factors and heuristic factors to balance global search and local search capabilities;

[0046] Through multiple iterations of optimization, the solution converges to the optimal path.

[0047] Preferably, the present invention also includes an ultra-high voltage substation inspection route planning system, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the above-described ultra-high voltage substation inspection route planning method.

[0048] Compared with the prior art, the beneficial effects of the present invention are:

[0049] This invention collects multi-dimensional monitoring data from ultra-high voltage substations, including equipment operating status data, environmental parameter data, and historical equipment maintenance data. It then dynamically evaluates this data to generate equipment operational health, environmental impact factor values, and equipment maintenance urgency. This multi-dimensional data fusion approach comprehensively and accurately reflects the true operating status of the equipment.

[0050] Traditional substation inspections often focus only on certain operating parameters of equipment, failing to comprehensively consider environmental factors and historical maintenance records, easily leading to the neglect of potential equipment problems. The multi-dimensional monitoring data acquisition and dynamic evaluation method employed in this invention overcomes this deficiency. Equipment operating status data directly reflects the current working condition of the equipment, environmental parameter data reflects the impact of the external environment on equipment operation, and historical maintenance data helps us understand past fault conditions and maintenance records, thereby more accurately judging the equipment's health status and maintenance needs. Through the comprehensive analysis of these multi-dimensional data, reliable evidence can be provided for inspections, enabling inspection personnel to conduct more targeted equipment checks and promptly identify potential equipment faults.

[0051] A major innovation of this invention is the construction of a three-dimensional evaluation space based on equipment operational health, environmental impact factors, and equipment maintenance urgency, followed by adaptive mesh generation within this space. Traditional path planning methods often fail to adequately consider the interrelationships between equipment status, environmental factors, and maintenance needs, resulting in a lack of scientific rigor and flexibility in path planning. The three-dimensional evaluation space constructed in this invention, however, incorporates these three crucial factors into a unified framework for analysis, enabling a more comprehensive reflection of the actual needs of substation inspections.

[0052] Adaptive mesh generation in the three-dimensional evaluation space allows path planning to be flexibly adjusted according to the characteristics of different areas. For areas with low equipment health, significant environmental impact, or high maintenance urgency, finer meshes can be used to increase the frequency and detail of inspections. Conversely, for areas with good equipment operation and a relatively stable environment, the mesh size can be appropriately relaxed to reduce unnecessary inspection workload. This adaptive planning method dynamically adjusts inspection strategies based on the actual conditions of the substation, improving the scientific rigor and flexibility of path planning, and enabling a more rational allocation of inspection resources, thereby better meeting the complex and ever-changing inspection needs of ultra-high voltage substations.

[0053] This invention identifies the optimal inspection interval within a divided grid using a density clustering algorithm, generates an inspection path sequence based on the optimal interval, and then smooths the inspection path sequence using a path optimization algorithm to output the final inspection path scheme. This combination of clustering and optimization algorithms significantly improves inspection efficiency. Attached Figure Description

[0054] Figure 1 This is a schematic diagram illustrating the working principle of the ultra-high voltage substation inspection path planning method described in this invention.

[0055] Figure 2 A flowchart for dynamic evaluation of multi-dimensional monitoring data;

[0056] Figure 3 A flowchart for adaptive mesh generation. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Please see Figure 1This invention provides a method and system for planning inspection routes in ultra-high voltage substations. The method includes the collection of multi-dimensional monitoring data covering equipment operating status data, environmental parameter data, and historical equipment maintenance data. This data is acquired in real time through a sensor network and historical database deployed within the substation. Equipment operating status data includes current, voltage, and temperature parameters; environmental parameter data includes humidity, wind speed, and pollution index; and historical equipment maintenance data records maintenance time, maintenance type, and number of failures. A dynamic evaluation process processes the multi-dimensional monitoring data to generate equipment operating health, environmental impact factor values, and equipment maintenance urgency. These evaluation indicators are used to construct a three-dimensional evaluation space. The three-dimensional evaluation space establishes a spatial coordinate system with equipment operating health as the X-axis, environmental impact factor values ​​as the Y-axis, and equipment maintenance urgency as the Z-axis. Historical monitoring data samples are projected onto this spatial coordinate system to form a data point cloud distribution. Adaptive grid partitioning sets the grid partitioning granularity based on the density characteristics of the data point cloud. A density clustering algorithm identifies the optimal inspection interval within the partitioned grid, generating an inspection route sequence. The path optimization algorithm smooths the inspection path sequence and outputs the final inspection path scheme. The path optimization algorithm uses dynamic programming to solve the shortest path and combines curve fitting technology.

[0059] Example 1: See Figure 2 The dynamic evaluation process of multi-dimensional monitoring data begins with the processing of equipment operating status data. A sliding window analysis method is used, setting a fixed time window size. The window size is adjusted based on the data sampling frequency; higher frequency data corresponds to smaller windows, and lower frequency data to larger windows. The sliding window analysis method calculates the coefficient of variation (COP) of data characteristics within each time window. The COP is obtained by dividing the standard deviation of the data within the window by the mean. Equipment operating status data includes current, voltage, and temperature parameters. The selection of data characteristics covers mean, variance, and peak values. A lower COP indicates a more stable equipment operating status, and the equipment's health is directly assigned by the COP value; the lower the value, the higher the health. Environmental parameter data analysis uses the entropy method. The entropy method calculates the probability distribution of environmental parameter data based on information theory principles. Environmental parameter data includes humidity, wind speed, and pollution index. The entropy calculation process first normalizes the parameter values ​​to the zero-to-one range, then calculates the probability of each parameter value. The entropy formula uses negative probability logarithmic summation. Environmental influencing factor values ​​are quantified by the entropy results; a higher entropy value indicates greater environmental fluctuations, and correspondingly higher influencing factor values. The processing of historical equipment maintenance data utilizes a time decay model, which assigns weights using an exponential decay function. The closer the maintenance time is to the current time, the greater the weight. The weighting function is exponentially decaying. Historical equipment maintenance data records maintenance time, maintenance type, and number of failures. The urgency of equipment maintenance is calculated through weighted averages, with the weight inversely proportional to the difference in maintenance time. A higher urgency value indicates that the equipment needs to be inspected as soon as possible.

[0060] The three-dimensional assessment space is constructed using a spatial coordinate system with equipment operational health as the X-axis, environmental impact factors as the Y-axis, and equipment maintenance urgency as the Z-axis. The coordinate system units are normalized to ensure dimensional consistency. The numerical ranges of equipment operational health, environmental impact factors, and equipment maintenance urgency are adjusted to the zero-to-one interval. Historical monitoring data samples are projected onto the spatial coordinate system to form a data point cloud distribution. Analysis of the data point cloud distribution involves preliminary grouping using clustering algorithms. The point cloud distribution reveals concentrated areas of equipment status. The spatial grid granularity is adaptively set according to the density characteristics of the data point cloud, with finer granularity in densely populated areas and coarser granularity in sparsely populated areas. During the data point cloud projection process, the timestamps of historical monitoring data samples are aligned, and the values ​​of equipment operational health, environmental impact factors, and equipment maintenance urgency are standardized to the same numerical range. The establishment of the spatial coordinate system considers the dynamic range of data values, and the coordinate axis scales are adjusted according to data extrema. Visual analysis of the data point cloud distribution helps identify outliers. Filtering outliers improves the accuracy of grid division. The adaptive grid division granularity setting is automatically completed by an algorithm, and the grid division results are stored as a data structure. The implementation of the sliding window analysis method involves setting the window sliding step size, which is proportional to the window size to ensure complete data coverage. In the coefficient of variation calculation, the selection of data features is based on domain knowledge: the mean reflects the average level, the variance reflects the degree of fluctuation, and the peak reflects extreme cases. The assessment results of equipment operating health are used for subsequent spatial projection. When analyzing environmental parameter data using the entropy method, the priority of parameter types considers actual impact; temperature parameters may have higher weight, followed by humidity parameters. The normalization of entropy calculation results ensures that environmental impact factor values ​​are within a standard range. The weight decay rate of the time decay model is configurable, and the decay constant is adjusted based on maintenance history. The calculation of equipment maintenance urgency integrates multiple maintenance records, and the urgency value is dynamically updated. The coordinate system of the three-dimensional evaluation space is fixed, with the X, Y, and Z axes orthogonal. The density calculation of the data point cloud distribution uses a kernel density estimation method; high-density areas have dense point clouds, and low-density areas have sparse point clouds. The adaptive mechanism for spatial grid division is based on a density threshold, which is determined by the overall point cloud density.

[0061] The sliding window analysis of equipment operating status data allows for real-time updates. As new data flows in, the window slides and the coefficient of variation is recalculated. The trend of equipment health is tracked through continuous windows. Entropy-based processing of environmental parameter data supports multi-parameter fusion, with weighted merging of entropy values ​​from different environmental parameters, weights allocated based on parameter importance. The time decay model for historical equipment maintenance data has an adjustable exponential decay constant; the decay rate affects the urgency sensitivity. The point cloud projection in the 3D assessment space uses a Cartesian coordinate system, with point positions determined by 3D coordinates. Feature extraction of the point cloud distribution includes centroid calculation and boundary determination. The adaptive meshing uses cubic grid cells with variable sizes: fine-grained meshes are small, and coarse-grained meshes are large. Grid cell identifiers include 3D indexes for fast lookup. The outputs of the dynamic assessment process—equipment health, environmental influencing factor values, and equipment maintenance urgency—serve as 3D spatial coordinate inputs. The establishment of the spatial coordinate system ensures the independence of each axis. Cluster analysis of the point cloud distribution identifies high-density regions, and the adaptive meshing handles non-uniform distributions. Window size optimization in the sliding window analysis method is achieved through cross-validation, and stability testing of the coefficient of variation ensures the reliability of the health data. The preprocessing of environmental parameter data using the entropy method includes imputation of missing values, and the probability estimation for entropy calculation uses a histogram method. The weight allocation of the time decay model considers differences in maintenance type, with higher weights for critical maintenance and lower weights for routine maintenance. The point cloud projection algorithm for the 3D assessment space uses matrix operations, and projection efficiency is improved through parallel processing. The granularity adjustment algorithm for mesh generation iteratively optimizes the mesh size, and the counting function is used for data point statistics in the mesh cells. The dynamic evaluation of multi-dimensional monitoring data integrates multiple algorithms, with sliding window analysis, the entropy method, and the time decay model working collaboratively. The evaluation results—equipment operational health, environmental impact factor values, and equipment maintenance urgency—are transmitted to the space construction stage. The establishment of the 3D assessment space provides a foundation for visualization analysis, pattern recognition of data point cloud distribution assists decision-making, and adaptive mesh generation enhances spatial resolution. The processing frequency of equipment operating status data is synchronized with data updates, the entropy calculation cycle for environmental parameter data is configurable, and the time decay model for historical equipment maintenance data supports real-time weight updates. Normalization of the spatial coordinate system eliminates the influence of dimensions, the generation of the density map of the data point cloud distribution uses a Gaussian kernel, and the adaptive threshold for mesh generation is dynamically calculated.

[0062] The implementation details of the sliding window analysis method include window boundary handling, data overlap to avoid information loss, robust handling of outliers in the coefficient of variation calculation, and using the median instead of the mean. The entropy method analyzes the fluctuation patterns of environmental parameter data considering time correlation, and multi-dimensional extension of entropy calculation is used to process vector data. The time decay model calculates the equipment maintenance urgency using a linear weighted formula, with the weight decay function in exponential form to ensure smoothness. Affine transformation is used for the projection coordinate transformation of the data point cloud in the 3D evaluation space, and the point cloud distribution visualization tool integrates 3D rendering. The algorithm complexity optimization for adaptive mesh generation is achieved through spatial indexing, and a tree structure is used for mesh cell management. Parameter configuration for the dynamic evaluation process is completed through configuration files; the sliding window size, entropy parameter, and decay constant can be externally adjusted. The coordinate system orientation of the 3D evaluation space is customizable, and the coordinate origin of the data point cloud projection is set based on the data center. The adaptive granularity algorithm for mesh generation is based on machine learning, with density clustering guiding mesh adjustment. The numerical range standardization of equipment operating health, environmental impact factor values, and equipment maintenance urgency uses minimum-maximum scaling, and a filtering algorithm is used for point cloud denoising in spatial projection. The collection and dynamic evaluation of multi-dimensional monitoring data are conducted in a closed loop, with new data flowing in to re-evaluate and update the three-dimensional space. The construction of the three-dimensional evaluation space and the setting of the grid are automated, and the grid division results are used for subsequent cluster analysis. The real-time performance of the sliding window analysis method meets the needs of high-speed data flow, the entropy method has high processing efficiency for environmental parameter data, and the time decay model calculates weights quickly. Historical data backup of the point cloud distribution in the spatial coordinate system is used, and the adaptive parameters of the grid division learn from changes in the data distribution.

[0063] The sliding window analysis of equipment operating status data adapts the window size to the data flow rate, and the coefficient of variation calculation optimizes numerical stability. For environmental parameter data, entropy-based parameter selection is used, and entropy results are calibrated against historical benchmarks. The time decay model for historical equipment maintenance data dynamically adjusts the decay constant, and urgency calculation caching improves performance. The point cloud projection coordinates in the 3D assessment space are highly accurate, and compressed storage of the point cloud distribution saves space. The adaptive mesh generation features continuously adjustable grid cell sizes and smoothed grid boundaries. The dynamic assessment outputs standardized numerical formats for equipment operating health, environmental impact factor values, and equipment maintenance urgency, using floating-point representations of 3D spatial coordinates. Point cloud statistics are updated in real-time, and the adaptive mesh generation algorithm is executed in parallel. The sliding window analysis, entropy method, and time decay model algorithms are implemented using programming language library functions, and computational resources are evenly allocated during the construction of the 3D assessment space. The dynamic assessment of multi-dimensional monitoring data is seamlessly integrated with 3D space construction, and the data flow pipeline ensures data consistency. Error control in the calculation of equipment operating health, environmental impact factor values, and equipment maintenance urgency is achieved through numerical verification. The spatial coordinate system is clearly defined with coordinate axis labels, and color coding of the data point cloud distribution enhances visualization. Adaptive grid generation allows for fast retrieval of grid cell indexes, and grid data is exported in a standard format.

[0064] Example 2: See Figure 3The adaptive mesh generation process begins with a quantitative analysis of the distribution characteristics of the data point cloud within the 3D evaluation space. The variance of the data point cloud distribution along each coordinate axis is calculated using standard statistical methods. The numerical sequences of the X-axis equipment operational health, the Y-axis environmental influencing factor values, and the Z-axis equipment maintenance urgency are input into the variance calculation function. The variance calculation function uses the sample variance formula to process the data point coordinate values. The magnitude of the distribution variance directly determines the density of the mesh generation; coordinate axes with larger variances have higher mesh densities, while those with smaller variances have lower mesh densities. Fine-grained mesh generation is used in densely populated data point areas, with smaller mesh sizes and more elements. Coarse-grained mesh generation is used in sparsely populated data point areas, with larger mesh sizes and fewer elements. Each mesh element is assigned a unique identifier, and the identifier encoding scheme uses a 3D index structure. The number of data points contained within a mesh element is counted using a traversal mapping algorithm. The density-based clustering algorithm initiates the region growth process from the grid cell with the highest data point density. Data point density is calculated based on the ratio of the number of data points within a grid cell to its volume. The region growth process checks the density values ​​of adjacent grid cells. The criteria for determining adjacent grid cells include three connectivity types: face adjacency, edge adjacency, and point adjacency. The density threshold is set as the arithmetic mean of the global data point density. Adjacent grid cells whose density reaches the threshold are merged into clusters. The boundaries of these clusters are determined by the spatial connectivity of the grid cells. The centroid coordinates of each cluster are calculated using the arithmetic mean method, processing the coordinates of all data points within the cluster. The grid cell closest to all centroids is determined using the Euclidean distance formula. The core grid cell is selected based on having the smallest sum of distances to the centroids of all clusters.

[0065] The calculation of variance requires independent processing for each coordinate axis. The numerical sequence of the X-axis device health status is first normalized to eliminate the influence of dimensions. The numerical range of the variance calculation results is used to guide the grid partitioning strategy. More grid division points are set in the high-variance coordinate axis direction, and fewer grid division points are set in the low-variance coordinate axis direction. The density of the grid partitioning is directly proportional to the variance value. When the variance value exceeds a set threshold, the number of grid divisions is automatically increased; when the variance value is below the set threshold, the number of grid divisions is reduced. Fine-grained grid partitioning improves spatial resolution in dense data point areas, and the grid size is reduced to a preset minimum value. Coarse-grained grid partitioning reduces computational complexity in sparse data point areas, and the grid size is expanded to a preset maximum value. The encoding rule for the unique identifier of the grid cell adopts a hierarchical index structure. The first layer identifies the X-axis coordinate segment, the second layer identifies the Y-axis coordinate segment, and the third layer identifies the Z-axis coordinate segment. The identifier generation algorithm ensures global uniqueness. The data point counting process establishes a mapping relationship between grid cells and data points. Each data point is assigned to a corresponding grid cell according to its coordinate value, and the data point counter within the grid cell is automatically updated. The region growth process of the density clustering algorithm uses a queue data structure to manage the grid cells to be processed. The initial seed cell is selected from the cell with the maximum density, and the region growth direction follows the principle of spatial proximity.

[0066] The dynamic adjustment mechanism for the density threshold considers the overall distribution characteristics of the data point cloud. The threshold calculation uses a weighted average method to handle local density values. The merging operation of adjacent grid cells whose density reaches the threshold is optimized using a disjoint-set data structure. The centroid coordinate calculation of clusters is iterative; the cluster centroid is recalculated after each merging operation, and the accuracy of the centroid coordinates is controlled within the allowable error range. The determination process for core grid cells calculates the Euclidean distance from each grid cell to the centroids of all clusters. Distance summation is accelerated using vectorized computation, and the grid cell with the minimum distance is marked as the core grid cell. The adaptive grid partitioning algorithm for grid size adjustment is based on the variance change rate. Regions with a large variance change rate experience large grid size changes, and a smooth transition technique is used to handle grid partitioning boundaries to avoid abrupt changes. Fine-grained grid partitioning in densely populated data point areas increases the number of grid cells, and a spatial index tree structure is used for grid cell management to improve query efficiency. Coarse-grained grid partitioning in sparsely populated data point areas reduces the number of grid cells, and the grid cell merging rules consider spatial continuity. The allocation process for grid cell identifiers is synchronized with grid partitioning, and the identifier encoding includes spatial location information to support rapid positioning. The density-based clustering algorithm's region growth termination conditions include a density threshold and spatial boundary constraints. The region growth process records the access status of grid cells to prevent duplicate processing. Connectivity checks between adjacent grid cells use a 3D neighborhood template, and 26-neighborhood connectivity ensures growth in all spatial directions. The cluster merging operation maintains a list of cluster attributes, including cluster size, density range, and centroid coordinates. Distance calculations for core grid cells are optimized using spatial pruning techniques to reduce unnecessary distance computations.

[0067] The variance calculation preprocessing for adaptive mesh generation includes outlier filtering, which uses the three-standard-deviation rule to ensure data quality. Configurable parameters control the mesh density, adjustable via a graphical interface. A lower limit for fine-grained mesh size prevents over-subdivision, while an upper limit for coarse-grained mesh size avoids excessively coarse division. Compressed encoding is used to store unique identifiers for each mesh cell, saving memory. Parallel computation for data point counting utilizes a multi-core processor architecture, with results updated in real-time. Region growth speed in density-based clustering is optimized through load balancing, and a block-based strategy is used for large-scale point cloud processing. Density threshold settings consider data distribution skewness, which is addressed using logarithmic transformation. Weighting factors are incorporated into the calculation of cluster centroid coordinates, allocated based on data point importance. A secondary candidate mechanism is added for core mesh cell selection, activating a backup core mesh cell when the primary core cell fails. Adaptive mesh generation uses integer coordinate mapping for mesh boundary alignment, avoiding floating-point precision errors. The grid subdivision rules for densely populated data point areas support multi-level refinement, and the depth of fine-grained grid division is configurable. For sparsely populated data point areas, grid merging conditions consider spatial proximity, and the attribute inheritance rules for merged grid cells are clearly defined. The query function for grid cell identifiers supports reverse location, and the conversion from identifier to spatial coordinates is fast and accurate. The region growth path record of the density clustering algorithm is used for visual debugging, and the growth process animation helps understand the algorithm's behavior. Density comparisons between adjacent grid cells use a relative difference threshold to avoid the inadequacy of absolute thresholds. Morphological operations are used for smoothing the boundaries of clusters, improving the jagged edge phenomenon. The core grid cell verification process checks the cell density value; low-density cells, even those with the smallest distance, are not selected.

[0068] The variance calculation window for adaptive mesh partitioning is adjustable, and the window size affects the stability of variance estimation. Adaptive learning for mesh density is trained using historical data, and machine learning algorithms optimize partitioning parameters. Fine-grained mesh partitioning automatically triggers garbage collection based on memory usage monitoring, while coarse-grained mesh partitioning uses a compressed format for disk storage. The indexing of unique mesh cell identifiers uses a B-tree structure, improving identifier lookup speed. Distributed computing for data point count statistics supports cluster deployment, and sharding the statistical task improves efficiency. Parallelization of region growing in density clustering algorithms utilizes GPU acceleration, significantly reducing processing time for large-scale point clouds. Adaptive adjustment of density thresholds is based on real-time data stream changes, with dynamic thresholds adapting to changes in data distribution. The centroid coordinate update mechanism for clusters supports incremental calculations, eliminating the need to recalculate all centroids when new data points are added. A caching strategy for core mesh cells stores frequently used distance calculation results, improving response speed for repeated queries. Adaptive mesh size adjustment considers computational resource limitations, automatically switching to coarse-grained partitioning when memory is insufficient. Density gradient judgment is added to the mesh subdivision conditions in densely populated areas, increasing subdivision intensity in areas with drastic density changes. The grid merging strategy for sparse data points considers the shape of the grid cells, avoiding excessive merging of long, narrow grid cells. The algorithm for assigning unique identifiers to grid cells avoids identifier conflicts, and in a distributed environment, machine numbers are incorporated into identifier generation.

[0069] The density clustering algorithm prioritizes the growth direction of regions, prioritizing growth in high-density directions to improve clustering quality. The density transfer model for adjacent grid cells considers spatial correlation, with the density transfer coefficient based on distance decay. Stability checks are incorporated into cluster merging decisions to avoid frequent merging and splitting. Robustness processing is added to the core grid cell selection algorithm, ensuring that outlier centroid coordinates do not affect core cell selection. Confidence interval estimation is added to the variance calculation for adaptive grid partitioning, quantifying the reliability of the variance results. A manual intervention interface is added to control the density of grid partitioning, with expert experience assisting in automatic partitioning. Data structure optimization for fine-grained grid partitioning reduces memory fragmentation, while batch operations are used for disk read / write in coarse-grained grid partitioning. The encoding scheme for unique grid cell identifiers supports spatial range queries, allowing for quick retrieval of spatial regions corresponding to identifier ranges. Real-time visualization of data point counts helps monitor system operation status, with dynamically updated statistical charts. Progress indicators are added to the region growth process of the density clustering algorithm, allowing for interrupted and resumed tasks running for extended periods. A multi-threshold strategy is used for global and local balance of density thresholds, with different thresholds applicable to different regions. Convergence checks are added to the calculation of cluster centroid coordinates, automatically stopping iterative calculations when the required accuracy is reached. The historical record function of the core grid cell supports trend analysis, and changes in the core cell reflect the evolution of data distribution.

[0070] Example 3: The process of generating the inspection path sequence begins with the extraction of historical inspection interval data corresponding to the core grid unit. The unique identifier of the core grid unit serves as the database query key. Historical inspection interval data is retrieved from the inspection record database, and a mapping relationship is established between the timestamp and the spatial coordinates of the core grid unit. The candidate interval set contains multiple consecutive time interval values, with the unit of the time interval values ​​uniformly set to minutes. Time series analysis applies a frequency statistics algorithm, which uses a weighted moving average method to process the frequency of interval values. The weight coefficients of the weighted moving average are allocated based on the near and far time, with higher weights for recent data and lower weights for older data. The interval value with the highest frequency is identified as the benchmark interval. The determination of the benchmark interval must meet the minimum support threshold. The equipment distribution topology is obtained digitally from the two-dimensional plan view of the substation. The geographical coordinates of the equipment nodes are converted to planar coordinates. The selection of the starting point comprehensively considers the equipment's operational health value and location accessibility. The order of the waypoints is determined using an improved nearest neighbor greedy algorithm. The output format of the inspection path sequence is an ordered list of equipment numbers.

[0071] Data preprocessing for the candidate interval set includes outlier removal and missing value imputation. Outlier identification uses the box plot rule, with values ​​exceeding 1.5 times the interquartile range considered outliers. Missing value imputation uses linear interpolation to ensure temporal continuity. The frequency statistics formula for time series analysis is expressed as:

[0072]

[0073] in: This represents the weighted frequency of the i-th interval value. Indicates the total number of time windows. This represents the weighting coefficient for the t-th time window. It is an indicator function when the interval value The value is 1 when it appears in the t-th window, and zero otherwise. The selection criteria for the baseline interval include the maximum weighted frequency and historical stability testing. Historical stability testing is performed by calculating the variance of the interval values; only intervals with a variance less than a set threshold are eligible as candidates. The graph model of the equipment distribution topology is constructed with equipment as nodes. The connections between nodes are determined based on electrical connection logic and physical paths. The starting point selection algorithm prioritizes devices with high operational health values ​​and proximity to entrances / exits. The optimization objective for the waypoint order is to minimize the total path length and balance the load across each segment.

[0074] The association between core grid cells and historical inspection interval data is achieved through spatiotemporal indexing. The spatial index uses grid cell identifiers, and the temporal index uses inspection record timestamps. The time interval values ​​of the candidate interval set are sorted and stored in an array structure. The weighted moving average window size of the frequency statistics algorithm is configurable, and the weight coefficient allocation uses an exponential decay model to ensure the dominance of recent data. The verification process of the baseline interval compares with historical data from the same period. Seasonal factors are handled through year alignment, and the edge weight calculation of the equipment distribution topology includes actual path length and turning difficulty factors. The multi-objective optimization for starting point selection uses Pareto front solving. A greedy algorithm for the order of waypoints selects the nearest unvisited point to the current point each time, while simultaneously checking path continuity constraints. The query statement for extracting historical inspection interval data includes time range filtering, with data from the most recent twelve months as the primary analysis object. The construction of the candidate interval set merges data corresponding to multiple core grid cells, and the merging operation uses union processing to avoid double counting. The frequency statistics results of time series analysis are visualized to help intuitively identify peaks. The final determination of the baseline interval includes a manual review process, with expert experience assisting in automatic selection. The digitization of the equipment distribution topology utilizes CAD drawing analysis algorithms, achieving an accuracy of 0.1 meters for equipment node coordinates. The scoring rules for starting point selection include equipment importance weights, which are set based on equipment type and voltage level. An upper limit is set on the number of iterations in the optimization algorithm for the pathpoint order to avoid indefinite searching.

[0075] The core grid cell data mapping table maintains the correspondence between cell identifiers and inspection record numbers, and the mapping table update mechanism supports incremental data addition. The candidate interval set data structure uses a dynamic array, with the array capacity automatically expanding with the amount of data. The weighted moving average coefficients for time series analysis are optimized through grid search, and the coefficient combination verification uses historical data backtesting. The application scope of the benchmark interval considers the differences in equipment types; different types of equipment can be set with different benchmark intervals. The connectivity check of the equipment distribution topology map uses a depth-first search algorithm, and disconnected subgraphs are handled specially. Multiple candidate points are generated for the alternative scheme of the starting point selection, and a backup point is activated when the main starting point is unreachable. The local optimization of the path point order uses a pairwise swapping method, and the swapping of adjacent point positions verifies path improvement. The storage format of historical inspection interval data includes three basic fields: timestamp, equipment number, and interval duration. Data integrity checks are implemented through checksums. Outlier handling of the candidate interval set records the reasons for removal, and linear interpolation for missing value imputation considers the time span factor. The frequency statistics parallel computing for time series analysis utilizes multi-threading technology, making the processing time for large-scale data controllable. The stability test of the baseline interval incorporates time series decomposition technology, and the stationarity of the residuals is assessed after separating the trend component from the seasonal component. The dynamic adjustment mechanism of the edge weights in the equipment distribution topology map supports real-time traffic updates, while the weight adjustment range is limited to avoid drastic fluctuations. The comprehensive evaluation function for starting point selection includes three dimensions: equipment operational health, environmental impact factor values, and equipment maintenance urgency, and is in the form of a linear weighted sum. The global search algorithm for the pathpoint order incorporates simulated annealing to avoid the local optimum trap of greedy algorithms.

[0076] After establishing data associations for core grid units, the association results are cached to improve the efficiency of repeated queries. Data cleaning rules for candidate interval sets are configurable, and cleaning parameters can be modified via configuration files. The confidence interval calculation for frequency statistics results in time series analysis uses the bootstrap method; the interval range helps assess the reliability of the baseline interval. The applicability check of the baseline interval compares it with actual inspection capabilities, ensuring the maximum interval does not exceed the daily work limit of the inspection team. The equipment distribution topology map is drawn using automated mapping tools, and the drawings are output at a standard scale. The decision support system for starting point selection integrates multi-source information, incorporating weather forecast data into the selection of entrances and exits. The waypoint sequence verification module checks whether the path covers all necessary equipment; any missing equipment is automatically inserted into the nearest path segment. A backup mechanism for historical inspection interval data ensures data security, with backup cycles synchronized with the business system. Version management of the candidate interval set records every change, and version rollback supports error recovery. The frequency statistics algorithm selection for time series analysis is based on data characteristics; highly periodic data undergoes Fourier transform preprocessing. The baseline interval adjustment process includes an application and approval stage, and adjustment records are archived for future reference. The equipment distribution topology map is updated based on the addition and removal of equipment, and the path is recalculated after changes to the map structure. Real-time optimization of starting point selection considers temporary equipment states, and faulty equipment is automatically excluded from candidate points. The graphical display of the path point order supports interactive adjustments; manually dragging equipment points allows for real-time path recalculation.

[0077] Data quality monitoring indicators for core grid units include mapping integrity and timeliness, with quality reports generated regularly. Data analysis reports for candidate interval sets include statistical descriptions and distribution visualizations, and reports are exported in PDF format. The parameter tuning interface for time series analysis provides slider controls, with real-time previews of parameter adjustment effects. Cross-station application of the baseline interval requires standardized processing, with data from different stations being normalized and compared. The storage of the device distribution topology map utilizes graph database technology, optimizing neighbor query efficiency. The machine learning model for starting point selection uses historical path records for training, and the model predicts starting point preferences. Performance analysis of the optimization algorithm for waypoint order includes time and space complexity assessments; approximate algorithms are used in big data scenarios. Access permissions for historical inspection interval data are managed hierarchically, and sensitive data is stored encrypted. The construction log of the candidate interval set records data sources and processing steps, and audit trails meet compliance requirements. Unit tests for the frequency statistics module of time series analysis cover boundary conditions, and abnormal inputs are handled correctly. The baseline interval push mechanism sends the final results to the path execution system, with message queues ensuring reliable transmission. The version comparison function of the device distribution topology map displays change details, and change impact assessment assists decision-making. The starting point selection feature supports multi-user collaboration and annotation, with annotations linked to specific devices. The exported waypoint sequence format is compatible with common navigation software, and the GPX standard format supports device import. The core grid cell data archiving strategy migrates historical data to cold storage; archived data is queryable but has high latency. Metadata management for candidate interval sets records data dictionaries and business meanings, with a metadata API providing service discovery. The frequency statistics database table design for time series analysis optimizes query performance, and indexing strategies are combined with query patterns. The baseline interval monitoring dashboard displays current values ​​and historical trends, with threshold alarms triggering color-coded alerts. Editing permissions for the device distribution topology map are assigned by role, and editing operations are traceable. The starting point selection simulation function predicts path execution time, and simulation results include time estimates and resource consumption. The waypoint sequence optimization algorithm supports adding constraints, and custom constraints can be extended using scripting languages.

[0078] Example 4: Smoothing the inspection path sequence using a path optimization algorithm requires establishing a directed graph model of the inspection path. Nodes in the directed graph model represent inspection points and have coordinate attributes, while edges represent inspection path segments and have weight attributes. Dynamic programming is used to find the shortest Hamiltonian path in the directed graph, which requires visiting each node exactly once and minimizing the total path length. Curve fitting uses cubic spline interpolation to make the discrete path points continuous, generating a smooth parametric curve. The fitted path needs to be checked to ensure it meets the mechanical motion constraints of the inspection equipment, including minimum turning radius and maximum acceleration limits. Path segments that do not meet the constraints are locally adjusted, including inserting auxiliary path points or modifying the positions of existing path points. The node attributes of the directed graph model are initialized using the equipment's geographical coordinates, and node numbers are bound to equipment identifiers. The edge weights of the directed graph model are calculated based on the Euclidean distance between nodes and the equipment priority weights, which are dynamically adjusted according to the equipment's operational health. The state transition equation of the dynamic programming algorithm records the set of visited nodes and the current node information. State compression techniques are used to handle large-scale node scenarios. The implementation of cubic spline interpolation requires determining boundary conditions and control points; natural boundary conditions are used. Verification of mechanical motion constraints is achieved by calculating path curvature and acceleration; curvature calculation uses differential geometry methods. The local adjustment algorithm iteratively optimizes the path point positions, and the adjusted path undergoes re-curve fitting and constraint checks.

[0079] The directed graph model is constructed using an adjacency matrix storage structure, with matrix elements storing edge weights. Node coordinate data is imported from a geographic information system, achieving millimeter-level accuracy. The calculation formula for the edge weight matrix combines distance and priority weights, with weight coefficients determined experimentally. The number of states in the dynamic programming algorithm grows exponentially with the number of nodes; large-scale problems are solved using heuristic algorithms for approximate solutions. Control point selection for cubic spline interpolation is based on path point sequences, with the number of control points matching the number of path points. Thresholds for mechanical motion constraints are set according to the performance of the inspection equipment, with different thresholds for different equipment types. The convergence condition of the local adjustment algorithm sets a maximum number of iterations to avoid infinite loops. The node attribute table of the directed graph model contains detailed equipment information, which is associated with path planning parameters. The dynamic update mechanism for edge weights supports real-time condition changes, triggering path recalculation upon weight updates. Memory usage optimization for the dynamic programming algorithm is achieved through state encoding, which compresses storage space. Curve smoothness evaluation for cubic spline interpolation uses a curvature continuity standard; refitting occurs when the rate of curvature change exceeds a threshold. The frequency of checking mechanical motion constraints is configurable, with high-frequency checks ensuring path safety. The step size of the local adjustment algorithm controls the adjustment range, with adaptive step sizes improving convergence speed. Visualization of the directed graph model helps verify model correctness, displaying the relationships between nodes and edges in a graphical interface. The solution progress of the dynamic programming algorithm is displayed in real time, with a progress bar indicating the remaining time. The parameter adjustment interface for cubic spline interpolation provides interactive control, with real-time preview of parameter modifications. Violation reports for mechanical motion constraints record the location and severity of violations in detail, and the reports can be exported as structured documents. The local adjustment algorithm's history function saves the adjustment trajectory, and trajectory playback supports process analysis.

[0080] The directed graph model's serialization storage supports model saving and loading, with a serialization format compatible with different platforms. The parallel computation of the dynamic programming algorithm utilizes multi-core processors, achieving a parallel speedup close to the ideal linear value. The numerical stability of the cubic spline interpolation method is controlled by the condition number; the algorithm automatically switches when the condition number is too large. Mechanical motion constraints are handled hierarchically, with hard constraints being mandatory and soft constraints providing optimization objectives. The local adjustment algorithm's multi-objective optimization balances path length and smoothness, with adjustable weight coefficients. Version management of the directed graph model records model change history, displaying version differences and changes. The dynamic programming algorithm's caching mechanism stores intermediate results, returning cached values ​​directly for repeated queries. Error analysis of the cubic spline interpolation method calculates fitting error, and error distribution statistics guide parameter adjustment. Simulation verification of mechanical motion constraints uses a physics engine, with simulation results visualized. The local adjustment algorithm's user intervention interface allows for manual adjustment, combining manual and automatic adjustments. The directed graph model's extensibility design supports dynamic node addition and deletion, with incremental updates avoiding full reconstruction. The dynamic programming algorithm's approximation algorithm provides a precision-efficiency tradeoff, allowing users to specify precision requirements. The real-time performance of cubic spline interpolation meets the requirements of online planning, with computational latency controlled within milliseconds. Adaptive adjustment of mechanical motion constraints is implemented based on equipment status, relaxing constraints when equipment performance degrades. The quality evaluation metrics for the local adjustment algorithm include multiple dimensions, with evaluation scores comprehensively judging the adjustment effect. Directed graph model access control restricts access to sensitive data, with role-based hierarchical management of permissions. Dynamic programming algorithms monitor and track memory and CPU usage, issuing alerts when resource limits are exceeded. Hardware acceleration of the cubic spline interpolation method utilizes GPU computation for large-scale data parallel processing. A knowledge base for mechanical motion constraints stores historical cases, recommending constraint parameters based on case matching. Automated testing of the local adjustment algorithm covers typical scenarios, with test cases updated regularly.

[0081] Referring to Table 1, the backup and recovery mechanism of the directed graph model ensures data security, and the backup strategy is executed regularly. The dynamic programming algorithm handles abnormal inputs with fault tolerance, and gracefully degrades in abnormal situations. The cubic spline interpolation method's cross-platform compatibility is verified against mainstream systems, with comprehensive compatibility testing coverage. The standardized interface for mechanical motion constraints supports third-party extensions, and the plug-in mechanism enhances flexibility. Performance analysis of the local adjustment algorithm identifies bottlenecks, and optimization measures target performance hotspots.

[0082] Table 1: Path Optimization Parameter Configuration Table

[0083]

[0084] The node data for the directed graph model originates from a device database, and the database connection parameters are configurable. The distance formula used for edge weight calculation employs high-precision spherical distance calculation, taking into account the Earth's curvature. The state transition table for the dynamic programming algorithm uses a hash table for storage, and chaining is used to resolve hash collisions. The matrix solution for the cubic spline interpolation method uses the LU decomposition algorithm, and its numerical stability has been rigorously verified. The mechanical motion constraints are checked in order of severity, with severe constraints checked first. The initial step size of the local adjustment algorithm is set proportionally to the path length, and the step size decay coefficient is dynamically adjusted. The verification process for the directed graph model includes node connectivity checks, and connected component analysis to ensure path reachability. The state space pruning strategy of the dynamic programming algorithm reduces invalid searches, and the pruning rules are based on the lower bound of the path length. The derivative calculation for the cubic spline interpolation method uses the central difference method, with the difference step size adaptively changing. The multi-objective optimization weight coefficients for mechanical motion constraints are determined through sensitivity analysis, and the coefficient adjustment step size is finely controlled. The restart mechanism of the local adjustment algorithm avoids local optima, and the restart condition is based on the improvement stagnation judgment.

[0085] Performance testing of the directed graph model uses a standard test dataset, recording response time and memory usage. The approximation ratio of the dynamic programming algorithm is provably guaranteed, and worst-case analysis provides performance boundaries. The smoothness metric of the cubic spline interpolation method is quantitatively evaluated, and the smoothness index guides parameter tuning. Spatial indexing is used for collision detection in mechanical motion constraints, accelerating collision detection. The parallel design of the local adjustment algorithm utilizes multi-threading technology, with thread synchronization ensuring data consistency. The maintenance tool for the directed graph model provides a graphical user interface, offering intuitive and easy operation. Debugging of the dynamic programming algorithm supports single-step tracking of state transitions and real-time monitoring of state values. Error visualization for the cubic spline interpolation method displays the error distribution, and error heatmaps pinpoint problem areas. An early warning mechanism for mechanical motion constraints detects potential violations in advance, with three levels of warning management. The version comparison function for the local adjustment algorithm displays differences before and after adjustments, with highlighted differences for easy confirmation. Importing and exporting the directed graph model supports multiple data formats, and format conversion maintains data integrity. The memory mapping technique of the dynamic programming algorithm handles extremely large state spaces, and disk I / O optimization reduces latency. The adaptive piecewise segmentation of the cubic spline interpolation method adjusts according to curvature changes, with curvature largely segmented. Relaxation variables for mechanical motion constraints handle boundary conditions and are incorporated into the optimization objective. For the robustness testing of the local adjustment algorithm, random perturbations are introduced, with the perturbation strength gradually increasing to test stability.

[0086] Example 5: The process of establishing a directed graph model begins with the accurate acquisition of the geographical coordinates of the equipment. Taking the inspection area of ​​an ultra-high voltage substation as an example, the main equipment in the substation includes five types: main transformers, circuit breakers, disconnect switches, current transformers, and voltage transformers, totaling thirty pieces of equipment constituting the inspection node set. The geographical coordinates of the equipment are obtained through total station measurements. The measurement points are located at the center point of the equipment foundation. The coordinate system adopts the independent coordinate system of the substation area, with the origin set at the center point of the substation control building. The X-axis points due north, and the Y-axis points due east. The coordinate units are accurate to millimeters. Euclidean distance calculation is based on the two-dimensional plane coordinates of the equipment. The formula for the straight-line distance between two equipment points A(x1,y1) and B(x2,y2) is a straight-line distance calculation. The actual path distance needs to consider the influence of obstacles between equipment. Obstacles include fixed facilities such as structural foundations, cable trenches, and fire pipelines. The actual path distance is corrected through the substation area path network model, which includes the centerlines of all walkable paths. The importance of equipment is quantified through a multi-dimensional evaluation system. Evaluation indicators include four dimensions: voltage level, load rate, fault history, and maintenance cost. Voltage levels are divided into three levels: 500kV, 220kV, and 110kV, each with corresponding weight coefficients. Load rate is weighted according to the ratio of actual operating capacity to rated capacity. Fault history statistics include the number of faults over the past three years, weighted by severity. Maintenance cost comprehensively considers both routine maintenance and fault repair costs. The final value of the equipment access priority weight is the normalized product of the weights of each dimension. Priority weight values ​​are set between 0.5 and 2.0, with higher-weighted equipment receiving priority access in path planning. The weight matrix of each edge in the directed graph is constructed using a distance-priority product model. The edge weight calculation formula is the product of the standardized Euclidean distance and the priority weight. Weight coefficients are used to adjust the relative influence of distance and importance.

[0087] The improved ant colony algorithm requires setting pheromone evaporation factor and heuristic factor parameters during the initialization phase. The pheromone evaporation factor controls the residual pheromone level on the path; a smaller evaporation factor value results in more residual pheromone, leading to faster convergence but a higher risk of getting trapped in local optima. Conversely, a larger evaporation factor value leads to faster pheromone evaporation, enhancing the algorithm's exploration capabilities but slowing down convergence. The heuristic factor influences the ants' preference for path length; a larger value indicates a greater tendency for ants to choose shorter paths, while a smaller value suggests stronger randomness in ant exploration. The ant colony size is set to 1.5 times the number of device nodes. Each ant constructs a complete path from a randomly selected starting node, using a roulette wheel selection rule where the selection probability is proportional to the product of the pheromone concentration and the heuristic factor. In practice, device coordinate data preprocessing includes coordinate offset correction and accuracy verification. Station control point coordinates are linked with device coordinates, and coordinate transformation parameters are calculated using least squares adjustment. The Euclidean distance calculation results were verified by comparing actual path lengths. Ten typical equipment paths were selected for on-site measurement, and the error between the measurement results and the calculation results was controlled within three per thousand. Equipment importance evaluation data were collected from the production management system. Voltage level weights were set as follows: 500kV equipment weight coefficient was 2.0, 220kV equipment weight coefficient was 1.5, and 110kV equipment weight coefficient was 1.0. Load rate weights were segmented as follows: below 50%, 1.0; 50%-80%, 1.2; and above 80%, 1.5. Fault history weights increased linearly according to the number of faults, and maintenance cost weights were segmented according to monetary value.

[0088] The node attribute table of the directed graph model contains three fields: device number, coordinate value, and priority weight. The edge weight matrix is ​​a symmetric matrix storing the weights between all node pairs. The improved ant colony algorithm's iterative optimization process sets a maximum of 500 iterations, with a convergence condition of no improvement to the optimal path after 50 consecutive iterations. The pheromone update rule adopts an elite ant strategy, where only the pheromone on the iteratively optimal path and the globally optimal path is enhanced, and the pheromone enhancement is inversely proportional to the path quality. Heuristic information is calculated based on the standardized distance between nodes; the smaller the distance, the greater the heuristic information value. The device geographic coordinate update mechanism considers device relocation; coordinate changes trigger the reconstruction of the directed graph model. Dynamic adjustment of priority weights supports real-time data updates; weight values ​​are automatically recalculated when device operating status changes. The parallelization of the ant colony algorithm utilizes multi-threading technology, with each thread managing the path search of one ant colony. The pheromone matrix is ​​stored in a sparse matrix format to reduce memory usage. A tabu list for path construction records visited nodes to avoid repeated visits.

[0089] In a specific application, the substation example includes two 500kV main transformers, six 220kV circuit breakers, twelve 110kV disconnect switches, five current transformers, and five voltage transformers. The main transformers are located in the central area of ​​the substation. The circuit breakers are arranged along the distribution equipment, the disconnect switches are distributed on both sides of the busbar, and the current and voltage transformers are installed close to their respective equipment. Priority weight calculation results show that the weight value of the 500kV main transformer is 2.0, the weight value of the 220kV circuit breaker is 1.8, the weight value of the 110kV disconnect switch is 1.2, the weight value of the current transformer is 1.3, and the weight value of the voltage transformer is 1.3. Euclidean distance calculation shows that the farthest equipment spacing is 320 meters, and the closest equipment spacing is 8 meters. The parameter optimization of the improved ant colony algorithm was completed through orthogonal experimental design. The optimal parameter combination is a pheromone evaporation factor of 0.3, a heuristic factor of 2.5, and a pheromone enhancement coefficient of 1.0. The path search process shows that the algorithm converged after two hundred iterations, with the optimal path length being 1.5 kilometers. The path sequence starts from the control building, passes through the main transformer, circuit breaker, disconnector, and instrument transformer, and finally returns. The pheromone distribution visualization shows that the pheromone concentration is high along the paths between the main equipment, forming obvious pheromone trajectories.

[0090] Quality control of equipment coordinate data includes repeated measurement verification, ensuring the maximum difference between three measurements of the same equipment does not exceed three millimeters. Priority weight normalization ensures the total weight equals the number of equipment, avoiding the impact of weight inflation. The ant colony algorithm's convergence curve records the optimal path length for each iteration, showing a rapid decline followed by smooth convergence. Path result verification is performed through simulated inspection processes, demonstrating that the path meets time constraints and operation sequence requirements. Extensibility testing of the directed graph model, with the addition of ten virtual devices, shows the algorithm can still solve the problem within a reasonable timeframe. Sensitivity analysis of priority weights involves observing path changes by altering weight values; increasing the weight of important equipment prioritizes access to these devices. Experiments on the influence of pheromone evaporation factors show that excessively small factors lead to premature convergence, while excessively large factors result in slow convergence. Heuristic factor adjustment experiments confirm that excessively large factors easily lead to local optima, while excessively small factors result in low efficiency for random search. Coordinate system transformation for equipment geographic coordinates supports multiple coordinate systems, including the Beijing 54 coordinate system, the Xi'an 80 coordinate system, and the WGS84 coordinate system. The customizable priority weight function allows users to adjust the weight calculation formula, and the formula editor supports mathematical expression input. The ant colony algorithm's progress monitoring displays the remaining iterations and the quality of the current optimal solution; users can terminate the algorithm midway. The exported path results include KML format for import into GIS systems and DXF format for processing in CAD software.

[0091] The specific implementation's anomaly handling mechanisms include interpolation methods for missing coordinate data, default value settings for priority weight calculation anomalies, and restart strategies when the ant colony algorithm fails to converge. Performance optimization measures include a caching mechanism for the directed graph model, pre-calculated storage of priority weights, and the ant colony algorithm's memory function to retain historical optimal solutions. The user interface provides a parameter setting panel, real-time display of path planning progress, and result visualization. The frequency of collecting equipment geographic coordinates is set to once a month to ensure the timeliness of coordinate data. The priority weight update cycle is synchronized with the equipment maintenance cycle, and the weights are recalculated after maintenance. The random number seed for the ant colony algorithm is configurable to ensure reproducible results. The path planning result evaluation report includes indicators such as path length, time estimation, and number of turns, and the report is automatically archived after generation. Detailed data for the substation example is recorded in the equipment information table, which includes five fields: equipment number, equipment type, X-coordinate, Y-coordinate, and priority weight. The improved ant colony algorithm's operation log records detailed data for each iteration, and the log file is used for subsequent analysis and optimization. The final path after path smoothing is saved as a path point sequence file, with a file format compatible with the inspection robot navigation system.

[0092] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for planning an inspection path of an ultra-high voltage substation, characterized in that, The method comprises the following steps: Collecting multi-dimensional monitoring data of an ultra-high voltage substation, the multi-dimensional monitoring data comprising equipment operation state data, environmental parameter data and equipment historical maintenance data; Performing dynamic evaluation on the multi-dimensional monitoring data to generate equipment operation health degree, environmental influence factor value and equipment maintenance urgency; Constructing a three-dimensional evaluation space based on the equipment operation health degree, environmental influence factor value and equipment maintenance urgency, and performing adaptive grid division in the three-dimensional evaluation space; Identifying an optimal inspection interval in the divided grid through a density clustering algorithm, and generating an inspection path sequence according to the optimal inspection interval; Performing smoothing processing on the inspection path sequence by using a path optimization algorithm, and outputting a final inspection path scheme; The method for identifying the optimal inspection interval in the divided grid through the density clustering algorithm comprises the following steps: Starting region growing from a grid unit with the largest data point density; Setting a density threshold, and merging adjacent grid units with a density reaching the threshold into a clustering cluster; Calculating the centroid coordinates of each clustering cluster, and determining a grid unit closest to all the centroids as a core grid unit; The method for generating the inspection path sequence comprises the following steps: Extracting historical inspection interval data corresponding to the core grid unit as a candidate interval set; Performing time series analysis on the candidate interval set, and taking the interval value with the highest occurrence frequency as a reference interval; According to the reference interval and the equipment distribution topological structure, determining a starting point and a passing point sequence of the inspection path. 2.The method of claim 1, wherein, The method for performing dynamic evaluation on the multi-dimensional monitoring data comprises the following steps: Processing the equipment operation state data by using a sliding window analysis method, calculating the variation coefficient of data characteristics in each time window as the equipment operation health degree; Analyzing the fluctuation law of the environmental parameter data by using an entropy method, and quantifying the environmental influence factor value; Processing the equipment historical maintenance data by using a time decay model, giving different weights according to the maintenance time, and calculating the equipment maintenance urgency. 3.The method of claim 2, wherein, The method for constructing the three-dimensional evaluation space comprises the following steps: Establishing a space coordinate system with the equipment operation health degree as an X-axis coordinate, the environmental influence factor value as a Y-axis coordinate and the equipment maintenance urgency as a Z-axis coordinate; Projecting historical monitoring data samples into the space coordinate system to form a data point cloud distribution; According to the density characteristics of the data point cloud, setting an adaptive space grid division granularity. 4.The method of claim 3, wherein, The adaptive grid division comprises the following steps: Calculating the distribution variance of the data point cloud on each coordinate axis, and determining the density of grid division according to the variance; In the data point dense area, fine-grained grid division is adopted, and in the data point sparse area, coarse-grained grid division is adopted; Assigning a unique identifier to each grid unit, and counting the number of data points contained in each grid unit. 5.The method of claim 1, wherein, The method for performing smoothing processing on the inspection path sequence by using the path optimization algorithm comprises the following steps: Establishing a directed graph model of the inspection path, wherein a node represents an inspection point, and an edge represents an inspection path segment; Using a dynamic programming algorithm to solve the shortest Hamilton path in the directed graph, and performing curve fitting on the obtained shortest path to eliminate sharp angle turns in the path; The method for performing curve fitting on the obtained shortest path to eliminate sharp angle turns in the path comprises the following steps: The discrete path points are continuously processed by using a cubic spline interpolation method; It is checked whether the fitted path meets the mechanical movement constraint condition of the inspection equipment; The path segment that does not meet the constraint condition is locally adjusted to ensure the executable of the path.

6. The ultra-high voltage substation inspection path planning method of claim 5, wherein, The establishment of the directed graph model comprises: The Euclidean distance between each inspection point is calculated according to the geographic position coordinates of the equipment; Different access priority weights are set for different equipment nodes considering the difference in the importance of the equipment, and the weight matrix of each edge in the directed graph is calculated in combination with the distance and the priority weight.

7. The ultra-high-voltage substation inspection path planning method of claim 6, wherein, The shortest Hamilton path in the directed graph is solved by using a dynamic programming algorithm, which comprises: An improved ant colony algorithm is used to search the path in the directed graph; The pheromone evaporation factor and the heuristic factor are set to balance the global search and local search capabilities; Through multiple iterations and optimization, the optimal path solution is converged.

8. An ultra-high voltage substation inspection path planning system comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, The processor, when executing the computer program, realizes the steps of the ultra-high voltage substation inspection path planning method in any one of claims 1 to 7.

Images