Electric power intelligent inspection method and system based on multi-dimensional sensor

Abstract

The invention relates to the technical field of electric power system monitoring, and particularly discloses an electric power intelligent inspection method and system based on a multi-dimensional sensor, and the method comprises the steps: obtaining multi-dimensional inspection data, actual environment data and equipment distribution data in an electric power inspection scene; obtaining a standardized inspection data set according to the multi-dimensional inspection data; according to the standardized patrol inspection data set, obtaining patrol inspection detection information of the power equipment, and according to the patrol inspection detection information of the power equipment, obtaining patrol inspection defect category information and patrol inspection identification information; acquiring inspection defect data according to the inspection defect category information and the inspection identification information; and generating a dynamic adjustment inspection path according to the actual environment data and the equipment distribution data, and generating a multi-dimensional inspection mapping matrix according to the dynamic adjustment inspection path and the inspection defect data. According to the method, the problems of data isomerism, detection lag, path solidification, information isolation and the like of traditional electric power inspection are effectively solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system monitoring, in particular to a power intelligent inspection method and system based on multi-dimensional sensors. BACKGROUND

[0002] As a key infrastructure of national economy, the safe and stable operation of power system directly affects the power supply reliability of power grid, and power inspection is the core means to timely find equipment defects and prevent faults. With the expansion of power grid scale and the diversification of equipment types (such as transformer substation main transformer, transmission line tower, wind and photovoltaic station photovoltaic array, etc.), the traditional manual inspection method gradually exposes the limitations of low efficiency, strong subjectivity, high risk of working in dangerous environment, etc., and it is difficult to meet the operation and maintenance needs of modern power grid. The power intelligent inspection technology based on multi-dimensional sensors emerges as the times require.

[0003] The existing power intelligent inspection technology based on multi-dimensional sensors still has many defects to be solved: the multi-source sensor data has serious heterogeneity, the timestamp deviation, spatial coordinate dislocation and format confusion problems of data collected by different sensors are prominent, such as the different time synchronization of visible light image and infrared temperature data leading to the inability to accurately associate the defect position, the large space registration error of laser radar point cloud and image, which directly affects the defect detection accuracy. SUMMARY

[0004] The purpose of the present application is to provide a power intelligent inspection method and system based on multi-dimensional sensors to solve the technical problems raised in the background.

[0005] To achieve the above purpose, the present application provides the following technical solutions: The power intelligent inspection method based on multi-dimensional sensors comprises: Obtaining multi-dimensional inspection data, actual environment data and equipment distribution data in a power inspection scene; Obtaining a standardized inspection data set according to the multi-dimensional inspection data; Obtaining power equipment inspection detection information according to the standardized inspection data set, and obtaining inspection defect category information and inspection recognition information according to the power equipment inspection detection information; Obtaining inspection defect data according to the inspection defect category information and the inspection recognition information; Generating a dynamically adjusted inspection path according to the actual environment data and the equipment distribution data, and generating a multi-dimensional inspection mapping matrix according to the dynamically adjusted inspection path and the inspection defect data; Generating an optimized inspection path according to the multi-dimensional inspection mapping matrix and the inspection defect data.

[0006] Preferably, the step of obtaining a standardized inspection data set according to the multi-dimensional inspection data comprises: According to the multi-dimensional inspection data, original collection records of various types of data are extracted, and multi-source heterogeneous original data is obtained according to the original collection records; According to the multi-source heterogeneous original data, a time reference calibration is performed to extract the collection time stamp of each original data, all time stamps are unified to the reference clock of the edge computing node by using the network time protocol (NTP), and the data segment with large time deviation is corrected by the interpolation method to obtain time axis aligned time series data; According to the time axis aligned time series data, spatial coordinate registration is obtained, and spatial fusion data is obtained according to the spatial coordinate registration; According to the spatial fusion data, noise and redundancy processing is performed to obtain denoising and simplification data; According to the denoising and simplification data, differential compression is implemented to obtain compressed data; According to the compressed data, format standardization conversion is performed to obtain a format unified standardized inspection data set.

[0007] Preferably, the step of obtaining power equipment inspection detection information according to the standardized inspection data set, and obtaining inspection defect category information and inspection recognition information according to the power equipment inspection detection information, comprises: According to the standardized inspection data set, equipment key features are extracted to obtain equipment multi-dimensional feature information; According to the equipment multi-dimensional feature information, image contour features, temperature abnormal area, three-dimensional structure features and discharge light spot features are obtained, the extracted image contour features, temperature abnormal area, three-dimensional structure features and discharge light spot features are spliced according to a preset dimension, the feature scale is unified through feature normalization processing, an input vector suitable for a defect detection model is formed, and model input data is obtained; According to the model input data, a lightweight defect detection model deployed on an edge node is called to output a preliminary detection result containing the location of the defect, the name of the component, and the characteristic parameters of the equipment, and power equipment inspection detection information is obtained; According to the power equipment inspection detection information, a preset power equipment defect type library is called, the defect feature parameters in the detection information are compared with the type library templates in terms of similarity, the type to which the defect belongs is determined according to the principle of the highest similarity, and inspection defect category information is obtained; According to the power equipment inspection detection information, the defect matching probability value output by the detection model is extracted, and the inspection recognition confidence identifier is obtained in combination with the operating environment of the equipment where the defect is located; The defect category information and the confidence identifier are associated with the inspection record of the corresponding equipment to obtain inspection recognition information.

[0008] Preferably, the step of obtaining inspection defect data according to the inspection defect category information and the inspection recognition information comprises: According to the inspection defect category information, a preset power equipment defect classification standard is obtained, and specific defect types such as insulator contamination and conductor fracture in the inspection defect category information are labeled with categories one by one according to the classification standard to form a defect type list with classification labels; According to the defect type list with classification labels, a trusted defect subset is obtained; According to the trusted defect subset, a power equipment account is retrieved, and through matching the defect occurrence position in the trusted defect subset with the equipment position in the account, the equipment number, the commissioning time and the interval to which the defect belongs corresponding to the defect are obtained, and a defect record associated with equipment information is obtained; According to the defect record associated with equipment information, a collection log of multi-dimensional sensor data is obtained, and through matching the equipment number in the defect record with the monitoring object in the collection log, the collection time and the sensor working mode when the defect occurs are extracted and supplemented into the defect record to obtain a defect detail containing collection background; According to the defect detail containing collection background, a structured defect item is obtained; According to the structured defect item, it is determined whether a key field is missing in each structured defect item, and for the item with a missing field, the inspection defect data is supplemented by backtracking the equipment account or the collection log.

[0009] Preferably, the step of generating a dynamic adjustment inspection path according to the actual environment data and the equipment distribution data, and generating a multi-dimensional inspection mapping matrix according to the dynamic adjustment inspection path and the inspection defect data, comprises: According to the actual environment data, environmental constraint information is extracted, and through the geographic marking tool, terrain obstacle information, electromagnetic interference strength and climate influence are filtered from the actual environment data based on the environmental constraint information to obtain an environmental constraint list; According to the environmental constraint list, the inspection passing area level is divided, and the recommended inspection speed and the equipment adaptation type of each level area are labeled based on the inspection passing area level to obtain a region map with passing level; According to the equipment distribution data, the equipment inspection priority is determined, and the equipment type information, the core degree information and the historical operation and maintenance record information are extracted from the equipment distribution data, and the equipment inspection priority list is generated according to the equipment type information, the core degree information and the historical operation and maintenance record information; According to the region map with passing level and the equipment inspection priority list, an initial inspection path is obtained; According to the initial inspection path and the real-time environment, a dynamic adjustment path is obtained, and a dynamic adjustment inspection path is obtained according to the dynamic adjustment path; According to the dynamic adjustment inspection path, path node information is obtained; According to the inspection defect data, obtain defect classification, defect corresponding equipment parts, and defect influence degree, and obtain defect feature dimension list according to the defect classification, defect corresponding equipment parts, and defect influence degree; According to the path node information and the defect feature dimension list, obtain a matrix basic framework; According to the matrix basic framework, generate filled matrix content, mark the inspection execution state information of each node based on the record information of each node corresponding equipment in the inspection defect data, and correct the structured record in the inspection defect data according to the inspection execution state information to obtain a multi-dimensional inspection mapping matrix.

[0010] Preferably, the step of generating an optimized inspection path according to the multi-dimensional inspection mapping matrix and the inspection defect data comprises: According to the multi-dimensional inspection mapping matrix, extract high defect risk path nodes, and generate a high defect risk node list according to the high defect risk path nodes; According to the high defect risk node list, determine an inspection priority sequence, and obtain the occurrence frequency and influence range of defects according to the inspection priority sequence, and obtain a priority sequence according to the occurrence frequency and influence range; According to the priority sequence, generate path redundancy information by dynamically adjusting the inspection path; According to the path redundancy information, obtain node access sequence, and generate a preliminary optimized path based on the node access sequence; According to the preliminary optimized path, generate an environment adaptation optimized path; According to the environment adaptation optimized path, obtain supplementary path nodes, obtain electrical association information between equipment according to the supplementary path nodes, and generate a complete optimized path based on the electrical association information; According to the complete optimized path, obtain total inspection time length, high priority node coverage rate, and low priority node reasonable omission rate, and generate an optimized inspection path based on the total inspection time length, the high priority node coverage rate, and the low priority node reasonable omission rate.

[0011] The application also provides a power intelligent inspection system based on a multi-dimensional sensor, characterized by comprising: A data acquisition module is configured to acquire multi-dimensional inspection data, actual environment data, and equipment distribution data in a power inspection scene; An inspection data set acquisition module is configured to acquire a standardized inspection data set according to the multi-dimensional inspection data; An inspection recognition information acquisition module is configured to acquire power equipment inspection detection information according to the standardized inspection data set, and acquire inspection defect category information and inspection recognition information according to the power equipment inspection detection information; The inspection defect data acquisition module is configured to acquire inspection defect data according to the inspection defect category information and the inspection identification information. The inspection mapping matrix generation module is configured to generate a dynamic adjustment inspection path according to the actual environment data and the equipment distribution data, and generate a multi-dimensional inspection mapping matrix according to the dynamic adjustment inspection path and the inspection defect data. The optimized inspection path acquisition module is configured to generate an optimized inspection path according to the multi-dimensional inspection mapping matrix and the inspection defect data.

[0012] Preferably, the inspection data set acquisition module comprises: The first acquisition unit is configured to extract original collection records of each type of data according to the multi-dimensional inspection data, and acquire multi-source heterogeneous original data according to the original collection records. The second acquisition unit is configured to extract collection time stamps of each original data according to the multi-source heterogeneous original data by performing time reference calibration, unify all the time stamps to a reference clock of an edge computing node by using a network time protocol (NTP), correct data segments with large time deviation by using an interpolation method, and obtain time axis aligned time series data. The third acquisition unit is configured to acquire spatial coordinate registration according to the time axis aligned time series data, and acquire spatial fusion data according to the spatial coordinate registration. The fourth acquisition unit is configured to obtain denoised and simplified data by performing noise and redundancy processing according to the spatial fusion data. The fifth acquisition unit is configured to implement differential compression to obtain compressed data according to the denoised and simplified data. The sixth acquisition unit is configured to obtain format unified standardized inspection data set by performing format standardization conversion according to the compressed data.

[0013] The application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the power intelligent inspection based on the multi-dimensional sensor when executing the computer program.

[0014] The application further provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of the power intelligent inspection based on the multi-dimensional sensor.

[0015] The beneficial effects of this application are as follows: This invention effectively solves the problems of heterogeneous data, detection lag, fixed paths, and isolated information in traditional power inspection, and has several significant advantages: First, relying on multi-dimensional sensors to collect data, and through time synchronization, spatial registration, and standardization processing, it eliminates the time, space, and format heterogeneity of multi-source data, avoiding misjudgments caused by traditional data fragmentation, and providing a high-quality data foundation for defect detection; Second, a lightweight deep learning model is deployed on edge nodes to achieve real-time and accurate identification of equipment defects, reducing the false positive rate, without relying on high-performance cloud hardware, meeting the real-time alarm requirements of inspection, and improving defect response efficiency; Third, it dynamically optimizes the inspection path based on the actual environment and defect data, prioritizing coverage of high-risk nodes, eliminating redundant sections, and supplementing the inspection of electrical related equipment, which shortens the total time and avoids the omission of hidden defects, improving the targeting and efficiency of inspection; Fourth, it constructs a multi-dimensional inspection mapping matrix to associate paths and defect information, achieving full-process traceability, supporting operation and maintenance decisions, and forming a "collection-processing-detection-optimization-decision" process. Closed-loop operation reduces reliance on manual labor, facilitates digital operation and maintenance of the power grid, ensures safe equipment operation, and reduces operation and maintenance costs. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a method flow according to an embodiment of this application.

[0017] Figure 2 This is a schematic diagram of the system structure according to an embodiment of this application.

[0018] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] like Figure 1 As shown, this application provides a power intelligent inspection method based on multi-dimensional sensors, including: S1. Acquire multi-dimensional inspection data, actual environmental data, and equipment distribution data in power inspection scenarios; S2. Obtain a standardized inspection data set based on the multi-dimensional inspection data; S3. Obtain power equipment inspection and testing information based on the standardized inspection data set, and obtain inspection defect category information and inspection identification information based on the power equipment inspection and testing information; S4. Obtain inspection defect data based on the inspection defect category information and inspection identification information; S5, generating a dynamic adjustment inspection path according to the actual environment data and the equipment distribution data, generating a multi-dimensional inspection mapping matrix according to the dynamic adjustment inspection path and the inspection defect data; S6, generating an optimized inspection path according to the multi-dimensional inspection mapping matrix and the inspection defect data.

[0021] As described in steps S1-S6 above, the present application systematically collects multi-dimensional inspection data in the power inspection scene, actual environment data and equipment distribution data, eliminates data heterogeneity through standardized processing, realizes accurate detection and type identification of power equipment defects relying on a lightweight deep learning algorithm, integrates structured inspection defect data, generates a dynamic adjustment inspection path combined with environmental constraints and equipment priority and constructs a multi-dimensional inspection mapping matrix, finally optimizes the inspection path, forms a complete intelligent inspection closed loop of "data collection-processing-analysis-decision", solves the problems of low efficiency, insufficient defect identification accuracy, poor path adaptability and lack of data support for operation and maintenance decision-making in traditional power inspection, realizes the intelligent, accurate and efficient power inspection, and guarantees the safe and stable operation of the power grid.

[0022] Three types of core data required for power inspection are collected to provide a basic support for subsequent intelligent analysis and decision-making. Power inspection needs to master the equipment running state, on-site environmental constraints and equipment spatial layout: the equipment running state determines the direction and focus of defect detection, the on-site environmental constraints affect the adaptability and path feasibility of the inspection equipment, and the equipment spatial layout determines the inspection coverage and priority, all of which are indispensable. Traditional power inspection relies on manual carrying of single equipment to collect local data, such as using an infrared thermal imager to collect temperature or a visible light camera to take pictures of the appearance, without systematically integrating environmental and equipment layout data, resulting in single data dimension and poor correlation, which is easy to cause defect misjudgment or unreasonable path planning, for example, without collecting electromagnetic interference data, the infrared sensor deviates in the high interference area, or without mastering the latest equipment layout, the inspection path misses the core equipment.

[0023] By eliminating the heterogeneity of multi-dimensional inspection data, a standardized data set with unified format and precision is formed, providing high-quality input for subsequent deep learning algorithms. Multi-dimensional inspection data comes from different types of sensors, with problems such as time asynchronization, spatial misplacement, format confusion, and data redundancy: for example, the timestamp deviation between visible light and infrared data is large, making it difficult to accurately correspond to the defect location; the spatial coordinates of laser radar point cloud and visible light image are misaligned, making it difficult to associate structural defects with appearance defects; different sensor data formats are heterogeneous (such as JPG for images, PCD for point clouds, and TXT for parameters), which need to be converted before being called by algorithms; the amount of original data is too large (such as high-definition video data), which can easily cause transmission congestion and storage pressure. If such data is used directly, it will seriously reduce the defect detection accuracy and algorithm running efficiency, so it is necessary to unify the data dimension, format, and precision through standardization processing.

[0024] From the standardized inspection data set, the equipment defect features are extracted, and the power equipment inspection detection information containing the defect location and feature parameters is generated, and then the defect type and recognition reliability are determined, providing clear basis for subsequent defect data sorting and operation and maintenance decision-making. Although the standardized data set has eliminated heterogeneity, it still needs to extract effective defect features, identify defect types, and evaluate reliability through intelligent algorithms: without recognizing defects, only mastering data cannot provide clear fault direction for operation and maintenance personnel; without evaluating reliability, only identifying defect types can easily cause waste of operation and maintenance resources due to "false positive" defects (such as misjudgment caused by electromagnetic interference). Traditional defect recognition methods rely on manual experience or complex deep learning models, which have low efficiency and strong subjectivity, complex models require high-performance hardware support and cannot be deployed on edge nodes, and only recognize single-dimensional data (such as only visible light images), which can easily miss complex defects (such as insulation sub-pollution accompanied by partial discharge), and lack a reliability evaluation mechanism, making it difficult to distinguish between real defects and misjudgment results.

[0025] The scattered inspection defect category information and inspection recognition information are integrated into inspection defect data containing device association information, collection background, and structured fields, making it have the properties of being statistical, traceable, and directly used for operation and maintenance decision-making. Inspection defect category information only specifies "defect type", and inspection recognition information only provides "reliability", both of which lack device basic information (such as device number and operation time), collection background (such as sensor model and collection parameters), and structured format, making it difficult to be directly used for device health assessment, defect frequency statistics, or historical tracing: for example, only knowing "joint overheating" but not knowing the corresponding device number and operation time, it is difficult to judge whether the defect is related to device aging; only knowing "reliable recognition" but not knowing the collection sensor parameters, it is difficult to investigate defects caused by sensor problems. Traditional defect records are mostly in free text form, with unfixed fields and incomplete information, which need to be manually sorted before use, resulting in low efficiency and easy errors.

[0026] The dynamic adjustment inspection path is generated by combining the actual environment constraints and the equipment priority to adapt to the field conditions, and a multi-dimensional inspection mapping matrix is constructed by associating the path with the defect data, so that the visual association of "path-equipment-defect" is realized, and an intuitive basis is provided for subsequent path optimization and operation and maintenance decision. The inspection path needs to meet two requirements of "covering all equipment to be inspected" and "adapting to the field environment": if only the equipment distribution is considered and the environment constraints (such as terrain obstacles and electromagnetic interference) are ignored, it is easy to cause the inspection equipment to be unable to reach the specified position or the data collection to be distorted; if only the environment is considered and the equipment priority is ignored, it is easy to miss the core equipment (such as the main transformer and the bus) inspection. The traditional path planning is mostly fixed route, which cannot dynamically respond to environmental changes (such as temporary construction fences), and the association between the path and the defect is not established, so the defect distribution rule cannot be traced back through the path, which is not conducive to subsequent path optimization.

[0027] Based on the path-defect association information in the multi-dimensional inspection mapping matrix and the defect distribution rule in the inspection defect data, the dynamic adjustment inspection path is optimized and adjusted to form an optimized inspection path covering high-risk equipment, avoiding redundant road sections, and adapting to environmental constraints, thereby improving the inspection efficiency and defect coverage. Although the dynamic adjustment inspection path can adapt to environmental changes, there may be problems such as "path redundancy" (such as low-priority nodes interspersed between high-priority nodes, resulting in detours) or "insufficient coverage of high-risk equipment" (such as not focusing on covering equipment with frequent defects). If the path has redundancy, the inspection time will be increased; if the high-risk equipment is not covered enough, key defects are easy to be missed. The traditional path optimization mostly relies on manual experience adjustment and lacks data support, and cannot be dynamically optimized according to the defect distribution, so it is difficult to balance the inspection efficiency and defect coverage.

[0028] In one embodiment, the step of obtaining a standardized inspection data set according to the multi-dimensional inspection data comprises: S201, extracting original collection records of each type of data according to the multi-dimensional inspection data, and obtaining multi-source heterogeneous original data according to the original collection records; S202, performing time reference calibration to extract collection time stamps of each original data, uniformly unifying all time stamps to a reference clock of an edge computing node by using a network time protocol (NTP), and correcting data sections with large time deviation by using an interpolation method to obtain time axis aligned time series data; S203, obtaining spatial coordinate registration according to the time axis aligned time series data, and obtaining spatial fusion data according to the spatial coordinate registration; S204, performing noise and redundancy processing on the spatial fusion data to obtain denoised and simplified data; S205, implementing differential compression on the denoised and simplified data to obtain compressed data; S206, performing format standardization conversion on the compressed data to obtain a set of format-unified standardized inspection data.

[0029] As described in steps S201-S206, the application classifies and extracts original collection records from multi-dimensional inspection data, clearly defines data sources and types, and provides traceable and classifiable basic data for subsequent processing. The multi-dimensional inspection data is derived from visible light sensors, infrared thermal imagers, ultraviolet sensors, laser radars, and state quantity sensors deployed in the previous steps, and has been temporarily stored in the edge computing node. In implementation, first, an independent storage directory is created in the edge computing node according to the sensor type, corresponding to the storage of original records of various sensors; then each record is extracted in ascending order of collection timestamp, and the sensor number, collection timestamp, and core parameter are embedded in the file name to ensure that the data source and collection conditions can be directly traced; finally, the extracted records are integrated to form multi-source heterogeneous original data, and the heterogeneity of the data is reflected in the format, dimension, and content. This step solves the problem of unknown source and confused type in traditional data processing, and ensures that the subsequent time synchronization, spatial registration, and other links can accurately associate data ownership through classification extraction and traceability labeling, laying a clear data foundation for the entire standardization process.

[0030] Eliminate the time heterogeneity of multi-source data and ensure the consistency of data timing. Due to the differences in sensor triggering mechanism and transmission delay, multi-source heterogeneous original data has timestamp deviation, which easily leads to the inability to associate multi-dimensional data of the same device. In implementation, first, the collection timestamp of each original data is extracted, and the visible light data with the highest collection frequency is used as a reference to analyze the time deviation and mark data segments with excessive deviation; then a high-precision reference clock is constructed through the Beidou time service module of the edge computing node, and an NTP server is deployed to unify the timestamps of all sensors to the reference, thereby preliminarily reducing the deviation; for the continuous deviation data segments that still exist, linear interpolation method is used for gradual correction to avoid timing confusion; finally, the data is arranged in ascending order of reference clock timestamp to form time axis aligned timing data. This step solves the problem of low time synchronization accuracy and uncorrected deviation in traditional time synchronization, and through the combination of NTP and interpolation method, the time deviation of multi-source data is controlled within a very small range, ensuring that the multi-dimensional state of the same device can be associated under the same timestamp, and providing time-consistent basic data for subsequent spatial registration.

[0031] Eliminate the spatial heterogeneity of data and realize the spatial correlation of multi-source data. Although the time axis alignment of time series data solves the time problem, the coordinate system of each sensor is independent, and the multi-dimensional data of the same part of the device cannot be correlated. In implementation, first, determine the WGS84 geodetic coordinate system commonly used in the power industry as the global reference, and call the WGS84 coordinate of the device to be inspected as the spatial reference point; for the wearable inspection device, combine the attitude parameters obtained by the built-in IMU module to convert the visible light and infrared data from the local coordinate system to the transition coordinate system, and then map it to the WGS84 coordinate system to realize pixel-level spatial correlation; for the unmanned aerial vehicle inspection system, convert the laser radar point cloud and ultraviolet data to the WGS84 coordinate system through the position data and IMU attitude data obtained by the GPS to determine the corresponding relationship between the discharge light spot and the device structure; for the fixed sensor, combine the device account and installation position offset to calculate its geodetic coordinate. Finally, integrate all the converted data to form spatial fusion data. This step solves the problem of low accuracy and single scene adaptation of traditional spatial registration, and through the scene-based registration scheme, the registration error is controlled in a very low range, which adapts to mobile and fixed inspection equipment and ensures that the multi-dimensional data of the same part of the device can be accurately correlated.

[0032] In order to eliminate the quality heterogeneity and volume redundancy of data and provide high-quality and lightweight data for subsequent processing. Although spatial fusion data solves the problem of space-time heterogeneity, there is noise due to environmental interference, and the amount of original data is still large, which affects the algorithm efficiency and transmission real-time. In implementation, different types of data are processed differently: for two-dimensional images such as visible light, infrared, and ultraviolet, Gaussian filtering, median filtering, and wavelet threshold denoising are used to remove noise, and adaptive downsampling is used to simplify the redundant background; for three-dimensional point cloud of laser radar, statistical filtering is used to remove outliers, and voxel grid is used to reduce point cloud density and retain device structure features; for one-dimensional parameters of state variables, sliding window smoothing is used to remove high-frequency fluctuations, and difference redundancy is used to simplify repeated parameters. Finally, denoising and simplification data is formed. This step solves the problem of traditional processing method generalization and separate processing, and through targeted processing, it greatly removes noise and redundancy on the premise of ensuring the complete retention of key features of device defects (such as insulator crack, joint overheating, and tower tilt), and lays a foundation for efficient data for subsequent compression and algorithm call.

[0033] Further optimize data volume, adapt edge storage and 5G transmission requirements, while retaining key defect information. Although the denoised and simplified data has been simplified, the data volume still cannot fully adapt to the limited storage of edge nodes and the fluctuating 5G bandwidth. When implementing, select an adaptive compression algorithm according to the characteristics of different data: for visible light video, use H.265 encoding, dynamically adjust the quantization parameter for device area and background area to balance the compression ratio and detail retention; for infrared temperature map, use wavelet transform compression, directly process the temperature value matrix to avoid temperature gradient loss; for laser radar point cloud, use octree structure compression combined with difference encoding to retain the spatial structure while simplifying the data volume; for state quantity parameters, use difference encoding combined with Huffman encoding to retain the parameter trend while reducing storage occupancy. Finally, form the compressed data. This step solves the problem of key information loss and imbalance between compression ratio and feature retention caused by traditional general compression algorithms. Through differentiated compression, the device defect key information is ensured to be complete while maximizing data volume reduction, adapting to the actual needs of edge computing and 5G transmission.

[0034] The final link of data standardization aims to eliminate format heterogeneity and achieve unified algorithm calling. Although the compressed data has been simplified, the format is still not unified and the metadata is stored in scattered form, increasing the complexity of algorithm calling. When implementing, use the encapsulation structure of "main JSON file + associated data file": the main JSON file defines six modules of data identification, collection information, sensor parameters, spatial information, data address, and verification information according to the specification, extracts data from the previous steps to fill in, and ensures field completeness and format specification; the associated data file stores the compressed original data and is associated through the "data address" field of the main JSON file; then, through the format verification module of the edge computing node, the main JSON file is verified for field integrity and format specification, and the associated data file is verified for integrity, and missing or non-standard data is supplemented and improved in the previous steps. Finally, the encapsulated data that passes the verification is stored in a standardized inspection data set, forming a format-unified standardized inspection data set. This step solves the problem of simple format conversion and non-standard fields in traditional format conversion. Through unified JSON encapsulation and verification, data format is unified and metadata is complete, ensuring that algorithms can be called through a unified interface, improving calling efficiency, while ensuring data traceability, providing standardized data input for subsequent defect detection and path optimization.

[0035] In one embodiment, the steps of obtaining power equipment inspection detection information from the standardized inspection data set, and obtaining inspection defect category information and inspection identification information from the power equipment inspection detection information, include: S301, extracting device key features from the standardized inspection data set to obtain device multi-dimensional feature information; S302, acquire image contour features, temperature abnormal area, three-dimensional structure features and discharge light spot features according to the device multi-dimensional feature information, splice the extracted image contour features, temperature abnormal area, three-dimensional structure features and discharge light spot features according to a preset dimension, unify the feature scale through feature normalization processing, form an input vector suitable for a defect detection model, obtain model input data; S303, call a lightweight defect detection model deployed on an edge node according to the model input data, output a preliminary detection result containing a device position where a defect is located, a component name and feature parameters of the defect, and obtain power equipment inspection detection information; S304, call a preset power equipment defect type library according to the power equipment inspection detection information, compare defect feature parameters in the detection information with templates of the type library in terms of similarity, determine a type to which a defect belongs according to a highest similarity principle, and obtain inspection defect category information; S305, extract a defect matching probability value output by the detection model according to the power equipment inspection detection information, and obtain an inspection recognition confidence index in combination with an operating environment of a device where the defect is located; S306, associate the defect category information and the confidence index to an inspection record of a corresponding device to obtain inspection recognition information.

[0036] As described in steps S301-S306 above, the application extracts key features of four dimensions of equipment appearance, temperature, structure and discharge from the standardized inspection data set with uniform format, providing high recognition and low redundancy feature input for subsequent defect detection models. The standardized inspection data set includes visible light images, infrared temperature maps, laser radar point clouds and ultraviolet discharge images, and has realized time alignment and space registration, which can ensure that the extracted features point to the same equipment component. In implementation, first, the four core data of appearance, temperature, structure and discharge are disassembled according to data types; for the appearance dimension, an edge detection algorithm is used to extract the equipment contour and describe the surface details through a texture quantization algorithm to capture appearance defect features such as insulator contamination and wire breakage; for the temperature dimension, a region growing algorithm is used to locate the temperature abnormal area, quantify the temperature extreme value, range and gradient of the abnormal area, and identify thermal defect features such as joint overheating and transformer oil temperature anomaly; for the structure dimension, a structure fitting algorithm is used to process the laser point cloud, calculate the structure deviation parameters such as tower tilt and conductor sag, and extract structure defect features; for the discharge dimension, a threshold segmentation algorithm is used to extract the ultraviolet discharge spot, quantify the discharge intensity, shape and position, and identify electrical defect features such as partial discharge. Finally, the four types of dimension features are spliced and normalized to form multi-dimensional feature information of the equipment. This step solves the problem of missing complex defects in traditional single-dimensional feature extraction by cross-dimension feature collaborative extraction and adaptation to power equipment defect characteristics, ensuring that the features cover all state dimensions of the equipment, and at the same time, through targeted algorithm quantization of defect features, the relevance of features and defects is improved, laying a foundation for subsequent model accurate detection.

[0037] The device multi-dimensional feature information is converted into an input tensor of the adaptive edge node lightweight defect detection model, solving the problems of mismatch between feature dimensions and model input and weak cross-dimensional feature correlation. In implementation, first, the image contour feature, temperature abnormal area, three-dimensional structure feature and discharge light spot feature are split from the device multi-dimensional feature information, and various features are associated to the same pixel plane (based on the spatial range parameter in the standardized data) through spatial coordinate mapping, ensuring that the temperature abnormal area, discharge light spot and other features are aligned with the spatial position of the image contour feature; then, the features are distributed according to the model input requirements, the image contour feature is taken as the basic feature channel, the temperature abnormal area and discharge light spot features are fused as the abnormal feature channel, and the three-dimensional structure feature and cross-dimensional association mark are taken as the association feature channel, forming a three-channel feature image; the bilinear interpolation method is used to uniformly scale the feature image to the fixed size required by the model, avoiding feature distortion caused by scaling; finally, the dynamic normalization processing is performed on the features of each channel, the feature distribution is adjusted by calculating the mean and standard deviation of the features in the channel, ensuring that the feature scales of each channel are consistent and meet the model training assumptions, and the extreme values are cut to ensure stable reasoning. Through this step, the effective conversion of multi-dimensional features to model input tensors is realized, the spatial correlation of cross-dimensional features is strengthened, the model can learn the composite features of device defects, and the normalization processing avoids the imbalance of model weights caused by the difference in feature scales, providing input data with strong adaptability for subsequent efficient reasoning.

[0038] The lightweight defect detection model deployed by the edge node converts the model input data into a defect detection result with clear physical meaning, meeting the real-time and detection accuracy requirements of power inspection. Before implementation, the basic lightweight model needs to be optimized for power equipment defect features: replace the model backbone network with a network structure with depth separable convolution to reduce parameter quantity and computing power requirements; increase the detection head adapted to small defect detection and adjust the anchor box size to match the typical size of power equipment defects; introduce the boundary box center point distance loss to optimize the loss function and improve defect positioning accuracy. The model training uses a special dataset containing multiple types of power equipment defects, and transfer learning is used to improve training efficiency and detection accuracy. After training, the model is converted into a format supported by the edge node and deployed on the inference engine. During implementation, the edge node inference scheduling module receives the model input data, calls the inference interface to pass the data into the model; the model extracts multi-scale features through the backbone network, strengthens the key defect features through the feature fusion network, and finally outputs the defect class probability, boundary box coordinates, quantitative feature parameters, and equipment component code by the detection head; the model output results are analyzed, the component name is converted through the preset component name coding dictionary, the quantitative parameters are restored to the actual value through inverse normalization, the metadata such as inference delay and model version are supplemented, and the power equipment inspection detection information containing the location of the defect, the component name, and the feature parameters of the equipment is formed. This step optimizes and deploys the model adapted to the power scene, ensures that the inference delay of the model on the edge node meets the real-time requirements, the detection accuracy can identify small defects, and the output information is complete, providing detailed detection basis for subsequent defect classification and credibility evaluation.

[0039] According to the defect management specification of the power industry, the standardization classification of defect types is realized through the similarity comparison of characteristic parameters, solving the problem of low efficiency and strong subjectivity of traditional manual classification. The preset defect type library of power equipment is constructed according to the industry specification, including four main classes of appearance damage, thermal fault, electrical discharge and structural deformation, and subdivided classes. Each sub-class is associated with priority, typical feature template and processing suggestion, and is stored in the local database of the edge node and updated regularly. In implementation, first, the defect characteristic parameters used for comparison are extracted from the power equipment inspection detection information, covering four dimensions of temperature, discharge, structure and appearance. The parameters are converted into the same format as the type library template, and the missing parameters are supplemented with default values by referring to the historical data of the same type equipment. Then, the weighted cosine similarity algorithm is used to calculate the similarity of the detection parameters and the templates in the type library. The different dimensions are assigned weights according to the influence of the parameters on the defect category, and the weights of temperature characteristics and discharge characteristics are higher than those of structure characteristics and appearance characteristics. The templates in the type library that match the main class of the detection information are traversed, and the template with the highest total similarity and meeting the threshold requirement is selected as the matching result. Finally, the defect category (main class-subclass-code), processing priority and processing suggestion corresponding to the template are integrated, and the device component name and location in the detection information are associated to form the inspection defect category information. Through the standardized type library and multi-dimensional weighted comparison, this step realizes the automation and standardization of defect classification, greatly improves the classification efficiency, and improves the classification accuracy of complex defects through multi-parameter collaborative comparison, providing clear category basis for the maintenance personnel to develop defect processing scheme.

[0040] The reliability of the defect detection result is dynamically evaluated combined with the equipment operating environment, false positive defects caused by environmental interference are filtered, and waste of maintenance resources is avoided. In implementation, first, the defect matching probability value output by the model is extracted from the power equipment inspection detection information. If there are multiple category prediction results, the standardization probability value is obtained by calculating the probability difference between the target category and the second highest category and weighted fusion, eliminating the probability misleading caused by multiple category competition. Then, the operating environment parameters of the equipment where the defect is located are called, including electromagnetic interference intensity, weather and lighting conditions, terrain and shielding conditions. The qualitative parameters are converted into quantitative scores, and the environmental interference comprehensive index is calculated according to the weight, quantifying the influence degree of the environment on the detection result. Based on the environmental interference comprehensive index, the confidence threshold is dynamically adjusted. The larger the environmental interference, the higher the threshold, ensuring that only high confidence results are determined as reliable in strong interference environment. Finally, the standardization probability value is compared with the dynamic threshold to divide into three levels of reliable identification, to-be-reviewed identification and unreliable identification. The reliability level, environmental interference comprehensive index, dynamic threshold and standardization probability value are integrated to form the inspection identification reliability identifier. Through the dynamic adaptation of model probability and environmental parameters, this step effectively reduces the false positive and missed judgment rate caused by traditional fixed threshold evaluation, making the reliability evaluation more suitable for the complex environment of power inspection, and guiding the artificial to prioritize processing high-value defects through the to-be-reviewed level, optimizing the efficiency of maintenance resource allocation.

[0041] Integrate defect detection, classification, and credibility assessment three types of information to form a complete information set for operation and maintenance decision-making, and solve the problem of low query efficiency and association error caused by traditional information fragmentation. When implementing, first, the corresponding inspection records of the equipment are called from the edge node inspection record database, including the basic information of the equipment (equipment number, operation time, and interval) and the historical inspection data (historical defect record, maintenance record), and the redundant fields are removed to form a structured record framework; then, the information association rules are formulated, the defect type, processing priority, and processing suggestion in the inspection defect category information are filled into the "current defect details" field of the record framework, and the credibility level, association parameter, and review suggestion in the inspection recognition credibility identification are filled into the "defect reliability description" field, to ensure that the information is consistent and logically coherent, and if there is inconsistency in the field, the previous spatial registration data is confirmed or marked for review; finally, the information is reorganized according to the operation and maintenance decision-making logic of "equipment basic information → current defect details → defect reliability description → historical inspection data → processing suggestion", and is packaged in JSON format, and the field naming uses general terms in the power industry, which is convenient for different roles to understand and call. The packaged inspection recognition information is stored in a special database, and the core information is pushed to the operation and maintenance personnel through a mobile terminal. Through the information association and structured integration based on the equipment inspection record, this step realizes "one-stop acquisition" of defect-related information, greatly improves the information acquisition efficiency of operation and maintenance personnel, and makes the decision-making more comprehensive by integrating historical inspection data. The general format also improves the cross-role collaboration efficiency, providing complete support for subsequent inspection defect data sorting and operation scheduling.

[0042] In one embodiment, the step of obtaining inspection defect data according to the inspection defect category information and the inspection recognition information comprises: S401, obtaining a preset power equipment defect classification standard according to the inspection defect category information, and labeling the specific defect types such as insulator contamination and conductor breakage in the inspection defect category information according to the classification standard to form a defect type list with classification labels; S402, obtaining a credible defect subset according to the defect type list with classification labels; S403, calling a power equipment account according to the credible defect subset, matching the defect occurrence position in the credible defect subset with the equipment position in the account to obtain the equipment number, operation time, and interval of the defect, and obtaining the defect record of the associated equipment information; S404, obtaining the collection log of multi-dimensional sensor data according to the defect record of the associated equipment information, matching the equipment number in the defect record with the monitoring object in the collection log to extract the collection time and sensor working mode when the defect occurs, and supplementing them to the defect record to obtain a defect detail containing collection background; S405, obtaining a structured defect item according to the defect details containing the collected background; S406, obtaining whether a key field is missing in each structured defect item according to the structured defect item, and supplementing the missing field item by tracing the equipment account or the collection log to obtain the inspection defect data.

[0043] As described in the above steps S401-S406, the present application realizes the standardization of defect classification, eliminates the differences in artificial expression, and establishes a "main class-subclass" hierarchical relationship in line with the specifications of the power industry. Although the inspection defect category information clearly indicates the specific type of defect, there are problems of subjective expression (such as the different expressions of "insulator contamination" and "insulator pollution") and missing classification hierarchy, which cannot adapt to the industry system docking and multi-dimensional statistical needs. The preset power equipment defect classification standard is based on industry specifications such as the "State Grid Corporation of China Power Equipment Defect Management Guide", and a three-level classification system of "one main class (appearance damage class, thermal fault class, electrical discharge class, structural deformation class)-two sub-classes-three sub-classes" is constructed. Each three sub-class is associated with a specification code, a defect description and a typical feature, stored in the edge node database and updated regularly to synchronize the industry specifications.

[0044] In implementation, first, the specific defect type is extracted from the inspection defect category information, and the unified expression of ambiguity is realized through the preset synonym mapping table (such as unified as "insulator pollution" to "insulator pollution"); then the classification labeling module of the edge node is called, and the unified specific defect type is matched with the three sub-classes in the classification standard, and the corresponding one main class, two sub-classes and specification code are automatically obtained after successful matching; finally, the specific defect type, classification hierarchy and specification code are integrated to form a defect type list with classification labels. If a new type of defect is encountered, it is labeled as "to be classified" and triggers manual review. The problems of low efficiency of traditional manual labeling and non-uniform terminology are solved, the classification accuracy is improved to meet the industry requirements, and the power industry defect management system can be directly connected, providing standardized basis for subsequent hierarchical defect frequency statistics and classification operation and maintenance strategy.

[0045] Screening real and valid defects, eliminating false positive defects caused by environmental interference or model misjudgment, and ensuring the quality of subsequent data. The defect type list with classification labels covers all detection results, including "credible", "to be reviewed" and "unreliable" three types of results. If low credibility defects are included in the subsequent process, it will lead to waste of operation and maintenance resources and distortion of data analysis.

[0046] In implementation, first, the defect type list with classification label is associated with the inspection identification information by "defect unique ID", and the classification label, credibility level, standardized probability value and environmental interference comprehensive index of the defect are integrated; then, multi-dimensional screening rules are developed: the core rule is to retain defects with "credibility level" as "credible", the supplementary rule is that defects with low environmental interference and high standardized probability value in "to be reviewed" can be upgraded to "credible", and "unreliable" defects are excluded regardless of other parameters; during the screening process, logs are recorded synchronously, and the exclusion reason (such as "to be reviewed and high environmental interference") is noted, and the log retention time meets the industry traceability requirements; finally, the key information (defect unique ID, classification label, credibility level) of the retained defects is integrated to form a credible defect subset and stored in a special database, and indexed according to the key field for subsequent calling. Through multi-dimensional automatic screening, the false positive defect rejection rate is greatly improved, invalid operation and maintenance costs are avoided, and at the same time, a high-quality data source is provided for subsequent device information association, ensuring the effectiveness and availability of defect data.

[0047] The "defect-equipment" association link is established to solve the problem that the credible defect subset only contains defect types without device attributes, and to provide support for analyzing the correlation between defects and device aging and developing interval-level operation and maintenance plans. Knowing only the defect type and credibility cannot locate the device to which the defect belongs and the location of the power grid topology. As the core carrier of device basic information, the power device account records the device number, location, commissioning time, and interval to which the device belongs, and is the key basis for realizing the association between defects and devices.

[0048] In implementation, first, the structured power device account is retrieved from the device distribution data of S1 step, and the device basic information, spatial location information and operation attribute information are extracted to establish an index to optimize query efficiency; then, the defect occurrence location (including text description and spatial coordinates) is extracted from the credible defect subset, and a two-step matching method of "text description priority matching + coordinate accurate verification" is adopted: in the first step, the defect location text description is matched with the account device location, and in the second step, for defects with fuzzy text description, the matching result is determined by calculating the distance between the defect coordinates and the coordinates of the same type device; after successful matching, the device number, commissioning time, and interval to which the device belongs are extracted, and are integrated with the classification label, credibility and other fields of the credible defect subset to form a defect record with associated device information; if the matching fails, it is marked as "device to be confirmed" and pushed to manual supplement. Through two-dimensional matching, the accuracy of device association is improved, and the association error caused by fuzzy location description is avoided, and at the same time, the automatic matching greatly improves the efficiency, providing key data in the device dimension for subsequent analysis of the correlation between device aging and defects and development of accurate operation and maintenance plans.

[0049] In the acquisition conditions of the defect occurrence, the complete traceability link of "defect-equipment-acquisition condition" is established, which provides basis for the traceability of subsequent defect detection results and sensor performance optimization. Although the defect record associated with equipment information clearly associates the defect with the equipment, it does not include the acquisition background (such as sensor parameters, acquisition time). If the subsequent defect detection result is questionable, it cannot be investigated whether it is caused by improper acquisition conditions.

[0050] In implementation, first, the acquisition log of multi-dimensional sensor data is called, which is automatically generated when the sensor acquires data in step S1, contains structured fields such as acquisition identification, sensor parameters, acquisition environment, and is stored in the edge node acquisition log database; then through the "equipment number" in the defect record associated with equipment information, the log in the acquisition log that "monitoring object = equipment number" and the acquisition time is consistent with the defect occurrence time is matched; after successful matching, the acquisition time, sensor model, working mode (such as infrared temperature measurement range, visible light resolution), calibration state and other key acquisition background information are extracted and supplemented to the defect record associated with equipment information; if no matching log is found, mark it as "acquisition background to be supplemented", retrieve the backup log from the sensor local cache, and still cannot obtain it. If the acquisition background is missing, record it. Through automatic matching and supplementing of acquisition background, a complete traceability link is established, the traceability of defect data is improved, and the sensor calibration state and working mode in the acquisition background can be used to check the data quality, which provides basis for subsequent sensor performance optimization and defect detection precision improvement.

[0051] The format of the defect data is standardized, the order of the defect detail fields containing acquisition background is chaotic, the name is different, the algorithm calling is difficult, the cross-system compatibility is poor, and other problems are solved, which provides standardized data for subsequent links that can be algorithmically called, batch statistics, and cross-system compatible. Although the defect information integrated in the early stage is complete, the format is unstructured, which cannot directly adapt to the input requirements of subsequent path optimization, mapping matrix construction and other algorithms, and it is also difficult to interface with other systems in the power industry.

[0052] In implementation, firstly, the principle of "full-dimension coverage, clear semantics, and unified format" is followed, 15 core fields covering five dimensions of "defect identification, defect classification, device information, collection background, and credibility" are designed, and the name, type, and format requirements of each field are specified (for example, the format of "defect occurrence time" is standard datetime type); then, a one-to-one mapping rule of "defect detail field → structured item field" is formulated, and the defect detail information containing the collection background is automatically filled into the corresponding field through the structured encapsulation module of the edge node; JSON format is used for encapsulation, which has the characteristics of strong readability, good scalability, and cross-platform compatibility, and the order of the encapsulated item field is fixed and the semantics is clear; finally, the structured defect item is stored in a special database, which supports quick query by any field. Through standardized field design and JSON encapsulation, the algorithm calling efficiency, statistical analysis capability, and cross-system compatibility of defect data are greatly improved, and the clear field name facilitates understanding by different roles, reducing system maintenance cost.

[0053] The final link to ensure the integrity and uniqueness of the inspection defect data ensures that the data can be directly used in subsequent core links (dynamic path adjustment, mapping matrix construction, etc.). Although the structured defect item is standardized in format, there may be missing key fields (such as device number, defect occurrence time) or data duplication (the same defect is collected by multiple sensors) due to previous collection or matching abnormalities. These problems will cause the data to be unusable or statistical analysis distorted.

[0054] In implementation, firstly, 8 key fields (including defect identification class, device association class, defect classification class, credibility class, and collection background class) are defined, and all structured defect items are traversed by the data verification module of the edge node to check whether they contain key fields and mark items with missing fields; for items with missing fields, they are supplemented according to the field source backtracking to the corresponding data source (for example, missing "device number" backtracking to device account, missing "sensor calibration state" backtracking to collection log), and if the supplement fails, it is pushed to manual verification; then, "multi-dimensional feature matching" is used to identify duplicate items (same device number, similar defect occurrence time, and consistent defect type), and the optimal item is retained according to the rule of "credibility first → collection time first → sensor calibration state first", and the duplicate data is removed; finally, the structured defect items that have passed the verification, are complete, and have been de-duplicated are integrated to form the final inspection defect data, which is stored in the edge node and cloud database simultaneously.

[0055] In one embodiment, the step of generating a dynamic adjustment inspection path according to the actual environment data and device distribution data, and generating a multi-dimensional inspection mapping matrix according to the dynamic adjustment inspection path and the inspection defect data, comprises: S501, extracting environment constraint information according to actual environment data, filtering terrain obstacle information, electromagnetic interference strength, and climate influence from the actual environment data based on the environment constraint information, and obtaining an environment constraint list by marking with a geographic marking tool; S502, dividing a patrol passing region level according to the environment constraint list, and marking a recommended patrol speed and a device adaptation type of each level region based on the patrol passing region level to obtain a region map with a passing level; S503, determining a device patrol priority according to device distribution data, extracting device type information, core degree information, and historical operation and maintenance record information from the device distribution data, and generating a device patrol priority list according to the device type information, the core degree information, and the historical operation and maintenance record information; S504, obtaining an initial patrol path according to the region map with the passing level and the device patrol priority list; S505, obtaining a dynamic adjustment path according to the initial patrol path and real-time environment, and obtaining a dynamic adjustment patrol path according to the dynamic adjustment path; S506, obtaining path node information according to the dynamic adjustment patrol path; S507, obtaining defect classification, defect corresponding device components, and defect influence degree according to patrol defect data, and obtaining a defect feature dimension list according to the defect classification, the defect corresponding device components, and the defect influence degree; S508, obtaining a matrix basic framework according to the path node information and the defect feature dimension list; S509, generating filled matrix content according to the matrix basic framework, obtaining record information of each node corresponding device in the patrol defect data based on the filled matrix content for each path node in the matrix row dimension, marking patrol execution state information of each node based on the record information, and obtaining a multi-dimensional patrol mapping matrix by correcting the patrol execution state information according to the structured record in the patrol defect data.

[0056] As described in steps S501-S509, the present application extracts and visually marks key constraints affecting patrol from actual environment data, providing clear environmental restriction basis for subsequent path planning. The actual environment data is derived from pre-acquired environment sensor data, GIS map data, and real-time feedback information, including three core constraint elements of terrain, electromagnetic, and climate. If these constraints are ignored, the planned path may not be able to pass through the patrol device due to terrain obstacles, or the sensor data may be distorted due to electromagnetic interference, affecting the accuracy of defect detection.

[0057] In implementation, first, the actual environment data is disassembled, and the terrain obstacle information such as steep slope, water accumulation area and equipment intensive area is screened from the terrain sensor and GIS map, the high interference area is screened from the electromagnetic interference sensor, and the climate influence information such as strong wind and heavy rain is screened from the climate sensor and real-time weather warning; then, the ArcGIS and other geographic marking tools are used to mark the boundaries and properties of various constraint areas according to the constraint types, such as marking the type and adaptive equipment of terrain obstacles, the intensity and influence sensor type of electromagnetic interference, and the level and safety suggestion of climate influence with specific colors; finally, the marked information is sorted to form an environment constraint list, which clearly defines the geographical range, constraint description and adaptive suggestion of each constraint. The problem of incomplete extraction and non-visual marking of traditional environment constraints is solved, and through multi-source data integration and geographic marking, the constraints cover all dimensions of terrain, electromagnetic and climate, are accurately positioned and intuitive, providing reliable environmental basis for subsequent division of passing area level and planning of feasible path.

[0058] The environment constraints are converted into quantified passing difficulty levels, and the equipment adaptation and inspection speed suggestions of each level area are clearly defined, solving the problems of equipment and area mismatch and unreasonable speed in path planning. The area passing difficulty under different environment constraints is significantly different, such as the inspection conditions of flat and non-interference area and steep slope and high interference area, which need to be clearly defined through grading to avoid inspection failure due to equipment mismatch or task timeout due to improper speed.

[0059] In implementation, according to the constraint intensity in the environment constraint list, four-level passing level standards are developed: the first level area is a low difficulty area without significant constraints, the second level is a medium-low difficulty area with slight constraints, the third level is a medium-high difficulty area with obvious constraints, and the fourth level is a high difficulty or prohibited passing area with serious constraints, and the "one vote veto system" is used to ensure accurate division of high-level constraint areas; then, combined with the power inspection specifications, the recommended inspection speed for each level area is marked, such as fast inspection for first level area and slow inspection for third level area, and the adaptive inspection equipment type is clearly defined, such as wearable device and fixed sensor for first level area and unmanned aerial vehicle and tracked robot for third level area; finally, different colors are used to distinguish each level area on the GIS map, and the device adaptation and speed suggestion text annotations are superimposed to form a regional map with passing levels.

[0060] Through quantitative grading and accurate adaptive marking, the problems of subjectivity and lack of adaptive suggestions in traditional passing area division are solved, making the passing difficulty objective and controllable, and the equipment and speed adaptation reasonable, providing clear environmental adaptation basis for subsequent initial path planning.

[0061] Based on the multi-dimensional information of equipment distribution data, an objective and unified equipment inspection priority is formulated to ensure that the path planning prioritizes covering core high-risk equipment. In the power system, the functions, importance, and defect risks of equipment are different. For example, the failure of core equipment such as main transformers has a wide impact, and high-defect-frequency equipment has high risk. Therefore, priority sequencing is needed to ensure that resources are tilted towards key equipment, avoiding excessive inspection of ordinary equipment or missing critical equipment.

[0062] In implementation, first, extract key information from equipment distribution data: obtain equipment type and installation location from equipment account, obtain equipment core degree (core / important / ordinary) from power grid dispatching system, and obtain historical operation and maintenance records (defect frequency, serious defect proportion) from defect management system; then develop standardized sequencing rules, calculate equipment comprehensive score according to core degree (weight 50%), historical defects (weight 30%), and equipment type (weight 20%), and divide equipment into first level (high priority), second level (medium priority), and third level (low priority) according to the score; finally, organize equipment information and priority level to form an equipment inspection priority list, mark the number, name, core degree, historical defect situation, and priority of each equipment, and index according to priority and installation location for subsequent calling. This solves the problem of subjective and non-uniform determination of traditional equipment priority, ensures objective and uniform priority through multi-dimensional weight sequencing, accurately covers key equipment, and provides clear basis for subsequent path planning to prioritize covering core high-risk equipment, improving inspection resource utilization efficiency.

[0063] Combined with environmental adaptability and equipment priority, an initial inspection path is generated that takes into account "passable" and "key coverage", laying the foundation for subsequent dynamic adjustment. The initial path needs to balance environmental constraints and equipment priority, avoiding impassable areas to ensure that the inspected equipment is reachable, and prioritizing covering first-level priority equipment to ensure that core high-risk equipment is not missed, avoiding the contradiction between feasibility and coverage caused by traditional path isolated planning.

[0064] In implementation, first, determine the planning parameters and constraints: determine the inspection starting point (such as the main control building of the substation, the take-off and landing point of the unmanned aerial vehicle), the combination of inspection equipment (adapt to the level of the region), and the termination condition (cover all first-level equipment, and the coverage ratio of second-level and third-level equipment meets the standard); then use the greedy algorithm for path planning, the core logic is "start from the starting point, preferentially select the closest uncovered high-priority equipment, select the lowest passable level path, and repeat until the termination condition is met", avoid four-level prohibited areas in the process, and set up surrounding observation points for equipment located in high-constraint areas; finally, verify the environmental adaptability (only first-level to third-level areas), priority coverage (first-level equipment full coverage), and time compliance (total time ≤ preset time) of the path, and form the initial inspection path after targeted adjustment and optimization, marking the path nodes, adaptive equipment, and estimated time.

[0065] A real-time environmental feedback and path adjustment mechanism is established to enable the path to respond to dynamic environmental changes during the inspection process and avoid inspection stagnation or safety risks caused by sudden environmental changes. The initial path is planned based on static environment, but temporary construction, sudden electromagnetic interference, strong wind and other dynamic changes may occur during actual inspection, and the path needs to be adjusted in time to ensure the continuity and safety of the inspection.

[0066] In implementation, first, a real-time environmental feedback mechanism is established: through the camera, IMU, and electromagnetic sensor of the inspection equipment (wearable device, unmanned aerial vehicle) to monitor obstacles and environmental parameter abnormalities in real time, combined with the data of the remote environmental monitoring system (fixed camera, weather station), the edge node analyzes in real time and triggers the path adjustment alarm; then standardized adjustment rules are developed, and node replacement (replacing obstacle area equipment with high priority equipment), sequential adjustment (changing the node access order to avoid obstacle sections), and path segment bypass (selecting low-level area bypass) are performed according to priority to ensure that the adjusted path still covers high-priority equipment and the time consumption is controllable; finally, the candidate adjustment path is generated and the optimal scheme is evaluated, which is synchronized to the inspection equipment and personnel terminal, and the path comparison and reasons before and after adjustment are recorded to form a dynamic adjustment inspection path. The traditional path adjustment lag and subjectivity problem is solved, and through real-time feedback and standardized rules, the rapid response to environmental changes is realized, the adjusted path still meets the key coverage and environmental adaptation requirements, and the inspection interruption rate is greatly reduced, ensuring the continuity of the task.

[0067] The standardized path node information is extracted from the dynamically adjusted inspection path to provide a structured row dimension basis for subsequent construction of a multi-dimensional inspection mapping matrix. The dynamically adjusted inspection path exists in the form of "path segment", and the specific attributes of each node (such as corresponding equipment, planned time, and adaptive equipment) are not explicitly defined, while the matrix construction needs to associate defect data with nodes, and the node attributes need to be explicitly defined through structured extraction to ensure that the information is callable and traceable.

[0068] In implementation, first, the dynamic adjustment inspection path (JSON format) is parsed, and the core information of each node is extracted in node order: the corresponding device number, name and location coordinates are obtained from the path node list, the adaptive inspection device is determined according to the area traffic level, the plan access period is allocated according to the total time consumption, and whether it is an adjustment node and the original node information are marked according to the adjustment record; then the standardized node field is designed, including node ID, corresponding device number / name / location, planned access period, adaptive device, whether it is an adjustment node, and adjustment association information, to ensure that the field is complete and the format is uniform; finally, the node information is integrated to form a path node information list, arranged in node order, stored in a special database and indexed, to facilitate subsequent matrix construction and calling. This step solves the problem of fragmentation and non-uniform format of traditional path node information, and through structured parsing and standardized field design, ensures that the node information is complete and callable, provides a clear basis for the row dimension of the matrix, and the node adjustment association information ensures traceability, facilitating subsequent analysis of path adjustment reasons. The standardized defect feature dimension is extracted from the inspection defect data, providing a structured column dimension basis for the multi-dimensional inspection mapping matrix. The matrix needs to associate path nodes and defect data through defect feature dimension to clearly determine whether a certain node device has a certain type of defect, which part of the device the defect occurs in, and how much impact it has, avoiding the confusion of the matrix information caused by incomplete or redundant traditional defect feature extraction.

[0069] In implementation, first, the core features are extracted from the inspection defect data (structured defect entries): the first-level main class, second-level sub-class, specific defect type and specification code are obtained from the defect classification field, the specific part of the device corresponding to the defect is obtained from the device part field, and the defect emergency level (minor, general, or serious) is obtained from the impact degree field; then the features are de-duplicated and integrated: duplicate features (such as ambiguous part names) are removed, and the features are integrated according to the "defect classification -> impact degree -> device part" hierarchy to avoid cross-dimension redundancy (such as not repeating the listing of defect classification implied parts); finally, the feature information is sorted to form a defect feature dimension list, labeling the ID, classification hierarchy, corresponding part, and impact degree of each feature, and arranging them in logical order for matrix column dimension calling. This solves the problem of incomplete and redundant traditional defect feature extraction, and through structured extraction and de-duplication integration, ensures that the defect feature dimension covers all dimensions of classification, part, and impact degree, is non-redundant and logically clear, providing an accurate basis for the matrix column dimension, and ensuring the effective association of subsequent paths and defects. The multi-dimensional inspection mapping matrix framework is constructed based on the association of "path node-defect feature-supplementary information", providing a structured carrier for subsequent content filling. The matrix framework needs to clearly define the row (path node) and column (defect feature) dimensions, while supplementing key information such as device status and execution, to avoid the single dimension and information fragmentation of traditional matrices, and to ensure that the matrix can not only associate paths and defects, but also reflect inspection progress and device status, supporting operation and maintenance decisions.

[0070] In implementation, first, the matrix dimension is determined: the row dimension is the path node information list (arranged in the order of inspection), each row marks the node ID, corresponding device number and name; the column dimension is the defect feature dimension list (arranged in the order of classification-impact degree-component), each column marks the feature ID, classification level, corresponding component and impact degree; then a supplementary column is designed, including the real-time running state of the device (running / hot standby, etc., from the state quantity sensor), the inspection execution state (completed / to be executed / interrupted, marking the time and reason), the defect treatment suggestion (associated with industry specifications), the risk level (combined with defect impact and device core degree); finally, the Python Pandas library is used to build the matrix framework, initialize the core column and the initial value of the supplementary column, set the format for easy viewing and calling, and store it in multiple formats for manual and algorithm use.

[0071] According to the matrix basic framework, the filling matrix content is generated, based on the filling matrix content, each path node in the matrix row dimension is obtained, and the record information of each node corresponding device in the inspection defect data is obtained, the inspection execution state information of each node is marked based on the record information, and the multi-dimensional inspection mapping matrix is obtained by correcting the structured record in the inspection defect data according to the inspection execution state information This step is the final forming link of the multi-dimensional inspection mapping matrix, the core is to automatically fill in the defect association content and real-time correct the inspection execution state, to ensure that the matrix information is accurate, real-time, and can directly support operation and maintenance decision. The matrix needs to be filled with content to clearly show “whether there is a defect in the device of a certain path node, and how the defect features are”, and the real inspection progress is reflected through state correction, avoiding the low efficiency of traditional manual filling and the poor usability of the matrix caused by state lag.

[0072] In implementation, first, the matrix core column is automatically filled in: each path node is traversed, the inspection defect data (structured defect items) is associated through the device number, the defect is marked “defect” in the corresponding feature column and supplemented with defect ID and occurrence time if the defect is queried, and “no defect” is marked if the defect is not queried, and the Python script is combined with SQL query to realize automation, and the accuracy is verified by manual sampling after filling; then the inspection execution state is real-time corrected: according to the inspection device feedback (executing, completed, interrupted) and the scheduling system record, the execution state column is updated, the completion time or interruption reason is marked, and the device running state column is synchronized and refreshed (every 10 seconds); finally, the structured record of the inspection defect data is compared to correct the possible information deviation (such as defect feature matching error) in the matrix, to ensure that the matrix content is consistent with the actual defect and execution situation, and the final multi-dimensional inspection mapping matrix is formed.

[0073] In one embodiment, the step of generating an optimized inspection path according to the multi-dimensional inspection mapping matrix and the inspection defect data comprises: S601, extract high-defect-risk path nodes according to the multi-dimensional inspection mapping matrix, and generate a high-defect-risk node list according to the high-defect-risk path nodes; S602, determine an inspection priority sequence according to the high-defect-risk node list, and obtain the occurrence frequency and influence range of defects according to the inspection priority sequence, and obtain the priority sequence according to the occurrence frequency and influence range; S603, generate path redundancy information according to the priority sequence and dynamically adjust the inspection path; S604, obtain a node access sequence according to the path redundancy information, and generate a preliminary optimized path based on the node access sequence; S605, generate an environment adaptation optimized path according to the preliminary optimized path; S606, obtain supplementary path nodes according to the environment adaptation optimized path, obtain electrical association information between devices according to the supplementary path nodes, and generate a complete optimized path based on the electrical association information; S607, obtain the total inspection time length, the high-priority node coverage rate, and the low-priority node reasonable omission rate according to the complete optimized path, and generate an optimized inspection path based on the total inspection time length, the high-priority node coverage rate, and the low-priority node reasonable omission rate.

[0074] As described in steps S601-S607, the present application accurately identifies and structures high-defect-risk path nodes from a multi-dimensional inspection mapping matrix, providing clear targeting for subsequent path optimization focusing on high-risk devices. The multi-dimensional inspection mapping matrix has associated path nodes with defect characteristics and device states, and contains node information about current serious defects and historically high-frequency defects. If such nodes are not covered in priority, they are likely to cause device failure due to defect expansion. Traditional methods rely on artificial subjective memory to identify high-risk nodes, which has the problems of low identification accuracy and no standardized record, leading to omission or misjudgment of high-risk devices.

[0075] In implementation, first, the multi-dimensional inspection mapping matrix is parsed, focusing on key columns such as defect impact degree, risk level, and defect frequency, and screening nodes with "serious impact", "high risk", and high-frequency defects; then, hierarchical rules are developed, and high-risk nodes are divided into first, second, and third levels according to the priority of "serious defect + high risk level > serious defect > high risk level + high frequency > high frequency"; finally, a structured list field is designed, including risk node ID, corresponding path node ID, device number, device name, risk level, risk reason, defect details, and device location coordinates, to ensure complete and traceable information. The list is stored in a special database and indexed for subsequent priority ordering and calling. Through matrix quantization extraction and structured list, the traditional subjective identification problem is solved, the identification accuracy of high-risk nodes is significantly improved, and accurate basis is provided for subsequent priority coverage of core risk devices, avoiding operation and maintenance risks caused by subjective omission.

[0076] Based on the high defect risk node list, combined with the defect occurrence frequency and the influence range, an objective priority sequence of the full quantity inspection node is constructed, solving the problem of subjectivity of traditional priority sorting and not integrating full quantity nodes. Identifying only high risk nodes cannot determine the inspection order between nodes, and does not cover non-high risk nodes, which can easily lead to confusion in the order of high risk nodes or interference of ordinary nodes with high risk inspection, affecting the operation and maintenance efficiency and risk response speed.

[0077] In implementation, first, the defect occurrence frequency (defect times in the last 3 months / 6 months) and the influence range (the size of the power grid area affected by equipment failure) are extracted from the inspection defect data, and are converted into quantitative scores. The higher the frequency and the wider the influence range, the higher the score. Second, the comprehensive score is calculated according to “frequency score (weight 60%) + influence range score (weight 40%)” for high risk nodes, and the nodes are ranked in descending order of score to form the internal sequence of first-level, second-level and third-level high risk nodes. Then, the non-high risk nodes are sorted according to “influence range score + frequency score”, and finally the high risk and non-high risk node sequences are integrated to form the full quantity inspection priority sequence of “first-level high risk→second-level high risk→third-level high risk→ordinary node”. The ordinary nodes adjacent to the high risk nodes are merged in order to reduce the subsequent redundancy. Through quantitative scoring and full quantity integration, the priority sequence is objective and unified, the consistency of understanding of the sequence by different operation and maintenance personnel is significantly improved, and the rationality of the high risk node sorting is enhanced, providing a clear basis for subsequent path redundancy identification and sequence optimization, and avoiding path confusion caused by subjective sorting.

[0078] By comparing the inspection priority sequence with the dynamically adjusted inspection path, the interlaced redundant nodes and detour redundant sections in the path are identified to provide accurate targets for subsequent redundancy elimination. Although dynamic adjustment of the inspection path adapts to the real-time environment, it may introduce low priority nodes interlaced with high risk nodes, detours without obstacles, and other redundancies, resulting in increased inspection time and delayed high risk nodes. Traditional methods rely on manual comparison and identification of redundancies, which is inefficient and prone to misjudgment of necessary detours and redundancies due to subjective judgment.

[0079] In implementation, first, the node sequence and section information of the dynamically adjusted inspection path are extracted, and the node ID, device number, section distance, and traffic level are determined; then, sequence comparison tools and geographic coordinate analysis are used to identify redundancies in two steps: first, sequence consistency comparison is performed to check whether the continuous high-risk nodes in the priority sequence are inserted by low-priority nodes in the dynamic path. If there are non-adjacent merged low-priority nodes, it is determined as an insertion redundancy; second, distance rationality comparison is performed to calculate the straight-line distance between adjacent high-risk nodes in the priority sequence and the actual section distance of the dynamic path. If the actual distance is much longer than the straight-line distance and there is no environmental obstacle, it is determined as a detour redundancy; finally, the redundancy node and section information are integrated, and the redundancy type, location, time consumption, and reason are labeled for visualization on the area map with traffic levels, facilitating intuitive understanding.

[0080] Through automatic comparison and quantitative determination, the efficiency and accuracy of redundancy identification are greatly improved, avoiding the inefficiency and subjective misjudgment of traditional manual identification. The clear redundancy information provides support for subsequent path optimization to target and remove redundancies, ensuring that the optimized path focuses on high-risk nodes without invalid detours.

[0081] Based on the path redundancy information, invalid redundancies are removed and the node access sequence is reconstructed to generate a preliminary optimized path without redundancy and reasonable sequence, solving the problem of subjective adjustment and easy addition of redundancies in traditional adjustment. Identifying redundancies alone cannot achieve path optimization. It is necessary to combine node location and priority sequence to reconstruct the sequence to ensure that high-risk nodes are continuously covered after removing redundancies and the path is the shortest, avoiding node sequence disorder or new detours caused by blind deletion.

[0082] In implementation, first, redundancies are classified and processed: non-adjacent ordinary nodes inserted between high-risk nodes are removed, and adjacent merged nodes along the path are retained; detour redundancy sections are corrected, and paths with short distances and low traffic levels are preferentially selected; then, based on the inspection priority sequence, the shortest path algorithm is used to reconstruct the node access sequence. The node is regarded as a graph node, and the section distance and traffic level are combined as edge weights to calculate the shortest path between adjacent nodes, ensuring that high-risk nodes are continuously covered according to priority; finally, the path redundancy removal and priority coverage are verified. If there are still redundancies or priority reversals, backtracking adjustment is performed to form a preliminary optimized path, and the node sequence, section information, adaptive device, and estimated time consumption are labeled. Through accurate redundancy removal and algorithmic reconstruction of the sequence, the total distance and total time of the path are significantly shortened compared to the dynamically adjusted path, high-risk nodes are continuously covered according to priority, and the subjective chaos of traditional adjustment is avoided, laying a foundation for efficient path for subsequent environmental adaptation and associated node supplementation.

[0083] Ensure that the preliminary optimization path adapts to the real-time environment and device capabilities, solving the problem of "theoretical optimization but actual unfeasibility". The preliminary optimization path may prioritize short paths, ignore real-time environmental obstacles, or mismatch devices and traffic levels (such as wearable devices unable to pass through a steep third-level slope), resulting in inspection stagnation. Traditional methods often adjust after discovering adaptation problems during execution, which is lagging and inefficient, and can delay inspection tasks.

[0084] During implementation, first verify the environmental adaptability of the preliminary optimization path: check if the road segment traffic level matches the adaptive device, and confirm whether there are obstacles such as construction or sudden electromagnetic interference based on real-time environmental feedback; then adjust according to the priority of "device replacement → road segment detour → node sequence fine-tuning": replace the adaptive device first (such as replacing a third-level road segment with a drone), if replacement is not possible, choose a detour in the same or lower level traffic area, and fine-tune the node sequence if the detour is too long; verify device adaptability, obstacle avoidance, and total time after adjustment, ensuring that the total time increment does not exceed 20% of the preliminary optimization path, forming an environmentally adapted optimization path, and synchronizing it to the inspection device terminal and labeling the adaptation notes. By verifying in advance and adjusting with multiple strategies, we avoid the inspection stagnation caused by traditional adaptation lag, significantly improve device adaptation accuracy and obstacle avoidance rate, and control the total time after adjustment, ensuring that the path can be implemented on the ground, enhancing the flexibility and reliability of on-site inspection.

[0085] Supplement the electrical association nodes of defective devices in the environmentally adapted optimization path, avoiding the omission of implicit defects caused by isolated coverage of defective devices. There are direct or indirect electrical associations between power equipment, such as main transformer defects that may affect incoming line breakers, and busbar defects that may be associated with multiple interval devices. Covering only defective devices can miss the implicit faults of associated devices, and traditional methods often arrange separate associated inspection paths, which are inefficient and time-consuming.

[0086] During implementation, first extract nodes with defects from the environmentally adapted optimization path as supplementary path nodes, as these nodes have devices with explicit defects, and their associated devices need to be checked; then obtain the electrical association information of the supplementary path nodes from the power grid topology database, distinguish between direct association (strong influence) and indirect association (weak influence), and prioritize direct association devices; finally, supplement the associated nodes to the adjacent position of the original defective node to avoid separate round trips, and if the total time exceeds the budget after supplementation, prioritize the retention of close-range, strongly-influenced associated nodes; verify the coverage of associated devices and the total time to form a complete optimization path, and label the association type and inspection focus. By accurately identifying associated devices and supplementing them in sequence, the accuracy of associated device identification and the discovery rate of implicit defects are significantly improved, the total time increment is controllable, and the inefficiency of traditional separate inspection is avoided, ensuring that inspection covers explicit defects and implicit risks, and reducing the probability of power grid failure.

[0087] The final verification and confirmation link of the path optimization ensures that the optimization effect meets the standard by quantifying key indicators, solving the problem of single and subjective traditional verification indicators. The complete optimization path needs to meet the core goals of "shortening the time, covering high-risk, and reasonably omitting low-risk", and the optimization cannot be ensured to be effective without verification, which may result in invalid optimization of time overrun or high-risk omission.

[0088] In implementation, first, three key indicators are defined: total inspection time (the sum of link time consumption and node stay time consumption), high-priority node coverage rate, and low-priority node reasonable omission rate, and the standard is formulated; then, the indicator values are calculated based on link distance, passing speed, and node type, and the verification is performed in the order of "total time → high-priority coverage rate → low-priority omission rate"; if the indicators do not meet the standard, targeted adjustments are made (such as deleting distant associated nodes and supplementing close high-priority nodes); after the standard is met, the path information, indicator results, and optimization comparison are integrated to form the final optimized inspection path, which is output in text and electronic formats and synchronized to the edge node and the cloud platform.

[0089] As shown in Figure 2 The application also provides a power intelligent inspection system based on a multi-dimensional sensor, which comprises: A data acquisition module 1 is configured to acquire multi-dimensional inspection data, actual environment data, and equipment distribution data in a power inspection scene. An inspection data set acquisition module 2 is configured to acquire standardized inspection data sets according to the multi-dimensional inspection data. An inspection identification information acquisition module 3 is configured to acquire power equipment inspection detection information according to the standardized inspection data sets, and acquire inspection defect category information and inspection identification information according to the power equipment inspection detection information. An inspection defect data acquisition module 4 is configured to acquire inspection defect data according to the inspection defect category information and the inspection identification information. An inspection mapping matrix generation module 5 is configured to generate a dynamically adjusted inspection path according to the actual environment data and the equipment distribution data, and generate a multi-dimensional inspection mapping matrix according to the dynamically adjusted inspection path and the inspection defect data. An optimized inspection path acquisition module 6 is configured to generate an optimized inspection path according to the multi-dimensional inspection mapping matrix and the inspection defect data.

[0090] In one embodiment, the inspection data set acquisition module 2 comprises: A first acquisition unit is configured to extract original collection records of each type of data according to multi-dimensional inspection data, and acquire multi-source heterogeneous original data according to the original collection records. The second acquisition unit is configured to extract collection time stamps of each original data according to time reference calibration of multi-source heterogeneous original data, unify all time stamps to a reference clock of an edge computing node by using a network time protocol (NTP), correct a data segment with a larger time deviation by using an interpolation method, and obtain time axis aligned time series data. The third acquisition unit is configured to acquire spatial coordinate registration according to the time axis aligned time series data, and acquire spatial fusion data according to the spatial coordinate registration. The fourth acquisition unit is configured to perform noise and redundancy processing on the spatial fusion data to obtain denoised and simplified data. The fifth acquisition unit is configured to implement differential compression on the denoised and simplified data to obtain compressed data. The sixth acquisition unit is configured to perform format standardization conversion on the compressed data to obtain a format unified standardized inspection data set.

[0091] The application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the power intelligent inspection based on the multi-dimensional sensor when executing the computer program.

[0092] The application further provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the power intelligent inspection based on the multi-dimensional sensor when being executed by a processor.

[0093] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, value library or other medium provided by the present application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0094] It should be noted that in this document, the terms "comprising", "including", or any other variant thereof are intended to cover a non-exclusive inclusion, so that a process, device, article or method that includes a list of elements not only includes those elements, but also includes other elements not expressly listed or inherent to such process, device, article or method. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, device, article or method that includes the element.

[0095] The above only describes the preferred embodiments of the present application, and does not limit the patent scope of the present application. Any equivalent results or equivalent process transformations obtained by using the content of the present application specification and drawings, or directly or indirectly applied to other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A power intelligent inspection method based on multi-dimensional sensors, characterized in that, include: Acquire multi-dimensional inspection data, actual environmental data, and equipment distribution data in power inspection scenarios; A standardized inspection data set is obtained based on the multi-dimensional inspection data. Based on the standardized inspection data set, power equipment inspection and testing information is obtained, and inspection defect category information and inspection identification information are obtained based on the power equipment inspection and testing information. Based on the inspection defect category information and inspection identification information, obtain inspection defect data; A dynamically adjusted inspection path is generated based on the actual environmental data and equipment distribution data, and a multi-dimensional inspection mapping matrix is ​​generated based on the dynamically adjusted inspection path and inspection defect data. An optimized inspection path is generated based on the multidimensional inspection mapping matrix and inspection defect data.

2. The power intelligent inspection method based on multi-dimensional sensors according to claim 1, characterized in that, The step of obtaining a standardized inspection data set based on the multidimensional inspection data includes: The original collection records of each type of data are extracted from the multidimensional inspection data, and multi-source heterogeneous raw data are obtained based on the original collection records. Based on the multi-source heterogeneous raw data, time base calibration is performed to extract the acquisition timestamps of each raw data. The Network Time Protocol (NTP) is used to unify all timestamps to the reference clock of the edge computing node. Data segments with large time deviations are corrected by interpolation to obtain time-axis aligned time-series data. Spatial coordinate registration is obtained based on time-series data aligned with the time axis, and spatial fusion data is obtained based on the spatial coordinate registration. Noise and redundancy are processed based on spatial fusion data to obtain denoised and simplified data. Compressed data is obtained by performing differentiated compression on the denoised and simplified data; The compressed data is then converted to a standardized format to obtain a unified set of inspection data.

3. The power intelligent inspection method based on multi-dimensional sensors according to claim 1, characterized in that, The steps of obtaining power equipment inspection and testing information based on the standardized inspection data set, and obtaining inspection defect category information and inspection identification information based on the power equipment inspection and testing information, include: Key features of equipment are extracted from standardized inspection data sets to obtain multi-dimensional feature information of the equipment; Based on the multi-dimensional feature information of the equipment, image contour features, temperature anomaly areas, three-dimensional structural features and discharge spot features are obtained. The extracted image contour features, temperature anomaly areas, three-dimensional structural features and discharge spot features are then spliced ​​together according to preset dimensions. The feature scale is unified through feature normalization processing to form an input vector that adapts to the defect detection model, thus obtaining the model input data. The model input data is used to call the lightweight defect detection model deployed on the edge node, and the output is a preliminary detection result containing the location of the defective equipment, the name of the component, and the characteristic parameters, thus obtaining the power equipment inspection and detection information. Based on the inspection and testing information of power equipment, the preset power equipment defect type library is retrieved, the defect feature parameters in the testing information are compared with the type library template, and the type of defect is determined according to the principle of the highest similarity, thus obtaining the inspection defect category information; Based on the inspection and testing information of power equipment, the defect matching probability value output by the detection model is extracted, and the inspection and identification credibility label is obtained by combining the operating environment of the equipment where the defect is located. By associating defect category information with credibility identifiers to the inspection records of the corresponding equipment, inspection identification information is obtained.

4. The power intelligent inspection method based on multi-dimensional sensors according to claim 1, characterized in that, The step of obtaining inspection defect data based on the inspection defect category information and inspection identification information includes: Based on the inspection defect category information, the preset power equipment defect classification standard is obtained. The specific defect types in the inspection defect category information, such as insulator contamination and conductor breakage, are labeled with their respective categories according to the classification standard, forming a list of defect types with classification labels. Obtain a subset of credible defects from a list of defect types with categorization labels; The power equipment ledger is retrieved based on the trusted defect subset. By matching the location of the defect in the trusted defect subset with the location of the equipment in the ledger, the equipment number, commissioning time and interval corresponding to the defect are obtained, and the defect record with associated equipment information is obtained. Based on the defect records of associated equipment information, the acquisition logs of multidimensional sensor data are obtained. By matching the equipment number in the defect record with the monitoring object in the acquisition log, the acquisition time and sensor working mode when the defect occurred are extracted and added to the defect record to obtain a defect detail including the acquisition background. Obtain structured defect entries based on the defect details including the collection background; Based on the structured defect entries, determine whether each structured defect entry is missing a key field. For entries with missing fields, supplement the inspection defect data by tracing back the equipment ledger or collecting logs.

5. The power intelligent inspection method based on multi-dimensional sensors according to claim 1, characterized in that, The steps of generating dynamically adjusted inspection paths based on the actual environmental data and equipment distribution data, and generating multi-dimensional inspection mapping matrices based on the dynamically adjusted inspection paths and inspection defect data, include: Environmental constraint information is extracted from actual environmental data. Based on the environmental constraint information, terrain obstacle information, electromagnetic interference intensity, and climate impact are filtered from the actual environmental data, and a list of environmental constraints is obtained by marking them with a geotagging tool. Based on the environmental constraint list, the inspection access area is divided into levels, and the recommended inspection speed and equipment compatibility type for each level are marked, resulting in an area map with access levels. The equipment inspection priority is determined based on the equipment distribution data, and equipment type information, core level information, and historical operation and maintenance record information are extracted from the equipment distribution data. An equipment inspection priority list is then generated based on the equipment type information, core level information, and historical operation and maintenance record information. The initial inspection path is obtained based on the area map with access levels and the equipment inspection priority list; A dynamically adjusted path is obtained based on the initial inspection path and the real-time environment, and a dynamically adjusted inspection path is obtained based on the dynamically adjusted path. Obtain path node information based on dynamically adjusted inspection paths; Based on the inspection defect data, obtain the defect classification, the corresponding equipment component, and the degree of defect impact, and obtain a list of defect feature dimensions based on the defect classification, the corresponding equipment component, and the degree of defect impact; The matrix basic framework is obtained based on the path node information and the list of defect feature dimensions; Based on the matrix basic framework, a filling matrix is ​​generated. Based on the filling matrix content, each path node in the matrix row dimension is processed, and the record information of the corresponding device in the inspection defect data is obtained. The inspection execution status information of each node is marked based on the record information. Based on the inspection execution status information, the structured records in the inspection defect data are compared and corrected to obtain a multi-dimensional inspection mapping matrix.

6. The power intelligent inspection method based on multi-dimensional sensors according to claim 1, characterized in that, The step of generating an optimized inspection path based on the multidimensional inspection mapping matrix and inspection defect data includes: High-defect-risk path nodes are extracted based on the multi-dimensional inspection mapping matrix, and a list of high-defect-risk nodes is generated based on the high-defect-risk path nodes. The inspection priority sequence is determined based on the list of high-defect-risk nodes, and the occurrence frequency and impact range of defects are obtained based on the inspection priority sequence. The priority sequence is then obtained based on the occurrence frequency and impact range. Path redundancy information is generated based on priority sequence and dynamic adjustment of inspection path; The node access order is obtained based on the path redundancy information, and a preliminary optimized path is generated based on the node access order. An environment-adapted optimization path is generated based on the initial optimized path; Supplementary path nodes are obtained based on the environment adaptation optimization path, and electrical association information between devices is obtained based on the supplementary path nodes. A complete optimization path is generated based on the electrical association information. The total inspection time, high-priority node coverage, and reasonable omission rate of low-priority nodes are obtained from the complete optimized path. An optimized inspection path is then generated based on the total inspection time, high-priority node coverage, and reasonable omission rate of low-priority nodes.

7. A power intelligent inspection system based on multi-dimensional sensors, characterized in that, include: The data acquisition module is used to acquire multi-dimensional inspection data, actual environmental data, and equipment distribution data in power inspection scenarios. The inspection data set acquisition module is used to acquire a standardized inspection data set based on the multidimensional inspection data. The inspection identification information acquisition module is used to acquire power equipment inspection and testing information based on the standardized inspection data set, and to acquire inspection defect category information and inspection identification information based on the power equipment inspection and testing information. The inspection defect data acquisition module is used to acquire inspection defect data based on the inspection defect category information and inspection identification information. The inspection mapping matrix generation module is used to generate a dynamically adjusted inspection path based on the actual environmental data and equipment distribution data, and to generate a multi-dimensional inspection mapping matrix based on the dynamically adjusted inspection path and inspection defect data. The optimized inspection path acquisition module is used to generate an optimized inspection path based on the multi-dimensional inspection mapping matrix and inspection defect data.

8. The power intelligent inspection system based on multi-dimensional sensors according to claim 7, characterized in that, The inspection data set acquisition module includes: The first acquisition unit is used to extract the original collection records of various types of data based on the multidimensional inspection data, and to acquire multi-source heterogeneous original data based on the original collection records. The second acquisition unit is used to perform time base calibration based on multi-source heterogeneous raw data to extract the acquisition timestamps of each raw data, and uses the Network Time Protocol (NTP) to unify all timestamps to the reference clock of the edge computing node. It then uses interpolation to correct data segments with large time deviations to obtain time-axis aligned time-series data. The third acquisition unit is used to acquire spatial coordinate registration based on time-series data aligned with the time axis, and to acquire spatial fusion data based on the spatial coordinate registration. The fourth acquisition unit is used to perform noise and redundancy processing on the spatial fusion data to obtain denoised and simplified data; The fifth acquisition unit is used to perform differentiated compression based on the denoised and simplified data to obtain compressed data; The sixth acquisition unit is used to perform format standardization conversion on the compressed data to obtain a standardized inspection data set with a unified format.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images