Intelligent robot inspection system in power distribution and substation intelligent auxiliary system

Abstract

This invention provides an intelligent robot inspection system for a power distribution and substation intelligent auxiliary system, relating to the field of intelligent sensing and control technology. It includes: a real-time operation and acquisition module for planning the global path of the intelligent inspection robot, generating an initial inspection path, performing real-time operation and acquisition, and obtaining a multimodal detection dataset; a fault feature identification module for identifying fault features, generating an abnormal feature set, and constructing a dynamic risk map; a simulated maintenance feedback module for performing simulated maintenance feedback based on the inspection correction path and equipment maintenance suggestions, and generating maintenance feedback results; and an inspection control module for generating an optimized inspection path to control the intelligent inspection robot. This invention solves the technical problem in existing technologies where inspection path planning often relies on a preset environmental map, and the intelligent inspection robot cannot automatically adjust its inspection path according to changes in equipment status, thus affecting inspection efficiency and fault diagnosis accuracy.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensing and control technology, specifically to an intelligent robot inspection system in a power distribution and substation intelligent auxiliary system. Background Technology

[0002] The distribution and transformation intelligent auxiliary system is an intelligent auxiliary system applied to the distribution and transformation links of the power system. Its core lies in the integration of advanced technologies such as computer hardware and software, artificial intelligence, and the Internet of Things to achieve comprehensive intelligent monitoring and management of power equipment, operating environment and safety protection.

[0003] Currently, intelligent robot inspection systems have been widely used in power equipment management. These robots achieve intelligent sensing and control by carrying various types of sensors. However, existing intelligent sensing and control technologies suffer from insufficient sensing accuracy due to sensor inherent limitations and environmental interference. This results in data collected by the sensors potentially containing noise or errors, affecting the accurate determination of equipment status and the accuracy of fault diagnosis. Furthermore, the path planning of intelligent inspection robots is typically based on the geographical distribution of equipment and inspection needs, lacking flexibility and the ability to dynamically adjust in real time according to changes in equipment status. This leads to high-risk equipment not receiving sufficient attention, increasing the risk of undetected faults and delaying repairs, ultimately impacting the operational safety of the power grid. Summary of the Invention

[0004] This application provides an intelligent robot inspection system for power distribution and substation intelligent auxiliary systems. It aims to solve the technical problem that in the prior art, inspection path planning often relies on a preset environmental map, and the intelligent inspection robot cannot automatically adjust the inspection path according to changes in equipment status, resulting in the inspection not being able to efficiently cover the faulty equipment area, thus affecting the inspection efficiency and fault diagnosis accuracy.

[0005] This application discloses an intelligent robot inspection system in a power distribution intelligent auxiliary system. The system includes: a real-time operation acquisition module, used to plan the global path of the intelligent inspection robot based on three-dimensional environmental data, generate an initial inspection path, and perform real-time operation acquisition of power distribution equipment in the power distribution intelligent auxiliary system to obtain a multimodal detection dataset; a fault feature identification module, used to identify fault features in the multimodal detection dataset, generate an abnormal feature set, and construct a dynamic risk map based on the abnormal feature set, the dynamic risk map including graded alarm signals and equipment maintenance suggestions; a simulated maintenance feedback module, used to correct the initial inspection path based on the graded alarm signals, perform simulated maintenance feedback based on the corrected inspection path and the equipment maintenance suggestions, and generate maintenance feedback results; and an inspection control module, used to generate an optimized inspection path according to the maintenance feedback results and send it to the power distribution intelligent auxiliary system to control the intelligent inspection robot during inspection.

[0006] One or more technical solutions provided in this application have at least the following beneficial effects:

[0007] By utilizing 3D environmental data to plan a global path, intelligent inspection robots can perform intelligent inspections of power distribution equipment based on the spatial distribution and environmental layout. Through the generation of more precise optimal inspection paths, the coverage and efficiency of inspections are improved. Simultaneously, global path planning reduces path repetition and wasted travel during inspections, lowering energy consumption and enabling efficient completion of inspection tasks. This process, through the application of intelligent sensing and control equipment, ensures the robot can perceive and adjust its path planning in real time, guaranteeing optimal execution of the inspection task. During the inspection process, real-time collection and comprehensive analysis of multimodal data makes the assessment of equipment status more comprehensive and accurate, enabling more effective identification of potential equipment faults. Automatic identification of fault characteristics generates a risk map, allowing for timely detection and handling of potential equipment problems, avoiding the limitations and incompleteness of manual inspections. The dynamic risk map provides a tiered alarm mechanism that issues corresponding warnings for faults of different severity levels, helping operators respond appropriately at different fault levels and ensuring timely and accurate intervention. The construction of dynamic risk maps can improve the intelligence level of equipment management and significantly enhance the timeliness and accuracy of fault early warning. Correcting the initial inspection path based on hierarchical alarm signals not only adjusts the priority of inspection tasks according to the actual health status of the equipment but also automatically adjusts the inspection path according to the alarm level, ensuring that high-risk equipment receives priority attention and optimizing inspection efficiency. The generation of simulated maintenance feedback allows the robot to make corrections based on equipment maintenance suggestions and simulate the effects of different maintenance measures. This process optimizes equipment maintenance strategies, predicts and adjusts inspection plans in advance, and improves equipment management efficiency. Finally, the optimized inspection path generated based on maintenance feedback allows the intelligent inspection robot to perform precise inspection control of the equipment. The optimized inspection path is more in line with the equipment status and maintenance needs, helping to improve the accuracy and efficiency of inspections while reducing ineffective robot paths. Overall, the application of intelligent sensing and control equipment ensures that the optimized path can be dynamically adjusted according to the real-time status of the equipment, making inspection tasks more efficient and accurate, and greatly improving the adaptive ability of the robot's behavior during the inspection process.

[0008] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0009] Figure 1This is a schematic diagram of the intelligent robot inspection system in the power distribution intelligent auxiliary system provided in the embodiments of this application.

[0010] Figure 2 This is a schematic diagram of the real-time data acquisition module in the intelligent robot inspection system of the power distribution intelligent auxiliary system provided in this application embodiment.

[0011] Figure labeling: Real-time operation acquisition module 10, fault feature identification module 20, simulated maintenance feedback module 30, inspection control module 40, three-dimensional spatial coordinate extraction unit 11, thermal feature annotation unit 12, path planning unit 13. Detailed Implementation

[0012] This application provides an intelligent robot inspection system in a power distribution and substation intelligent auxiliary system, which solves the technical problem that in the prior art, inspection path planning often relies on a preset environmental map, and the intelligent inspection robot cannot automatically adjust the inspection path according to changes in equipment status, resulting in the inspection not being able to efficiently cover the faulty equipment area, thus affecting the inspection efficiency and fault diagnosis accuracy.

[0013] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0014] like Figure 1 As shown in the embodiment of this application, an intelligent robot inspection system is provided in a power distribution and substation intelligent auxiliary system. The system includes:

[0015] The real-time operation acquisition module 10 is used to plan the global path of the intelligent inspection robot based on three-dimensional environmental data, generate the initial inspection path, and perform real-time operation acquisition of power distribution equipment in the power distribution intelligent auxiliary system to obtain a multimodal detection dataset.

[0016] Three-dimensional point cloud data is generated by scanning the substation with lidar. This point cloud data contains the spatial location and shape information of various objects in the environment. The three-dimensional spatial coordinates of the power distribution equipment are extracted from the point cloud data. Combined with thermal distribution data obtained from infrared thermal imaging sensors and the equipment spatial coordinate information extracted from the 3D point cloud, a thermal-spatial fusion array is generated. This array reflects the relationship between the thermal state and spatial location of the equipment. Based on the thermal-spatial fusion array, multi-objective path planning is performed on the intelligent auxiliary system for the power distribution substation to generate an initial inspection path.

[0017] The intelligent inspection robot is equipped with a visible light camera, an ultrasonic probe, and a partial discharge sensor. The visible light camera captures real-time images of the power distribution equipment, the ultrasonic probe captures ultrasonic signal data for structural health checks, and the partial discharge sensor captures partial discharge signals on the equipment for electrical checks. During real-time data acquisition, hardware trigger signals ensure the consistency of the acquisition timing from different sensors, providing a consistent timestamp for data fusion.

[0018] Sensor data preprocessing is performed to obtain a multimodal detection dataset. Specifically, real-time distortion correction is applied to the acquired visible light images to ensure geometric accuracy. An adaptive histogram equalization algorithm is used for illumination compensation to enhance image contrast and address image differences under different lighting conditions. Bandpass filtering is applied to the ultrasonic signals, and effective echo signals within a specific frequency band (e.g., 0.5-5MHz) are selected for analysis. This helps to eliminate noise signals outside the frequency band and enhance the recognition ability of effective signals. Amplitude-phase joint analysis is performed on the partial discharge pulse signals to eliminate electromagnetic interference pseudo-pulses and extract the true partial discharge signal features.

[0019] The fault feature identification module 20 is used to identify fault features in the multimodal detection dataset, generate an abnormal feature set, and construct a dynamic risk map based on the abnormal feature set. The dynamic risk map includes graded alarm signals and equipment maintenance suggestions.

[0020] The multimodal detection dataset is processed to extract their respective features, and the features provided by each sensor are converted into vectors, such as visible light image feature vectors, ultrasonic signal feature vectors, and partial discharge feature vectors. These feature vectors are then weighted and fused to obtain a multidimensional set of abnormal features. This step uses an attention mechanism to weight the features of different sensors, thereby more accurately reflecting the fault characteristics of different data sources.

[0021] By combining historical equipment defect data and using similarity analysis to compare currently collected anomaly features with historical defect data, potential failure modes are identified. Dynamic risk assessment is then performed based on these potential failure modes, generating tiered alarm signals categorized as low, medium, and high risk. Maintenance recommendations are provided based on these potential failure modes, including regular inspections, component replacement, and structural reinforcement. Failure modes, alarm signals, and equipment maintenance recommendations are integrated into a dynamic risk map, which reflects the real-time health status of the power distribution system, enabling operators to respond more quickly and accurately.

[0022] The simulated maintenance feedback module 30 is used to correct the initial inspection path based on the graded alarm signal, and to perform simulated maintenance feedback based on the corrected inspection path and the equipment maintenance suggestions to generate maintenance feedback results.

[0023] When a graded alarm signal is triggered, real-time analysis is performed first. The graded alarm signal reflects potential faults and risks in equipment operation. Based on the graded alarm signal, the target alarm devices that need to be inspected are determined, and their 3D coordinates and alarm risk levels are extracted. The 3D coordinates of the target alarm devices are used as nodes in the topology network, and the edge weights are calculated based on the alarm risk level and spatial distance to construct an alarm topology network. Correction analysis is performed based on the alarm topology network to generate a path correction factor set. This set contains path parameters that must be adjusted to avoid risk areas, collisions, or high-risk areas. The initial inspection path is corrected based on the path correction factor set to generate a corrected inspection path.

[0024] After obtaining the corrected inspection path, simulated maintenance feedback is generated based on equipment maintenance suggestions. This means that based on the current health status of the equipment (through analysis of alarm signals and failure modes), a simulated maintenance action plan is generated. According to the simulated maintenance process, the final simulated maintenance feedback result is generated, including information such as estimated maintenance time, required resources, and changes in equipment status. The simulated maintenance feedback result will serve as the basis for optimizing the inspection path, ensuring that the inspection robot can perform inspections according to the optimal path and sequence to minimize the risk of equipment failure.

[0025] The inspection control module 40 is used to generate an optimized inspection path according to the maintenance feedback results and send it to the power distribution intelligent auxiliary system to control the intelligent inspection robot.

[0026] By analyzing maintenance feedback results, the current health status of the equipment and maintenance recommendations are determined. Based on this, the inspection path is optimized. For example, for high-risk equipment, the inspection robot needs to increase the inspection frequency or change the inspection path. Specifically, a multi-objective evolutionary algorithm is used to optimize the inspection path. The optimization objectives include reducing path length, avoiding high-risk areas, and shortening inspection time.

[0027] After the path optimization is completed, the optimized inspection path is sent to the distribution substation intelligent auxiliary system. This path includes the revised path sequence and new inspection points. The distribution substation intelligent auxiliary system controls the inspection task of the intelligent inspection robot according to the optimized inspection path. The robot automatically performs inspections according to the optimized inspection path. This process ensures that the robot can perform inspection tasks along the optimal path, thereby improving inspection efficiency and the monitoring effect of equipment health status.

[0028] Furthermore, such as Figure 2As shown, the real-time data acquisition module includes:

[0029] The three-dimensional spatial coordinate extraction unit 11 is used to generate three-dimensional point cloud parameters by scanning the substation with lidar and extract the three-dimensional spatial coordinate set of the power distribution equipment; the thermal feature annotation unit 12 is used to fuse the thermal distribution data obtained by the infrared thermal imaging sensor with the three-dimensional spatial coordinate set, and to annotate the three-dimensional point cloud parameters with thermal features to generate a thermal-spatial fusion array; the path planning unit 13 is used to perform multi-objective path planning for the power distribution intelligent auxiliary system based on the thermal-spatial fusion array and generate the initial inspection path.

[0030] LiDAR uses laser beams to scan the environment, calculating distances by measuring the time it takes for a laser pulse to reflect back from a target object, thus obtaining the object's position and shape data. During the scanning process, LiDAR generates three-dimensional point cloud parameters, with each point containing the coordinates of a point in space, typically in three dimensions: X, Y, and Z. These points form the three-dimensional coordinate information of all objects in the environment. Based on the point cloud data, a set of three-dimensional spatial coordinates for power distribution equipment, including transformers, circuit breakers, switches, and lines, is extracted from a large number of points. This extracted set of equipment's three-dimensional spatial coordinates is used for subsequent path planning and analysis.

[0031] Infrared thermal imaging sensors can capture the temperature distribution on the surface of equipment. The sensor generates thermal distribution data by measuring the infrared radiation emitted by the surface of an object. Each point in the thermal distribution data represents the temperature at a certain location on the surface of the equipment, reflecting the temperature change of the equipment surface. In power distribution systems, the temperature change of equipment can reveal the working status of the equipment. For example, overheating indicates equipment failure or excessive load.

[0032] The thermal distribution data is fused with a set of three-dimensional spatial coordinates. Specifically, the thermal distribution data is converted into coordinates in three-dimensional space to match the three-dimensional spatial coordinates of the device. The three-dimensional spatial coordinates of each device location correspond to a certain temperature value in the thermal distribution, forming a thermal-spatial data point. The thermal information of each device is mapped to the three-dimensional space of the device to form a thermal-spatial fusion array. This array reflects the location of the device in space and its temperature status.

[0033] Based on a thermal-spatial fusion array, multi-objective path planning is performed to ensure that the intelligent inspection robot can efficiently inspect all equipment. Specifically, the first objective function is defined as maximizing equipment detection coverage, and the reachability weight coefficient of each equipment point is calculated. The second objective function is defined as prioritizing thermal anomaly areas, and the urgency of area detection is set according to the temperature rise rate. The third objective function is defined as minimizing path energy consumption, and the turning angle energy consumption factor is calculated based on the robot motion model. The weighted summation method is used to transform the multi-objective problem into a single-objective optimization problem. The constraints include obstacle avoidance radius and minimum equipment detection dwell time. Optimization algorithms such as A* algorithm, Dijkstra's algorithm, and genetic algorithm are used to calculate the optimal path. After planning, the initial inspection path is generated.

[0034] Furthermore, the thermal feature annotation unit includes:

[0035] The feature point matching channel is used to match feature points based on the thermal distribution data and the three-dimensional spatial coordinate set to construct a coordinate system transformation matrix; the visualization layer generation channel is used to retrieve the rated temperature threshold of the distribution and transformation equipment in the distribution and transformation intelligent auxiliary system to set dynamic pseudo-color mapping rules, and generate a temperature gradient visualization layer according to the dynamic pseudo-color mapping rules; the anomaly analysis channel is used to traverse the temperature gradient visualization layer and perform anomaly analysis in combination with the thermal distribution data to determine the temperature anomaly region, and extract the continuous thermal anomaly boundary contour based on the temperature anomaly region; the feature annotation channel is used to delineate the thermal anomaly region according to the continuous thermal anomaly boundary contour, and perform thermal feature annotation on the three-dimensional point cloud parameters based on the thermal anomaly region to generate the thermal-spatial fusion array.

[0036] Feature points are extracted from thermal distribution data and 3D spatial coordinate data. In 3D spatial coordinate data, these feature points are points with prominent characteristics in the equipment or environment (such as equipment surfaces, corners, and edges), while in thermal distribution data, they are points with significant temperature changes (such as locations with rapid temperature changes). Using spatial mapping methods, such as least squares or the RANSAC algorithm, the feature points in the thermal distribution data are matched with those in the 3D spatial coordinate data. This process ensures that the 2D temperature map of the thermal data can be correctly mapped to specific locations in 3D space.

[0037] Based on the matching feature points, a coordinate transformation matrix is ​​constructed. This matrix is ​​used to transform thermal distribution data from a two-dimensional plane to a three-dimensional space, so that the thermal data can be accurately fused with the three-dimensional spatial coordinate set of the device. Using this transformation matrix, the thermal map captured by the thermal imaging sensor (usually two-dimensional data) is converted into a thermal-spatial fusion array aligned with the lidar point cloud data (usually three-dimensional data), which provides accurate data support for subsequent analysis and path planning.

[0038] Power distribution equipment typically has a set rated operating temperature threshold. Equipment with a temperature exceeding this threshold may be at risk of failure. Therefore, the rated temperature threshold of the equipment should first be obtained from the power distribution intelligent auxiliary system. These temperature thresholds are the maximum safe operating temperatures specified in the equipment design.

[0039] Pseudo-color mapping maps temperature data to color values, allowing temperature changes to be visually displayed through color variations. Different temperature ranges can be mapped to different colors; for example, red represents high temperatures and blue represents low temperatures. Dynamic pseudo-color mapping rules automatically adjust the color mapping based on the device's real-time temperature data and temperature thresholds. For instance, when the device's temperature approaches or exceeds the threshold, it is displayed in a more prominent color for easier identification by inspection personnel. Based on dynamic pseudo-color mapping rules, the device's temperature data is transformed into a temperature gradient visualization layer. Temperature changes within this layer are visually represented through color gradients, enabling inspection personnel to quickly identify areas of abnormal temperature.

[0040] In the generated temperature gradient visualization layer, each region is traversed, focusing primarily on anomalies in temperature values. Anomalies are defined as temperatures exceeding the device's rated temperature threshold. Anomaly regions are identified by comparing the temperature at each device point with its rated temperature threshold. The boundary contours of these anomaly regions are extracted. For example, the Canny edge detection algorithm, the Sobel operator, or a region-growing method can be used to extract the boundaries of the anomaly regions, generating continuous thermal anomaly boundary contours. These contours identify the specific location and shape of the temperature anomalies, providing a reference for subsequent analysis.

[0041] Based on the extracted continuous thermal anomaly boundary contours, these contour areas are designated as thermal anomaly regions. Thermal anomaly regions focus on anomalies in heat transfer or distribution, considering not only temperature changes but also the source, transfer path, and distribution of heat energy across different areas, identifying anomalies from a thermodynamic perspective. Based on the designated thermal anomaly regions, their thermal information (such as temperature and rate of temperature change) is annotated into the corresponding 3D point cloud data. This annotation helps subsequent analysis identify which devices have overheating problems and categorizes them according to the severity of the problem. After annotation, the 3D point cloud data and thermal data are merged into a new data structure, namely a thermal-spatial fusion array. This array contains the 3D spatial coordinates and temperature information of the devices and indicates which areas belong to the temperature anomaly regions.

[0042] Furthermore, the fault feature identification module includes:

[0043] The system includes the following components: a feature extraction unit for cross-domain feature extraction of the multimodal detection dataset to generate multiple feature vectors; a weighted fusion unit for weighted fusion of the multiple feature vectors using an attention mechanism to construct a multidimensional abnormal feature set; a similarity analysis unit for retrieving a historical defect database, transferring the multidimensional abnormal feature set to the historical defect database for similarity analysis, identifying faults based on data similarity, and determining potential fault modes; a dynamic risk assessment unit for performing dynamic risk assessment based on the potential fault modes and constructing graded alarm signals; an equipment maintenance analysis unit for performing equipment maintenance analysis based on the potential fault modes and formulating equipment maintenance recommendations; and an association and integration unit for associating and integrating the graded alarm signals with the equipment maintenance recommendations to construct a dynamic risk map.

[0044] Based on the analysis of multimodal detection datasets, visible light image datasets, ultrasonic signal datasets, and partial discharge datasets are obtained. Different modalities have different feature spaces, enabling cross-domain feature extraction. Specifically, convolutional neural networks are used to extract spatial features from images, such as shape, color, and texture, generating image feature vectors. Spectral analysis is performed on ultrasonic signals, for example, through Fourier transform or wavelet transform, to extract frequency components and energy distribution, generating ultrasonic signal feature vectors. Based on the time-domain or frequency-domain features of partial discharge signals, features such as amplitude, frequency, and pulse clusters are extracted, generating partial discharge signal feature vectors. These feature vectors represent the state of the equipment in different modalities, providing support for subsequent fault identification and risk analysis.

[0045] Attention mechanisms assign weights based on the importance of different features in a task. In feature fusion of multimodal data, attention mechanisms help automatically determine which features of different modalities are more important, thereby strengthening the focus on important features. Multiple feature vectors are input into the attention mechanism, which assigns weights to each feature vector. These weights are adjusted based on the importance of the feature in the overall decision; features with higher weights have a greater impact on the final decision. Through weighted fusion, multiple feature vectors are combined into a multidimensional set of anomalous features, representing the comprehensive information from different modalities in potential fault detection.

[0046] Retrieve the historical defect database, which contains past equipment failure modes, failure types, failure conditions, and repair records. By querying this database, you can find historical failure modes that are similar to the current equipment status.

[0047] The currently generated multidimensional anomaly feature set is migrated to a historical defect database for comparison. This migration involves standardizing and normalizing the features to ensure they match those in the historical database. The similarity between these features and the historical data is calculated; for example, a cosine similarity metric is used to calculate the similarity between the current feature vector and the historical data vector. A higher similarity value indicates a higher similarity, meaning the current equipment faces a fault similar to a certain fault mode in the historical data. Based on the similarity analysis results, potential fault modes of the current equipment are identified, such as overheating, structural damage, and electrical discharge.

[0048] Risk assessments are conducted on potential failure modes to determine the likelihood of failures and their impact on the system. Specifically, based on historical data, equipment health status, and operating environment, the probability of a particular failure mode occurring is evaluated; the higher the probability, the greater the risk. The impact of a failure on the equipment or the entire power distribution system is analyzed; for example, some failures cause complete system shutdowns, while others have only localized effects or even no significant impact. The probability and impact of failures are quantified, for example, by using a risk matrix to assess different failure modes and calculate the risk value of potential failure modes. Based on the dynamic risk assessment results, corresponding alarm signals are generated for potential failure modes and classified according to different risk value thresholds to construct a tiered alarm signal system.

[0049] Based on the dynamic risk assessment results of potential failure modes, develop targeted equipment maintenance plans to prevent failures from occurring or reduce their impact on the system.

[0050] For example, failure modes with low risk values ​​typically have little impact on equipment or systems, generating low-risk alarms and recommending regular inspections or monitoring. Regular inspections include routine maintenance such as equipment cleaning, lubrication, and temperature monitoring. Failure modes with high risk values ​​require partial repair or maintenance, generating medium-risk alarms and recommending preventative maintenance. This type of maintenance focuses on prevention by detecting and replacing some parts or systems to reduce the likelihood of failure. Failure modes with very high risk values ​​may cause equipment damage or system shutdown, generating high-risk alarms and recommending immediate action, including shutting down the equipment, emergency repairs, or replacement of critical components.

[0051] By associating graded alarm signals with corresponding equipment maintenance recommendations, the level of the alarm signal (low, medium, high risk) determines the priority and handling method of the maintenance recommendations. The dynamic risk map is a visualization tool used to display the real-time health status, fault risk, and maintenance needs of various parts of equipment, systems, or networks. The map presents alarm signals and maintenance recommendations in a graphical way, helping staff to quickly identify equipment that needs to be focused on.

[0052] Furthermore, the feature extraction unit includes:

[0053] The system comprises the following channels: a parsing channel for parsing the multimodal detection dataset to obtain visible light image datasets, ultrasonic signal datasets, and partial discharge datasets; a dual-channel convolution analysis channel for performing dual-channel convolution analysis on the visible light image dataset to obtain visible light image features; a wavelet packet energy entropy analysis channel for performing wavelet packet energy entropy analysis on the ultrasonic signal dataset to obtain ultrasonic signal features; a pulse clustering channel for performing pulse clustering on the partial discharge dataset to extract partial discharge features; a feature vector construction channel for analyzing the temporal evolution trends of the visible light image features, ultrasonic signal features, and partial discharge features to construct visible light image feature vectors, ultrasonic signal feature vectors, and partial discharge feature vectors; and a data integration channel for integrating the visible light image feature vectors, ultrasonic signal feature vectors, and partial discharge feature vectors to obtain the multiple feature vectors.

[0054] During the intelligent inspection process, visible light image datasets, ultrasonic signal datasets, and partial discharge datasets are collected by the visible light camera, ultrasonic probe, and partial discharge sensor configured on the intelligent inspection robot.

[0055] A dual-channel convolutional network is a deep learning method that processes input data from two different sources simultaneously through two channels. In visible light image processing, a dual-channel convolutional network can extract image features from different processing levels. The first channel extracts surface defect texture features, while the second channel extracts structural deformation geometric features. Each channel extracts image features through a series of convolution operations. The features extracted from the two different channels are then fused to form the visible light image features.

[0056] Wavelet packet transform is a time-frequency analysis method that decomposes a signal into different frequency bands (high-frequency and low-frequency components), enabling it to better capture the subtle changes in the signal. It has a stronger frequency decomposition capability than traditional wavelet transform, extracting more frequency components from the signal. Energy entropy is an indicator used to measure the uncertainty or complexity of a signal. In ultrasonic signal analysis, energy entropy reflects the complexity and information distribution of the signal. Signals with high energy entropy indicate complex physical phenomena, such as cracks or corrosion, while signals with low energy entropy are simpler and more regular.

[0057] Wavelet packet transform is used to decompose the ultrasonic signal into multiple frequency bands, representing the signal's components within different frequency ranges. Wavelet packet decomposition captures the high-frequency and low-frequency components of the ultrasonic signal, which can be used to detect internal cracks or damage. Energy calculations are performed on the signal in each frequency band to obtain its energy value. By calculating the energy distribution across different frequency bands, the signal's energy spectrum is obtained. Based on the energy spectrum, the entropy value of each frequency band is calculated. A higher entropy value indicates a more complex signal within that band. Wavelet packet energy entropy analysis reveals ultrasonic signal characteristics that reflect the signal's complexity and frequency components.

[0058] Partial discharge refers to minute electrical discharge phenomena inside or on the surface of electrical equipment. It typically occurs in electrical insulation materials and can reflect the electrical health status of the equipment. Partial discharge signals often increase before equipment failure. Pulse clustering is a method for analyzing partial discharge signals. It clusters pulses in the partial discharge signal according to certain time intervals or similarities. Clustering can reveal the nature, frequency, and region of discharge events. Using clustering algorithms, such as K-means clustering and DBSCAN clustering, similar discharge events are grouped into the same cluster by extracting three-dimensional statistics such as discharge counts / second, average amplitude, and phase distribution skewness, thereby extracting the characteristic patterns of the discharge as partial discharge features.

[0059] Temporal evolution analysis refers to the analysis of the trends in equipment status data (including visible light images, ultrasonic signals, and partial discharge signals) over time. This analysis can reveal the dynamic changes in equipment health status and help identify potential failure modes. Specifically, by analyzing consecutive frames of visible light images, trends in changes to the equipment surface can be identified, including crack propagation and surface wear. During the analysis, techniques such as image differencing and optical flow are used to track the movement and changes of objects between consecutive frames, constructing visible light image feature vectors that reflect changes in the equipment's appearance. Ultrasonic signals are used to detect structural problems inside equipment, such as cracks and corrosion. Over time, the frequency, amplitude, and phase of the signal change. Temporal analysis yields ultrasonic signal feature vectors, providing quantitative information on the internal health status of the equipment. Partial discharge signals are important early signs of electrical equipment failure. As equipment ages or is damaged, the frequency, amplitude, and pulse pattern of partial discharges change. Temporal analysis yields partial discharge feature vectors that reflect changes in the equipment's electrical state.

[0060] By integrating the feature vectors of visible light images, ultrasonic signals, and partial discharge, multiple feature vectors are obtained, which simultaneously reflect the health status of the equipment in different dimensions (visual, structural, and electrical).

[0061] Furthermore, the simulated maintenance feedback module includes:

[0062] The real-time analysis unit is used to perform real-time analysis based on the hierarchical alarm signals, determine the target alarm device, and extract the three-dimensional alarm coordinates and alarm risk level based on the target alarm device; the correction analysis unit is used to construct an alarm topology network according to the three-dimensional alarm coordinates and the alarm risk level, perform correction analysis based on the alarm topology network, and generate a path correction factor set; the correction unit is used to correct the initial inspection path based on the path correction factor set and generate an initial inspection correction path; the correction path determination unit is used to perform collision detection according to the initial inspection correction path, determine the completion progress index of the intelligent robot inspection task based on the detection results, and determine the inspection correction path based on the completion progress index.

[0063] The system performs real-time analysis of tiered alarm signals to determine which device is malfunctioning. This analysis identifies target alarm devices that require priority handling; these are typically high-risk devices whose failures could severely impact the system. The location of the target alarm device is determined based on its 3D spatial coordinates (derived from LiDAR scan data), and the 3D alarm coordinates are extracted for subsequent path planning and correction. Simultaneously, the alarm risk level of the target alarm device is extracted for path correction and priority ranking.

[0064] An alarm topology network is constructed using the 3D alarm coordinates and alarm risk levels of the target alarm devices. In this network, each node represents a device, and the edges between devices represent their spatial or functional relationships. The weight of a node is determined by its risk level; for example, nodes of high-risk devices have higher weights. After constructing the alarm topology network, path correction analysis is performed. By calculating the distance between different devices, their risk levels, and the priority of alarm signals, it is analyzed which devices need to be prioritized for inspection and which devices can be temporarily skipped. Based on the topology network analysis, a set of path correction factors is generated. These correction factors represent the priority between devices, the adjusted inspection order, and the paths that need to avoid high-risk areas.

[0065] Based on the generated set of path correction factors, the initial inspection path is corrected, including prioritizing high-risk equipment at the beginning of the path; avoiding areas where high risk or malfunctions have already been detected to ensure the robot does not enter potentially hazardous areas; and optimizing the path to reduce unnecessary repetitive inspections and improve efficiency. After path correction, a revised initial inspection path is generated, which is more effectively able to address the health status and malfunction risks of the equipment.

[0066] Collision detection is performed along the initial inspection and correction path. This includes detecting whether the robot will collide with obstacles in the environment (such as equipment, walls, facilities, etc.) and whether collisions will occur due to confined space or equipment configuration issues during robot movement. Simulation technology is used to virtually run the correction path and detect potential collision risks.

[0067] Based on the collision detection results, the progress of the inspection task is determined. The progress index is a numerical value reflecting the degree of completion of the robot's inspection task. It is calculated based on the equipment that has been inspected and the remaining inspection tasks. For example, if there are obstacles on the path or equipment malfunctions, the progress index decreases; if the robot successfully completes the inspection without encountering obstacles, the progress index increases. Based on the calculated progress index, the inspection path is further optimized. For example, if the progress index is low, the path is adjusted or additional inspection points are added to ensure the task can be completed successfully. After optimization, the corrected inspection path is determined to improve inspection efficiency.

[0068] Furthermore, the correction analysis unit includes:

[0069] The system includes the following channels: an edge weight parameter acquisition channel, used to calculate node spatial distance data based on the 3D alarm coordinates as topology nodes, and using the alarm risk level combined with the node spatial distance data as edge weight parameters; an alarm topology network construction channel, used to construct an alarm topology network based on the topology nodes and the edge weight parameters; a calculation channel, used to calculate the initial inspection path according to the alarm topology network to obtain a correction cost matrix, which includes path correction weights; a descending order sorting channel, used to perform multi-objective evaluation based on the path correction weights, and sort the results in descending order to generate a path correction priority ranking list; and a simulation channel, used to determine the path correction factor set by combining the path correction priority ranking list with the motion trajectory simulation of the intelligent robot.

[0070] In the alarm topology network, the location of each device is represented by three-dimensional alarm coordinates. As a node in the topology network, the spatial distance between topology nodes is calculated using Euclidean distance. When constructing the topology network, the connection (i.e., edge) between each pair of devices is weighted according to its spatial distance and alarm risk level. The edge weight combines the physical distance between devices and the urgency of device failure. The shorter the distance, the smaller the weight of the connection, indicating the relative closeness between devices; the higher the risk level (e.g., high-risk alarm), the larger the weight, indicating that the device needs to be dealt with first.

[0071] Using the calculated edge weight parameters and topology nodes, an alarm topology network is constructed. This network consists of several nodes (alarm devices) and the edges between them (representing the relationship or connection between devices). The weight of each edge is determined by the spatial distance between devices and the alarm risk level.

[0072] The initial inspection path does not consider risk level and spatial distance. Based on the alarm topology network, the correction cost of the initial inspection path is calculated. The correction cost represents the path cost from one device to another, including the weighted impact of spatial distance and risk level. The correction cost matrix is ​​obtained after calculation, which represents the correction cost of each path.

[0073] A multi-objective evaluation is conducted, with optimization objectives including: minimizing path length (optimizing the total length of the inspection path to reduce the robot's travel distance and time); minimizing risk (prioritizing the inspection of high-risk equipment to ensure timely monitoring of equipment health); and collision avoidance (preventing collisions between the robot and obstacles or other equipment during inspection). Based on these objectives, a comprehensive score is calculated for each path, incorporating path correction weights. According to the results of the multi-objective evaluation, all corrected paths are sorted in descending order to generate a path correction priority ranking list. A higher priority path indicates greater importance during inspection; high-priority paths are selected for inspection first, while low-priority paths are postponed or skipped, especially when the path contains low-risk equipment or has already passed other inspection paths.

[0074] After determining the priority list for path correction, the robot's movement along the corrected path is simulated using motion trajectory simulation to identify potential problems during actual execution, such as collisions and route adjustments. Based on the simulation results, a set of path correction factors is determined. These factors include equipment inspection priorities, obstacle avoidance routes, and equipment health status. Each correction factor represents a portion of the path that needs adjustment during the inspection process.

[0075] Furthermore, the corrected path determination unit includes:

[0076] The collision detection channel is used to perform multimodal collision detection according to the initial inspection correction path and acquire obstacle spatial distribution data in real time; the progress evaluation channel is used to introduce the intelligent robot inspection task for progress evaluation, obtain the completion progress index of the intelligent robot inspection task, and calculate the remaining path risk assessment value based on the completion progress index; the local correction channel is used to perform local correction based on the obstacle spatial distribution data and the remaining path risk assessment value to generate an obstacle avoidance path candidate set; the update channel is used to update the initial inspection correction path according to the obstacle avoidance path candidate set and determine the inspection correction path.

[0077] During the robot's inspection process, following the initial inspection and correction path, real-time images of the power distribution equipment are captured by a visible light camera, ultrasonic signal data is captured by an ultrasonic probe for structural health detection, and partial discharge signals on the equipment are captured by a partial discharge sensor for electrical detection. This data provides real-time updates for collision detection, enabling the robot to dynamically adjust its path. Based on the sensor data, spatial distribution data of obstacles in the environment is generated. This data structure is an obstacle map with spatial coordinates, reflecting the type, location, and size of the obstacles.

[0078] The progress of the task is assessed based on the robot's actual inspection status. The progress index is a digital indicator representing the robot's percentage progress in the inspection task. For example, if the robot has completed 50% of the inspection path, the progress index is 0.5. Based on the current progress index, the risk of the remaining path is assessed. If the risk level of the remaining equipment is higher, the risk assessment value of the remaining path will be higher.

[0079] Based on obstacle spatial distribution data and remaining path risk assessment values, the inspection path is locally modified to avoid obstacles and ensure the smooth completion of the inspection task in high-risk areas. Specifically, where there are obstacles, a path planning algorithm, such as the A* algorithm, is used to find a new path to bypass the obstacle or high-risk area. If the path passes through a high-risk area, path modification is used to avoid entering that area, or the path is replanned to reduce the possibility of failure. Based on the local modifications, a candidate set of obstacle avoidance paths is generated, representing different detour methods or optimized paths.

[0080] After generating a candidate set of obstacle avoidance paths, the optimal path is selected based on the evaluation results and updated as the inspection correction path. The new inspection correction path has the characteristics of obstacle avoidance and low risk. This path will be provided to the intelligent inspection robot so that it can successfully complete the inspection task according to the new path.

[0081] Furthermore, the inspection control module includes:

[0082] The system comprises the following components: a state change analysis unit, used to analyze the state changes of the distribution equipment in the distribution substation intelligent auxiliary system based on the maintenance feedback results, and determine equipment state improvement parameters; a secondary detection analysis unit, used to traverse the distribution equipment in the distribution substation intelligent auxiliary system according to the equipment state improvement parameters, and perform secondary detection analysis to determine secondary detection priority coefficients; a health prediction unit, used to retrieve historical patrol data and combine it with the secondary detection priority coefficients to perform health prediction on the distribution equipment in the distribution substation intelligent auxiliary system, and generate equipment health prediction scores; a multi-objective evolution unit, used to perform multi-objective evolution on the inspection correction path according to the equipment health prediction scores, and generate the optimized inspection path; and an inspection control unit, used to send the optimized inspection path to the distribution substation intelligent auxiliary system to control the intelligent inspection robot, forming a closed-loop control circuit.

[0083] Based on maintenance feedback results, the status change analysis of power distribution equipment is carried out. The purpose of the analysis is to identify whether the status of the equipment has improved after maintenance, whether it has returned to normal operation, or whether there are still potential problems. For example, if the equipment returns to normal operation after maintenance, or if the equipment has not fully recovered after maintenance, or if the health status of the equipment fluctuates, the equipment status improvement parameters are determined based on the analysis results. The equipment status improvement parameters are indicators used to quantify the changes in equipment status.

[0084] After the initial maintenance, a secondary inspection and analysis is performed on the equipment based on its condition improvement parameters. This secondary inspection aims to further confirm whether the equipment has returned to its ideal state or to check for any new problems. All distribution and substation equipment in the intelligent auxiliary system is examined, and equipment requiring secondary inspection is determined according to its condition change parameters (such as temperature, efficiency, and failure rate). For example, if equipment condition improvement is not significant or the fault recurs, it is marked as high-priority equipment and undergoes secondary inspection; if equipment condition improvement is good, secondary inspection is not required, or a routine inspection is performed. To optimize resource and time allocation, a secondary inspection priority coefficient is assigned to each piece of equipment based on factors such as risk level, failure mode, and maintenance history. Equipment is then prioritized, with high-priority equipment being inspected first, while low-priority equipment can be inspected later.

[0085] Historical patrol data refers to equipment operation data collected by intelligent inspection robots during past inspection tasks. This data includes equipment operating status, inspection paths, maintenance records, equipment fault history, environmental data, and other information. Based on historical patrol data and secondary inspection priority coefficients, the health prediction of power distribution equipment in the power distribution intelligent auxiliary system is performed. Specifically, machine learning models based on historical data and current equipment status, such as regression analysis, time series analysis, or neural networks, are used to predict the health of each piece of equipment. By adjusting the focus of health prediction according to the secondary inspection priority of the equipment, the health status of high-priority equipment is ensured to be more accurately assessed. By analyzing the equipment's historical operating data, fault records, and environmental factors, the health status of the equipment in the future is predicted, generating an equipment health prediction score. The equipment health prediction score is a quantitative score based on the above prediction results, representing the equipment's current and future health status.

[0086] Based on equipment health prediction scores, multi-objective path optimization is performed on the inspection correction path. Optimization objectives include minimizing inspection time, minimizing risk, and optimizing resources (reducing unnecessary repetitive inspections and saving robot working time and energy). Evolutionary algorithms, such as genetic algorithms and particle swarm optimization, are used to optimize the inspection path. These algorithms, through an iterative process, integrate different objectives to find the optimal solution. The evolutionary algorithm adjusts the inspection path according to the equipment health prediction score, prioritizing equipment with low health scores and higher risks, and further optimizing the path accordingly. After multi-objective evolution, an optimized inspection path is generated. This path achieves an optimal balance in terms of time, risk, and resources. The optimized path will be used in subsequent inspection tasks to ensure the efficiency of robot inspection tasks and the health management of equipment.

[0087] The optimized inspection path is sent to the distribution substation intelligent auxiliary system. Based on this path, the system controls the movement of the intelligent inspection robot, which autonomously performs the inspection task. During the inspection, the robot continuously collects operational data and feedback information from the equipment using sensors (including LiDAR, cameras, and ultrasonic detectors), and adjusts the inspection path in real time. For example, if the health status of a piece of equipment changes significantly or a new alarm signal appears, the inspection path is dynamically adjusted based on real-time data to ensure the robot can respond to changes in equipment health in real time. This forms a closed-loop control system, ensuring dynamic optimization of the inspection task. This closed-loop control process allows the robot to flexibly adjust according to the constantly changing health status of the equipment, ensuring continuous monitoring and timely maintenance of the equipment's health.

[0088] Furthermore, the optimized inspection path is sent to the distribution substation intelligent auxiliary system to control the intelligent inspection robot during inspection. The inspection control unit includes:

[0089] The inspection control channel is used to send the optimized inspection path to the distribution substation intelligent auxiliary system to control the intelligent inspection robot and collect mechanical vibration feedback data of the intelligent inspection robot; the twin simulation channel is used to perform twin simulation on the intelligent inspection robot based on the mechanical vibration feedback data to obtain equipment state evolution simulation parameters; the closed-loop control channel is used to activate the online replanning mechanism to perform closed-loop control when the error between the equipment state improvement parameters and the equipment state evolution simulation parameters exceeds a preset threshold, thereby constructing the closed-loop control loop.

[0090] The optimized inspection path is sent to the power distribution and substation intelligent auxiliary system. The system controls the intelligent inspection robot according to the path instructions to ensure that the robot moves along the optimized path to cover all equipment that needs to be inspected. During the inspection, the intelligent inspection robot continuously monitors its own mechanical state through built-in vibration sensors, such as accelerometers and gyroscopes. These sensors can capture vibration information of the robot when performing inspection tasks, especially the vibration patterns of mechanical components (such as motors and gears).

[0091] Twin simulation refers to the virtual modeling and simulation of robots and equipment, using real-time data to simulate the equipment's state and predict its health status over a future period. Based on mechanical vibration feedback data, the robot's inspection process is simulated in the virtual model. By analyzing the robot's motion trajectory, vibration amplitude, and frequency changes, the potential state changes of the equipment during the inspection process are predicted, including the equipment's health status, fault evolution, and possible performance degradation. The simulation output includes equipment state evolution simulation parameters, including the predicted equipment health status and potential faults.

[0092] The equipment status improvement parameters are compared with the predicted equipment status evolution simulation parameters in the twin simulation. When the error exceeds a preset threshold, it indicates that the equipment status has not changed as expected, requiring further monitoring or intervention. At this point, an online replanning mechanism is activated. This mechanism adjusts the inspection path, priority, and task allocation in real time to ensure the robot's inspection task remains effective and to promptly identify and address potential equipment problems. After the replanning mechanism is activated, a closed-loop control loop is formed. Through real-time data acquisition, simulation feedback, and path optimization, the inspection plan can be dynamically adjusted based on the equipment's health status and the actual progress of the inspection task, ensuring continuous optimization and efficient execution of the inspection task.

[0093] In summary, the intelligent robot inspection system in the power distribution and substation intelligent auxiliary system provided in this application embodiment has the following technical effects:

[0094] By utilizing 3D environmental data to plan a global path, intelligent inspection robots can perform intelligent inspections of power distribution equipment based on the spatial distribution and environmental layout. Through the generation of more precise optimal inspection paths, the coverage and efficiency of inspections are improved. Simultaneously, global path planning reduces path repetition and wasted travel during inspections, lowering energy consumption and enabling efficient completion of inspection tasks. This process, through the application of intelligent sensing and control equipment, ensures the robot can perceive and adjust its path planning in real time, guaranteeing optimal execution of the inspection task. During the inspection process, real-time collection and comprehensive analysis of multimodal data makes the assessment of equipment status more comprehensive and accurate, enabling more effective identification of potential equipment faults. Automatic identification of fault characteristics generates a risk map, allowing for timely detection and handling of potential equipment problems, avoiding the limitations and incompleteness of manual inspections. The dynamic risk map provides a tiered alarm mechanism that issues corresponding warnings for faults of different severity levels, helping operators respond appropriately at different fault levels and ensuring timely and accurate intervention. The construction of dynamic risk maps can improve the intelligence level of equipment management and significantly enhance the timeliness and accuracy of fault early warning. Correcting the initial inspection path based on hierarchical alarm signals not only adjusts the priority of inspection tasks according to the actual health status of the equipment but also automatically adjusts the inspection path according to the alarm level, ensuring that high-risk equipment receives priority attention and optimizing inspection efficiency. The generation of simulated maintenance feedback allows the robot to make corrections based on equipment maintenance suggestions and simulate the effects of different maintenance measures. This process optimizes equipment maintenance strategies, predicts and adjusts inspection plans in advance, and improves equipment management efficiency. Finally, the optimized inspection path generated based on maintenance feedback allows the intelligent inspection robot to perform precise inspection control of the equipment. The optimized inspection path is more in line with the equipment status and maintenance needs, helping to improve the accuracy and efficiency of inspections while reducing ineffective robot paths. Overall, the application of intelligent sensing and control equipment ensures that the optimized path can be dynamically adjusted according to the real-time status of the equipment, making inspection tasks more efficient and accurate, and greatly improving the adaptive ability of the robot's behavior during the inspection process.

[0095] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent robot inspection system in a power distribution and substation intelligent auxiliary system, characterized in that, The system includes: The real-time operation acquisition module is used to plan the global path of the intelligent inspection robot based on 3D environmental data, generate the initial inspection path, and perform real-time operation acquisition of power distribution equipment in the power distribution intelligent auxiliary system to obtain multimodal detection datasets. The real-time data acquisition module includes: The three-dimensional spatial coordinate extraction unit is used to generate three-dimensional point cloud parameters by scanning the substation with lidar and extract the three-dimensional spatial coordinate set of the power distribution equipment. The thermal feature annotation unit is used to fuse the thermal distribution data obtained by the infrared thermal imaging sensor with the three-dimensional spatial coordinate set, perform thermal feature annotation on the three-dimensional point cloud parameters, and generate a thermal-spatial fusion array. The path planning unit is used to perform multi-objective path planning for the distribution substation intelligent auxiliary system based on the thermal-spatial fusion array, and generate the initial inspection path. The fault feature identification module is used to identify fault features in the multimodal detection dataset, generate an abnormal feature set, and construct a dynamic risk map based on the abnormal feature set. The dynamic risk map includes graded alarm signals and equipment maintenance suggestions. The simulated maintenance feedback module is used to correct the initial inspection path based on the graded alarm signal, and to perform simulated maintenance feedback based on the corrected inspection path and the equipment maintenance suggestions to generate maintenance feedback results. The simulated maintenance feedback module includes: The real-time analysis unit is used to perform real-time analysis based on the hierarchical alarm signal, determine the target alarm device, and extract the three-dimensional alarm coordinates and alarm risk level based on the target alarm device. The correction analysis unit is used to construct an alarm topology network according to the three-dimensional alarm coordinates and the alarm risk level, perform correction analysis based on the alarm topology network, and generate a set of path correction factors. The correction unit is used to correct the initial inspection path based on the path correction factor set and generate an initial inspection correction path. The path correction determination unit is used to perform collision detection according to the initial inspection correction path, determine the completion progress index of the intelligent robot inspection task based on the detection results, and determine the inspection correction path based on the completion progress index. The corrected path determination unit includes: The collision detection channel is used to perform multimodal collision detection according to the initial inspection correction path and acquire obstacle spatial distribution data in real time. The progress assessment channel is used to introduce the intelligent robot inspection task for progress assessment, obtain the completion progress index of the intelligent robot inspection task, and calculate the remaining path risk assessment value based on the completion progress index. The local correction channel is used to make local corrections based on the spatial distribution data of obstacles and the risk assessment value of the remaining path, and to generate a candidate set of obstacle avoidance paths. An update channel is used to update the initial inspection correction path based on the obstacle avoidance path candidate set, and to determine the inspection correction path; The inspection control module is used to generate an optimized inspection path based on the maintenance feedback results and send it to the power distribution and substation intelligent auxiliary system to control the intelligent inspection robot.

2. The intelligent robot inspection system in the power distribution and transformation intelligent auxiliary system as described in claim 1, characterized in that, The thermal feature annotation unit includes: The feature point matching channel is used to match feature points based on the thermal distribution data and the three-dimensional spatial coordinate set, and to construct a coordinate system transformation matrix. The visualization layer generation channel is used to retrieve the dynamic pseudo-color mapping rule for the rated temperature threshold of the power distribution equipment in the power distribution and substation intelligent auxiliary system, and generate a temperature gradient visualization layer according to the dynamic pseudo-color mapping rule. Anomaly analysis channel is used to traverse the temperature gradient visualization layer and combine it with the thermal distribution data to perform anomaly analysis, determine the temperature anomaly area, and extract the continuous thermal anomaly boundary contour based on the temperature anomaly area. The feature annotation channel is used to delineate the thermal anomaly region according to the continuous thermal anomaly boundary contour, and to perform thermal feature annotation on the three-dimensional point cloud parameters based on the thermal anomaly region to generate the thermal-spatial fusion array.

3. The intelligent robot inspection system in the power distribution and transformation intelligent auxiliary system as described in claim 1, characterized in that, The fault feature identification module includes: The feature extraction unit is used to perform cross-domain feature extraction on the multimodal detection dataset and generate multiple feature vectors; The weighted fusion unit is used to perform weighted fusion based on the multiple feature vectors using an attention mechanism to construct a multidimensional set of abnormal features. The similarity analysis unit is used to retrieve the historical defect database, migrate the multidimensional abnormal feature set to the historical defect database for similarity analysis, identify faults based on data similarity, and determine potential fault modes. A dynamic risk assessment unit is used to perform dynamic risk assessment based on the potential failure modes and construct hierarchical alarm signals. The equipment maintenance analysis unit is used to perform equipment maintenance analysis based on the potential failure modes and formulate equipment maintenance recommendations. The association and integration unit is used to associate and integrate the hierarchical alarm signals with the equipment maintenance suggestions to construct a dynamic risk map.

4. The intelligent robot inspection system in the power distribution and transformation intelligent auxiliary system as described in claim 3, characterized in that, The feature extraction unit includes: The parsing channel is used to parse the multimodal detection dataset to obtain visible light image dataset, ultrasonic signal dataset, and partial discharge dataset; A dual-channel convolutional analysis channel is used to perform dual-channel convolutional analysis based on the visible light image dataset to obtain visible light image features; The wavelet packet energy entropy analysis channel is used to perform wavelet packet energy entropy analysis based on the ultrasonic signal dataset to obtain ultrasonic signal characteristics. The pulse clustering channel is used to perform pulse clustering based on the partial discharge dataset and extract partial discharge features; A feature vector construction channel is used to analyze the temporal evolution trends of the visible light image features, the ultrasonic signal features, and the partial discharge features, and to construct visible light image feature vectors, ultrasonic signal feature vectors, and partial discharge feature vectors. The data integration channel is used to integrate the visible light image feature vector, the ultrasonic signal feature vector, and the partial discharge feature vector to obtain the multiple feature vectors.

5. The intelligent robot inspection system in the power distribution and transformation intelligent auxiliary system as described in claim 1, characterized in that, The correction analysis unit includes: The edge weight parameter acquisition channel is used to calculate node spatial distance data based on the three-dimensional alarm coordinates as topology nodes, and to use the alarm risk level combined with the node spatial distance data as edge weight parameters. An alarm topology network construction channel is used to construct an alarm topology network based on the topology nodes and the edge weight parameters. A calculation channel is used to calculate the initial inspection path according to the alarm topology network to obtain a correction cost matrix, wherein the correction cost matrix includes path correction weights; The descending order channel is used to perform multi-objective evaluation based on the path correction weight, and to generate a path correction priority sorting list by arranging the evaluation results in descending order. The simulation channel is used to determine the set of path correction factors by combining the path correction priority sorting list with the motion trajectory simulation of the intelligent robot.

6. The intelligent robot inspection system in the power distribution and transformation intelligent auxiliary system as described in claim 1, characterized in that, The inspection control module includes: The status change analysis unit is used to perform status change analysis on the power distribution equipment of the power distribution intelligent auxiliary system based on the maintenance feedback results, and to determine the equipment status improvement parameters. The secondary detection and analysis unit is used to perform secondary detection and analysis on the power distribution equipment of the power distribution and substation intelligent auxiliary system according to the equipment status improvement parameters, and to determine the secondary detection priority coefficient. The health prediction unit is used to retrieve historical cruise data and combine it with the secondary detection priority coefficient to perform health prediction on the power distribution equipment of the power distribution intelligent auxiliary system and generate equipment health prediction scores. A multi-objective evolutionary unit is used to perform multi-objective evolution on the inspection correction path based on the equipment health prediction score, and generate the optimized inspection path. The inspection control unit is used to send the optimized inspection path to the power distribution intelligent auxiliary system to control the intelligent inspection robot, forming a closed-loop control circuit.

7. The intelligent robot inspection system in the power distribution and transformation intelligent auxiliary system as described in claim 6, characterized in that, The inspection control unit includes: The inspection control channel is used to send the optimized inspection path to the power distribution intelligent auxiliary system to control the intelligent inspection robot and collect mechanical vibration feedback data of the intelligent inspection robot. The twin simulation channel is used to perform twin simulation on the intelligent inspection robot based on the mechanical vibration feedback data to obtain simulation parameters of equipment state evolution. A closed-loop control channel is used to initiate an online replanning mechanism to perform closed-loop control and construct the closed-loop control loop when the error between the equipment state improvement parameter and the equipment state evolution simulation parameter exceeds a preset threshold.

Images