Intelligent substation equipment state intelligent inspection method and device

Abstract

The application relates to the technical field of substation inspection, in particular to a smart substation equipment state intelligent inspection method and device, which comprises the following steps: constructing an intelligent visual inspection coordinate system based on three-dimensional space, optimizing the optimal inspection route, inspection distance and inspection angle of an inspection robot, and generating an inspection task list; acquiring equipment image feature data and electrical parameter data synchronously collected by the inspection robot according to the inspection task list to form a multi-source data set; performing collaborative diagnosis and verification of equipment states and faults through collaborative analysis of a random forest classification model and an improved YOLOv5 recognition model; and based on the collaborative diagnosis result, historical operation data and real-time collection data of the equipment, a multi-dimensional early warning model of limit early warning, mutation early warning, synchronous early warning and trend early warning is established to realize equipment life cycle early warning. The application helps to reduce the probability of equipment failure expansion.

Description

Technical Field

[0001] This application relates to the field of substation inspection technology, and in particular to a method and device for intelligent inspection of equipment status in smart substations. Background Technology

[0002] As the core hub of the power system, the safe and stable operation of substation equipment is directly related to the reliability of power supply. Equipment condition inspection is a key link to ensure the normal operation of substations.

[0003] Traditional substation equipment inspections mainly rely on manual on-site inspections, supplemented by some portable testing equipment. This inspection method has many technical defects and application drawbacks: First, manual inspections are affected by subjective experience, physical strength, and on-site environment, which can easily lead to missed inspections and misjudgments, and the ability to identify hidden equipment faults is insufficient. Second, the planning of inspection routes lacks scientific basis and relies heavily on the experience of maintenance personnel, resulting in problems such as route redundancy and unreasonable inspection frequency of key equipment, leading to low inspection efficiency. Third, equipment data collection is limited to a single dimension, making it difficult to accurately characterize the equipment's operating status from multiple dimensions. Fourth, fault diagnosis relies heavily on manual analysis of single data points, resulting in low accuracy and efficiency in fault identification.

[0004] Therefore, there is an urgent need for a smart substation equipment status intelligent inspection method that can improve inspection quality and efficiency and reduce the probability of equipment failure expansion, in order to make up for the shortcomings of existing technologies. Summary of the Invention

[0005] Therefore, it is necessary to provide a smart substation equipment status intelligent inspection method and device that can reduce the probability of equipment failure escalation, addressing the aforementioned technical problems.

[0006] Firstly, this application provides a method for intelligent inspection of the status of equipment in a smart substation, the method comprising: Construct a three-dimensional intelligent visual inspection coordinate system, and combine equipment importance level, historical failure rate and inspection scenario to optimize the inspection robot's optimal inspection route, inspection distance and inspection angle, and generate an inspection task list. Acquire equipment image feature data and electrical parameter data synchronously collected by the inspection robot according to the inspection task list, establish a three-dimensional correlation between equipment, image features, and electrical parameters, and form a multi-source dataset; Based on the image feature data and electrical parameter data in the multi-source dataset, the random forest classification model and the improved YOLOv5 recognition model are used for collaborative analysis to output equipment status level information and fault identification and location information, respectively. The equipment status level information and the fault identification and location information are combined to perform collaborative diagnosis and verification of equipment status and fault. Based on collaborative diagnostic results, historical equipment operation data, and real-time collected data, a multi-dimensional early warning model is established, including limit warning, sudden change warning, synchronization warning, and trend warning, to achieve early warning throughout the entire equipment lifecycle.

[0007] In one embodiment, the step of analyzing the image feature data and electrical parameter data from the multi-source dataset using a random forest classification model and an improved YOLOv5 recognition model to output equipment status level information and fault identification and location information, and then combining the equipment status level information and the fault identification and location information to perform collaborative diagnosis and verification of equipment status and faults, includes: The image feature data and electrical parameter data in the multi-source dataset are fused at the feature level to form a multi-dimensional fused feature set; The multi-dimensional fused feature set is input into the trained random forest classification model, and the device's state level information is output. The abnormal region image feature data and the corresponding associated electrical parameter data in the multi-source dataset are input into the improved YOLOv5 recognition model, and the fault identification and location information is output. The YOLOv5 recognition model realizes the identification of fault type and the location of fault by replacing conventional convolution with variable convolution, adding a substation equipment-specific feature attention module, and modifying the lightweight backbone network. The equipment status level information and the fault identification and location information are combined to perform collaborative diagnosis and verification of equipment status and faults.

[0008] In one embodiment, the collaborative diagnostic verification of equipment status and faults by combining the equipment status level information and the fault identification and location information includes: If the equipment status level information output by the random forest classification model matches the fault identification and location information output by the improved YOLOv5 recognition model, the diagnosis result is directly confirmed and the equipment is determined to have a corresponding fault. If the model output results do not match, the historical operating data of the equipment, data of similar fault cases, and equipment location information in the intelligent visual inspection coordinate system are retrieved as supplementary data. The random forest classification model and the improved YOLOv5 recognition model are re-inputted for iterative reasoning. Based on the iterative reasoning results, a comprehensive judgment result of equipment status and fault is generated to complete the collaborative diagnosis verification.

[0009] In one embodiment, the process of retrieving historical operating data of the equipment, data on similar fault cases, and equipment location information in the intelligent visual inspection coordinate system as supplementary data, re-inputting the random forest classification model and the improved YOLOv5 recognition model for iterative inference, and generating a comprehensive judgment result of equipment status and fault based on the iterative inference result to complete the collaborative diagnosis verification includes: Using laser scanning and oblique photography technology, a digital twin of the substation is constructed, and the digital twin is synchronized with the physical substation in real time in terms of operating status and data information. By integrating information on equipment working principles, historical fault cases, operation and maintenance procedures, and defect handling solutions, a substation equipment knowledge graph is constructed. Real-time data in the digital twin is then associated with the substation equipment knowledge graph to achieve rule-based preliminary fault reasoning. Historical equipment operation data, similar fault case data, and equipment location information in the intelligent visual inspection coordinate system are retrieved as supplementary data. The equipment operation status is simulated and deduced in the digital twin to analyze the reasons for the mismatch in the model output results. Combining the supplementary data, simulation results, and inference results, the random forest classification model and the improved YOLOv5 recognition model are re-inputted for iterative inference. Based on the iterative inference results, a comprehensive judgment result of equipment status and fault is generated to complete the collaborative diagnosis verification.

[0010] In one embodiment, generating a comprehensive judgment result of device status and fault based on the iterative reasoning results to complete the collaborative diagnostic verification includes: Based on the iterative equipment status level information output by the random forest classification model and the iterative fault identification and location information output by the improved YOLOv5 recognition model, a comprehensive judgment result of equipment status and fault is generated. The comprehensive judgment result is imported into the digital twin and compared with the virtual operating status of the device to verify whether the fault type and location conform to the physical operating mechanism. Based on the data of similar fault cases, the validity of the comprehensive judgment result is verified.

[0011] In one embodiment, the construction of a three-dimensional intelligent visual inspection coordinate system, combined with equipment importance level, historical failure rate, and inspection scenario, optimizes the inspection robot's optimal inspection route, inspection distance, and inspection angle, generating an inspection task list including: A three-dimensional intelligent visual inspection coordinate system is constructed with the geographical center point of the substation as the origin. Visual constraint parameters are added to the intelligent visual inspection coordinate system, including the optimal inspection distance threshold and the unobstructed inspection angle domain. Acquire the three-dimensional coordinates of all equipment in the substation, the equipment outline dimensions, and the spatial coordinates of key monitoring points to form a substation equipment spatial database; The substation equipment spatial database, the visual constraint parameters, the equipment importance level, historical failure rate data, and the inspection scenario are input into the improved genetic algorithm, which outputs the optimal inspection route, inspection distance, and inspection angle for the inspection robot. Based on the output of the genetic algorithm, an inspection task list is generated.

[0012] In one embodiment, the step of inputting the substation equipment spatial database, the visual constraint parameters, equipment importance level, historical failure rate data, and inspection scenario into an improved genetic algorithm to output the optimal inspection route, inspection distance, and inspection angle for the inspection robot includes: Based on the three-dimensional chromosome coding rules, the inspection route, inspection distance, and inspection angle are mapped to independent gene segments of the chromosome, and an initial population is constructed based on the independent gene segments and the visual constraint parameters. Differentiated weights are assigned to each device based on its importance level and historical failure rate data, and optimization constraints for algorithm iteration are set based on environmental obstacles and robot movement range in the inspection scenario. A multi-objective fitness function is constructed that integrates patrol route length, visual constraint matching degree, and robot energy consumption. The initial population, the differentiated weights, the optimization constraints, and the multi-objective fitness function are input into an improved genetic algorithm, and iterative calculations are completed through adaptive crossover mutation and elite retention strategies. When the algorithm iterates to the convergence condition, it outputs the global optimal solution, obtaining the optimal inspection route, inspection distance, and inspection angle of the inspection robot.

[0013] In one embodiment, the step of inputting the initial population, the differential weights, the optimization constraints, and the multi-objective fitness function into the improved genetic algorithm, and completing iterative calculations through adaptive crossover mutation and elite retention strategies, includes: The differentiated weights are assigned to the corresponding evaluation index of the multi-objective fitness function, and the encoding value of each chromosome in the initial population is optimized and the constraint conditions are checked to remove invalid chromosomes that have route collisions, distances or angles that exceed the preset visual constraint range. The processed initial population and the multi-objective fitness function are input into the improved genetic algorithm. The crossover probability is dynamically adjusted according to the iteration process to perform adaptive crossover operation on the same type of gene segments in the chromosome with inspection routes, inspection distances, and inspection angles. Mutation operation is then performed on the population after crossover. After crossover and mutation, the population is screened for a new generation through an elite retention strategy. The fitness calculation, adaptive crossover, mutation, and elite retention operations are repeated until the algorithm iterates to the convergence condition.

[0014] Secondly, this application also provides an intelligent inspection device for the status of equipment in a smart substation, the device comprising: The task list generation module is used to construct an intelligent visual inspection coordinate system based on three-dimensional space. Combining the equipment importance level, historical failure rate and inspection scenario, it optimizes the inspection robot's optimal inspection route, inspection distance and inspection angle, and generates an inspection task list. The data fusion module is used to acquire equipment image feature data and electrical parameter data synchronously collected by the inspection robot according to the inspection task list, establish a three-dimensional correlation between equipment, image features, and electrical parameters, and form a multi-source dataset; The collaborative diagnosis and verification module is used to analyze the image feature data and electrical parameter data in the multi-source dataset through a random forest classification model and an improved YOLOv5 recognition model, and output equipment status level information and fault identification and location information respectively. The module then combines the equipment status level information and the fault identification and location information to perform collaborative diagnosis and verification of equipment status and faults. The multi-dimensional early warning module is used to establish a multi-dimensional early warning model based on collaborative diagnostic results, historical equipment operation data, and real-time collected data, including limit warning, sudden change warning, synchronization warning, and trend warning, to achieve early warning throughout the entire equipment lifecycle.

[0015] In summary, this application includes the following beneficial technical effects: By constructing a three-dimensional intelligent visual inspection coordinate system, and combining equipment importance level, historical failure rate, and inspection scenario, the inspection route, inspection distance, and inspection angle are optimized. This avoids problems such as blurred image features and electrical data sensing deviations caused by improper acquisition angles, excessively far or too close distances, significantly improving the completeness and effectiveness of data acquisition. Establishing a three-dimensional correlation between equipment, image features, and electrical parameters breaks through the limitations of traditional single-dimensional data, comprehensively reflecting the actual operating status of equipment from both appearance and operating parameter perspectives, avoiding misjudgments due to incomplete data. Through collaborative analysis of a random forest classification model and an improved YOLOv5 recognition model, equipment status level information and fault identification and location information are output respectively, and collaborative diagnostic verification is performed. This effectively solves the problems of missed detections and misjudgments caused by single models, improving the accuracy of fault diagnosis. Based on the collaborative diagnostic results, historical equipment operating data, and real-time acquired data, a multi-dimensional early warning model is established, transforming post-event maintenance into pre-event early warning, significantly reducing the probability of equipment failure escalation. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating an intelligent inspection method for the status of equipment in a smart substation, as shown in one embodiment. Figure 2 This is a flowchart illustrating the intelligent inspection method for the status of smart substation equipment in another embodiment; Figure 3 This is a structural block diagram of an intelligent substation equipment status inspection device in one embodiment. Detailed Implementation

[0017] This invention provides a method and apparatus for intelligent inspection of equipment status in smart substations.

[0018] The embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0019] In the description of the embodiments disclosed in this invention, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0020] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the intelligent inspection method for the status of smart substation equipment in this invention includes: S100 constructs an intelligent visual inspection coordinate system based on three-dimensional space. Combining the equipment importance level, historical failure rate, and inspection scenario, it optimizes the inspection robot's optimal inspection route, inspection distance, and inspection angle, and generates an inspection task list.

[0021] Specifically, firstly, a three-dimensional intelligent visual inspection coordinate system is established with the geographical center of the substation as the origin. This coordinate system adopts a right-handed Cartesian coordinate structure, which can accurately represent all physical spatial locations within the substation, including the three-dimensional installation coordinates of equipment and the spatial locations of key monitoring points, providing a unified spatial benchmark for the positioning of subsequent inspection paths and data collection parameters. Secondly, combining the actual characteristics and inspection needs of substation equipment, the system integrates equipment importance levels, historical failure rates, and inspection scenario information. Intelligent algorithms are used to optimize the design of the inspection robot's optimal inspection route, inspection distance, and inspection angle. The optimal inspection route must meet the requirements of no redundant backtracking, no collisions, and coverage of all equipment to be inspected, while prioritizing the inspection of core equipment and equipment with high failure rates. The inspection distance and inspection angle must be adapted to the equipment's visual acquisition requirements, ensuring that the acquired image features are clearly visible, providing high-quality data for subsequent fault identification. Finally, a standardized inspection task list is generated based on the optimization results. The list includes the inspection equipment, inspection sequence, inspection time, inspection distance, inspection angle, and the types of data to be collected.

[0022] S200 acquires equipment image feature data and electrical parameter data synchronously collected by the inspection robot according to the inspection task list, establishes a three-dimensional correlation between equipment, image features, and electrical parameters, and forms a multi-source dataset.

[0023] Specifically, the inspection robot is equipped with a high-definition vision acquisition module and multiple types of electrical parameter sensing modules, and performs automated inspection operations according to a preset inspection task list. During the inspection, the robot strictly follows the route, distance, and angle requirements in the task list, and simultaneously collects image feature data and electrical parameter data of the equipment at the same inspection time and the same equipment monitoring point. The image feature data includes visual information such as the equipment's appearance outline, surface texture, and component installation status, which can intuitively reflect the physical state of the equipment. The electrical parameter data includes core operating parameters such as voltage, current, and temperature during equipment operation, which can quantitatively characterize the electrical operating status of the equipment. After the data is collected, a three-dimensional correlation relationship of "equipment-image feature-electrical parameter" is established using the equipment point in the three-dimensional intelligent visual inspection coordinate system as a unique index. That is, the visual data and electrical data of the same equipment and the same monitoring point are accurately bound together, completing the initial fusion processing of multi-source data. This correlation method breaks the limitations of separate acquisition and independent storage of image and electrical data in traditional inspections, ensuring the consistency of the two types of data in spatial and equipment dimensions, and ultimately forming a structured and correlated multi-source dataset, providing comprehensive and coherent basic data support for subsequent equipment status analysis and fault diagnosis.

[0024] S300, based on image feature data and electrical parameter data from multi-source datasets, uses a random forest classification model and an improved YOLOv5 recognition model for collaborative analysis, outputting equipment status level information and fault identification and location information respectively, and combines the equipment status level information and fault identification and location information to perform collaborative diagnosis and verification of equipment status and faults.

[0025] Specifically, the random forest classification model focuses on determining the overall operational status of equipment. By deeply mining and analyzing feature information from multi-source datasets, it outputs equipment status level information, such as good, average, and abnormal, providing a macro-level basis for judging whether there are operational risks. The improved YOLOv5 recognition model focuses on the precise location and type identification of faults. It focuses on analyzing abnormal feature regions in the dataset and outputs specific fault identification and location information, including the fault type and the specific location of the fault, providing precise guidance for fault handling. After the two models output their results, a collaborative diagnosis and verification of equipment status and faults is performed. By associating and matching the equipment status level information with the fault identification and location information, the consistency and rationality of the two types of results are verified, ensuring the reliability of the diagnostic results and avoiding misjudgments and omissions that may be caused by analysis of a single model.

[0026] S400, based on collaborative diagnostic results, historical equipment operation data, and real-time collected data, establishes a multi-dimensional early warning model including limit warning, sudden change warning, synchronization warning, and trend warning, realizing early warning throughout the entire equipment lifecycle.

[0027] Specifically, based on the collaborative diagnostic results of the model, historical operating data and real-time collected data of the equipment are integrated to establish a multi-dimensional early warning model, including limit warning, mutation warning, synchronization warning, and trend warning. Among them, the limit warning monitors the safety threshold of the equipment operating parameters and triggers an early warning when the real-time parameters exceed the preset safety range; the mutation warning focuses on the abnormal fluctuation of parameters within a unit of time and issues a warning when the rate of parameter change exceeds the normal range; the synchronization warning monitors the coordinated occurrence of abnormal visual characteristics and abnormal electrical parameters of the equipment and captures the associated signals of fault occurrence; the trend warning predicts the future operating status of the equipment by analyzing the time-series change patterns of parameters and identifies potential fault risks in advance. The limit warning system breaks through the traditional fixed threshold mode. Based on the equipment's rated parameters and industry standards, it dynamically adjusts the warning threshold according to the equipment's years of operation, aging coefficient, and real-time operating conditions. For example, the oil temperature warning threshold for newly commissioned main transformers is set at ±10% of the rated parameters, while for equipment that has been in operation for more than 5 years, it is adjusted to ±15% based on the aging coefficient. In high-temperature environments, it is further relaxed by 2%-3%. At the same time, the threshold is divided into three levels: safety, warning, and emergency, corresponding to different response strategies. By comparing the warning results with the actual state, the threshold correction coefficient is automatically adjusted to achieve closed-loop optimization. The sudden change warning adopts a fusion model of the first-order difference method, the sliding window variance method, and the isolated forest algorithm. It dynamically allocates weights according to equipment type to identify abnormal parameter fluctuations, automatically associates environmental data and equipment operation records to determine the type of sudden change, eliminates false warnings caused by external interference, and sets response priorities according to the magnitude of the sudden change. To ensure rapid response to emergency mutations, a synchronous early warning system is constructed based on equipment failure mechanisms and historical cases, building a cross-dimensional anomaly association rule base. The system dynamically sets the time window and association confidence threshold for synchronous early warnings using the Pearson correlation coefficient. Early warnings are triggered only when visual and electrical anomalies within the same time window meet the confidence requirements. Upon triggering, the system retrieves the digital twin 3D model, overlays the anomaly data, and generates a source tracing report to assist in fault location. A trend early warning system constructs a time-series prediction model integrating Long Short-Term Memory (LSTM) networks and attention mechanisms. Focusing on the dimensions of strong fault correlation, it predicts parameter changes and health decay trends over the next 30 and 90 days. Combined with digital twin simulation technology, it simulates the fault evolution path, outputting the possible consequences and impact range of the fault. Early warnings are categorized into three levels: mild, moderate, and severe, based on the predicted risk level. The prediction results are updated every 7 days based on the latest data to ensure timely warnings.

[0028] In one embodiment, such as Figure 2 As shown, S300 includes: S310 performs feature-level fusion of image feature data and electrical parameter data from multiple source datasets to form a multi-dimensional fused feature set; S320 inputs a multi-dimensional fused feature set into a trained random forest classification model and outputs the device's state level information. S330 inputs the abnormal region image feature data and the corresponding associated electrical parameter data from the multi-source dataset into the improved YOLOv5 recognition model and outputs fault identification and location information. S340 combines equipment status level information with fault identification and location information to perform collaborative diagnosis and verification of equipment status and faults.

[0029] Specifically, the heterogeneous image feature data and electrical parameter data from the multi-source dataset are first preprocessed to eliminate data noise and dimensionality differences. A feature-level fusion strategy is then used to deeply fuse the two types of data, forming a multi-dimensional fused feature set that simultaneously includes visual and electrical operation features. This provides comprehensive and unified feature input for subsequent model analysis. The multi-dimensional fused feature set is then input into a trained random forest classification model. This model is trained and optimized using the actual operating status of the equipment as a label. Through multi-decision tree hierarchical judgment and weighted voting, it accurately outputs the equipment's status level information, achieving a comprehensive assessment of the equipment's overall operating status. The quantitative evaluation is performed; for equipment areas in multi-source datasets that are identified as abnormal by the random forest classification model, the corresponding abnormal area image feature data and associated electrical parameter data are extracted and input into the improved YOLOv5 recognition model. This improved YOLOv5 recognition model is not a general object detection model, but a customized optimization based on the specific needs of substation equipment fault identification. First, it replaces the conventional convolution with a variable convolution, which can identify the irregular shape and fault characteristics (such as leakage marks, rust spots, loose parts) of different substation equipment (such as main transformers, switchgear, insulators, etc.). By leveraging the random distribution characteristics of faults (such as dynamic displacement), the sampling position and receptive field size of the convolution kernel are dynamically adjusted to accurately capture subtle fault features of different scales and shapes, avoiding the problem of insufficient extraction of irregular fault features by conventional convolution. Secondly, the model incorporates a substation equipment-specific feature attention module. After training with a large number of substation equipment fault samples, this module automatically focuses on high-frequency fault areas, strengthening the information weight of fault features in these areas while weakening the interference of irrelevant factors such as equipment background, changes in ambient lighting, and shooting angle deviations, significantly improving the recognition accuracy of small and weak-feature faults. In addition, by lightweighting the backbone network of the model, while ensuring that the fault identification accuracy is not reduced, the network channels are pruned and optimized and the feature fusion process is simplified, reducing the computational load and parameter count of the model, thus meeting the real-time requirements of fault identification in inspection operations. The equipment status level information output by the random forest classification model is correlated and matched with the fault identification and location information output by the improved YOLOv5 recognition model to initiate a collaborative diagnosis and verification process for equipment status and fault. If the two results match, the diagnosis result is directly confirmed; if the results do not match, the process of supplementing data and iterative reasoning is initiated.

[0030] In one embodiment, the collaborative diagnostic verification of equipment status and faults by combining equipment status level information and fault identification and location information includes: If the equipment status level information output by the random forest classification model matches the fault identification and location information output by the improved YOLOv5 recognition model, the diagnosis result is directly confirmed, and the equipment is determined to have a corresponding fault. If the model output results do not match, historical operating data of the equipment, data of similar fault cases, and equipment location information in the intelligent visual inspection coordinate system are retrieved as supplementary data. These are then re-input into the random forest classification model and the improved YOLOv5 recognition model for iterative reasoning. Based on the iterative reasoning results, a comprehensive judgment result of equipment status and fault is generated, completing the collaborative diagnosis verification.

[0031] Specifically, if the device status level information output by the random forest model matches the fault identification and location information output by the improved YOLOv5 recognition model, that is, the device status level is abnormal and the model accurately identifies the corresponding fault type and location, the diagnosis result is directly confirmed, and the device is determined to have a corresponding fault. If the model output results do not match, such as the device status level is abnormal but no specific fault is identified, the fault is identified but the device status level is normal, or the fault type and status level do not match, the supplementary data iterative reasoning process is initiated. First, the historical operating data of the device, the data of similar fault cases, and the device location information in the intelligent visual inspection coordinate system are retrieved as supplementary data. Then, the supplementary data and the original multi-source dataset are re-input into the random forest classification model and the improved YOLOv5 recognition model for iterative reasoning. Finally, based on the results of the model iterative reasoning, a comprehensive judgment result of the device status and fault is generated, completing this collaborative diagnosis verification.

[0032] In one embodiment, historical equipment operation data, similar fault case data, and equipment location information in the intelligent visual inspection coordinate system are retrieved as supplementary data. These are then re-input into the random forest classification model and the improved YOLOv5 recognition model for iterative inference. Based on the iterative inference results, a comprehensive judgment result of equipment status and fault is generated, completing the collaborative diagnostic verification, including: Using laser scanning and oblique photography technologies, a digital twin of the substation is constructed, with its operating status and data synchronized in real time with the physical substation. Information on equipment operating principles, historical fault cases, maintenance procedures, and defect handling schemes is integrated to build a substation equipment knowledge graph. Real-time data from the digital twin is then linked to this knowledge graph to achieve rule-based preliminary fault reasoning. Supplementary data includes historical equipment operating data, similar fault case data, and equipment location information from the intelligent visual inspection coordinate system. Equipment operating status simulations are then performed within the digital twin to analyze the reasons for mismatches in model output results. Combining the supplementary data, simulation results, and reasoning results, the random forest classification model and the improved YOLOv5 recognition model are re-inputted for iterative reasoning. Based on the iterative reasoning results, a comprehensive judgment result on equipment status and faults is generated, completing collaborative diagnostic verification.

[0033] Specifically, using laser scanning and oblique photography technologies, a digital twin is constructed that maps to the physical substation at a 1:1 scale. This enables real-time synchronization of the digital twin's operating status and data information with the physical substation, providing a virtual platform for equipment operation status simulation. Multi-dimensional information, including equipment working principles, historical fault cases, maintenance procedures, and defect handling solutions, is integrated to construct a standardized substation equipment knowledge graph. Real-time operating data from the digital twin is then fused with the knowledge graph to achieve rule-based preliminary fault reasoning. For example, when the partial discharge of a device in the digital twin exceeds the standard, the association rule "abnormal partial discharge - insulation aging - handling solution" in the knowledge graph is automatically applied to initially identify the possible fault type and handling direction. When the output results of the random forest classification model and the improved YOLOv5 recognition model do not match, historical operating data, similar fault case data, and equipment location information from the intelligent visual inspection coordinate system are retrieved as supplementary data. This supplementary dataset, along with the initial input data of the two models, is imported into the digital twin. Based on the high-fidelity virtual model, the current operating condition of the equipment is reconstructed, and the specific reasons for the mismatch in the output results of the two models are analyzed. Examples of potential causes include internal equipment faults, image acquisition angle obstruction, sensor data errors, and environmental interference. Finally, supplementary data, simulation results from the digital twin (including bias analysis), and preliminary inference conclusions from the knowledge graph are fused together to form an iterative inference dataset that is both complete and targeted. This dataset is then re-input into the random forest classification model and the improved YOLOv5 recognition model for secondary inference. For the random forest classification model, supplementary data, parameter correction values ​​verified by simulation, and fault association features from the inference results are integrated into the original dataset. Multi-decision tree hierarchical judgment and weighted voting are then performed again to correct and optimize the initial output equipment status level information, making the status level judgment more consistent with the actual operating conditions of the equipment. For the improved YOLOv5 recognition model, the presentation of fault features clearly defined in the simulation and the fault type indications in the inference results are combined to perform targeted feature enhancement and interference removal on the image feature data of abnormal areas. Simultaneously, the associated electrical parameter anomaly threshold is used as an auxiliary identification basis to optimize the model's accuracy in matching fault types and spatial positioning of fault locations, achieving iterative upgrades to fault identification and positioning information.

[0034] In one embodiment, based on the results of iterative reasoning, a comprehensive judgment result of equipment status and fault is generated to complete the collaborative diagnostic verification, including: Based on the iterative equipment status level information output by the random forest classification model and the iterative fault identification and location information output by the improved YOLOv5 recognition model, a comprehensive judgment result of equipment status and fault is generated. The comprehensive judgment result is imported into the digital twin and compared with the virtual operating status of the equipment to verify whether the fault type and location conform to the physical operating mechanism. Based on the data of similar fault cases, the effectiveness of the comprehensive judgment result is verified.

[0035] Specifically, after iterative inference, the iterative equipment status level information output by the random forest classification model and the iterative fault identification and location information output by the improved YOLOv5 recognition model are obtained. A comprehensive judgment result is generated according to the "status-fault correlation fusion" logic. To ensure the scientific validity and rationality of the comprehensive judgment result, a dual verification process is implemented. First, the comprehensive judgment result is imported into the substation's digital twin. After importation, it automatically locates the corresponding equipment and fault area, reconstructs the current operating conditions (such as load rate, voltage and current parameters, and ambient temperature and humidity), and simulates the fault based on the equipment's structural design parameters, working principle, and mechanical and electrical characteristics. The changes in operating status corresponding to fault type and location are used to verify whether the fault type and location conform to the physical operating mechanism of the equipment (e.g., whether a leakage fault matches the oil circuit route, and whether a damage fault is located in a weak stress area). The consistency between the comprehensive judgment result and the simulation result is compared. Subsequently, a secondary verification is carried out based on data from similar fault cases. Fault cases with the same equipment type, similar operating years, and similar operating conditions are retrieved from the operation and maintenance database. The matching degree of the comprehensive judgment result with the fault type, location, and characteristic manifestations in the cases is compared. If the matching degree reaches the preset standard (e.g., the core feature matching degree ≥ 80%), the comprehensive judgment result is confirmed to be valid. Through the verification of the physical operating mechanism of the digital twin and the validity verification of data from similar fault cases, the collaborative diagnosis and verification of equipment status and faults are finally completed, ensuring that the output results are accurate and reliable, and providing authoritative decision-making basis for subsequent operation and maintenance.

[0036] In one embodiment, a three-dimensional intelligent visual inspection coordinate system is constructed. Combining equipment importance level, historical failure rate, and inspection scenario, the optimal inspection route, inspection distance, and inspection angle of the inspection robot are optimized, generating an inspection task list including: Using the geographical center point of the substation as the origin, a three-dimensional intelligent visual inspection coordinate system is constructed, and visual constraint parameters are added to the intelligent visual inspection coordinate system. The visual constraint parameters include the optimal inspection distance threshold and the unobstructed inspection angle domain. The three-dimensional coordinates, equipment outline dimensions, and spatial coordinates of key monitoring points of all equipment in the substation are obtained to form a substation equipment spatial database. The substation equipment spatial database, visual constraint parameters, equipment importance level, historical failure rate data, and inspection scenario are input into an improved genetic algorithm to output the optimal inspection route, inspection distance, and inspection angle of the inspection robot. Based on the output results of the genetic algorithm, an inspection task list is generated.

[0037] Specifically, a right-handed Cartesian three-dimensional intelligent visual inspection coordinate system is constructed with the geographical center of the substation as the origin. Visual constraint parameters are added to this coordinate system, including the optimal inspection distance threshold and the unobstructed inspection angle domain. These parameters are calibrated based on the visual acquisition requirements of different types of equipment and the substation's on-site environment. Through LiDAR scanning and high-precision mapping technology, the three-dimensional coordinates, equipment outline dimensions, and spatial coordinates of key monitoring points of all equipment within the substation are acquired, constructing a substation equipment spatial database. Then, the substation equipment spatial database and the aforementioned visual constraint parameters are used as basic inputs, while also incorporating equipment importance levels based on equipment functional positioning and maintenance priorities, and statistical data on equipment performance. Historical failure rate data on the frequency and type of failures during the operational cycle, along with inspection scenario information covering the characteristics of indoor and outdoor areas of the substation, the distribution of environmental obstacles, and the limitations of the inspection robot's motion performance, are input into an improved genetic algorithm for multi-objective optimization calculations. The algorithm combines various input information to coordinate and output the globally optimal inspection route for the inspection robot, the precise inspection distance for each piece of equipment, and the optimal visual acquisition angle for each key monitoring point, while satisfying visual constraints and avoiding operational obstacles. Finally, based on the globally optimal solution output by the improved genetic algorithm, a standardized inspection task list is generated, providing clear execution instructions for the automated inspection of the inspection robot.

[0038] In one embodiment, an improved genetic algorithm is input using a substation equipment spatial database, visual constraint parameters, equipment importance levels, historical failure rate data, and inspection scenario data. The output of the optimal inspection route, inspection distance, and inspection angle for the inspection robot includes: Based on the three-dimensional chromosome coding rules, the inspection route, inspection distance, and inspection angle are mapped to independent gene segments of the chromosome, and an initial population is constructed based on the independent gene segments and visual constraint parameters. Differentiated weights are assigned to each device according to the importance level and historical failure rate data, and optimization constraints for algorithm iteration are set according to environmental obstacles and robot movement range in the inspection scenario. A multi-objective fitness function integrating inspection route length, visual constraint matching degree, and robot energy consumption is constructed, and the initial population, differentiated weights, optimization constraints, and multi-objective fitness function are input into the improved genetic algorithm. Iterative calculations are completed through adaptive crossover mutation and elite retention strategies. When the algorithm iterates to the convergence condition, the global optimal solution is output, obtaining the optimal inspection route, inspection distance, and inspection angle of the inspection robot.

[0039] Specifically, based on three-dimensional chromosome coding rules, the three core inspection parameters—inspection route, inspection distance, and inspection angle—are mapped to independent gene segments of the chromosome to avoid parameter coupling interference. Then, based on each independent gene segment and visual constraint parameters, an initial population conforming to the substation inspection scenario is constructed to provide basic samples for algorithm iteration. Differential optimization weights are assigned to each device according to its importance level and historical failure rate data, with core devices and high-failure-rate devices given higher weights to ensure their inspection priority. Simultaneously, optimization constraints for algorithm iteration are set based on the distribution of environmental obstacles in the inspection scenario and the inspection robot's movement range (such as maximum endurance and gimbal adjustment angle), eliminating obstacles... Excluding invalid solutions, a multi-objective fitness function is constructed that integrates the inspection route length, visual constraint matching degree, and robot energy consumption. This function is used as the evaluation basis for algorithm iteration to achieve coordinated optimization of inspection efficiency, data collection quality, and robot operating cost, avoiding the disadvantages caused by single-objective optimization. The initial population, differentiated weights, optimization constraints, and multi-objective fitness function are input into the improved genetic algorithm. Iterative calculations are completed through adaptive crossover mutation and elite retention strategies. When the algorithm iterates to the convergence condition (the number of iterations reaches a preset value or the fitness function value tends to stabilize), the global optimal solution is output, obtaining the optimal inspection route, inspection distance, and inspection angle of the inspection robot.

[0040] In one embodiment, the initial population, differential weights, optimization constraints, and multi-objective fitness function are input into the improved genetic algorithm, and iterative computation is completed through adaptive crossover and mutation and elite retention strategies, including: Differential weights are assigned to the corresponding evaluation index of the multi-objective fitness function, and the encoding values ​​of each chromosome in the initial population are optimized and the constraints are checked to remove invalid chromosomes that collide with routes or whose distances or angles exceed the preset visual constraint range. The processed initial population and the multi-objective fitness function are input into the improved genetic algorithm, and the crossover probability is dynamically adjusted according to the iteration process to perform adaptive crossover operations on the same type of gene segments in the chromosomes for inspection routes, inspection distances, and inspection angles. Mutation operations are then performed on the population after crossover. The population after crossover and mutation is then screened for a new generation of population through an elite retention strategy, and the fitness calculation, adaptive crossover, mutation, and elite retention operations are repeated until the algorithm iterates to the convergence condition.

[0041] Specifically, differentiated weights are assigned to the corresponding evaluation metrics of the multi-objective fitness function, tilting the algorithm iteration towards core equipment and high-failure-rate equipment. Simultaneously, the encoding values ​​of each chromosome in the initial population are optimized and constrained, eliminating invalid chromosomes due to route collisions, distances, or angles exceeding preset visual constraint ranges, retaining valid samples, and improving iteration efficiency. The processed initial population and multi-objective fitness function are input into the improved genetic algorithm, and the crossover probability is dynamically adjusted according to the algorithm iteration process. A high crossover probability is used in the early stages of iteration to expand the solution space search range, while a low crossover probability is used in the later stages to ensure convergence stability. Only [the algorithm] executes [the following steps]. The algorithm employs adaptive crossover operations on gene segments of the same type with the same inspection route, inspection distance, and inspection angle in the chromosome to avoid invalid crossovers of gene segments with different parameters. After crossover, a mutation operation is performed on the population to introduce random perturbations and prevent the algorithm from getting trapped in local optima. Then, an elite retention strategy is used to select a new generation of the population, directly retaining high-quality individuals with high fitness values ​​in each generation to ensure that the optimal solution is not lost. The fitness calculation, adaptive crossover, mutation, and elite retention operations are repeated. In each iteration, the new population is optimized and the constraints are checked, and invalid solutions are eliminated, until the algorithm iterates to the preset convergence condition, completes the optimization calculation of the inspection parameters, and outputs the global optimal solution.

[0042] In one embodiment, such as Figure 3 As shown, a smart substation equipment status intelligent inspection device is provided, including: a task list generation module 10, a data fusion module 20, a collaborative diagnosis and verification module 30, and a multi-dimensional early warning module 40, wherein: The task list generation module 10 is used to construct an intelligent visual inspection coordinate system based on three-dimensional space. Combining the equipment importance level, historical failure rate and inspection scenario, it optimizes the inspection robot's optimal inspection route, inspection distance and inspection angle to generate an inspection task list. The data fusion module 20 is used to acquire equipment image feature data and electrical parameter data synchronously collected by the inspection robot according to the inspection task list, establish a three-dimensional correlation between equipment, image features, and electrical parameters, and form a multi-source dataset. The collaborative diagnosis and verification module 30 is used to analyze the image feature data and electrical parameter data in the multi-source dataset through the collaborative analysis of the random forest classification model and the improved YOLOv5 recognition model, and output the equipment status level information and fault identification and location information respectively, and perform collaborative diagnosis and verification of equipment status and fault by combining the equipment status level information and fault identification and location information. The multi-dimensional early warning module 40 is used to establish a multi-dimensional early warning model based on collaborative diagnostic results, historical equipment operation data, and real-time collected data, including limit warning, sudden change warning, synchronous warning, and trend warning, to achieve early warning throughout the entire equipment lifecycle.

[0043] In one embodiment, the collaborative diagnostic verification module 30 is further configured to perform feature-level fusion of image feature data and electrical parameter data from multi-source datasets to form a multi-dimensional fusion feature set; input the multi-dimensional fusion feature set into a trained random forest classification model to output equipment status level information; input abnormal region image feature data and corresponding associated electrical parameter data from multi-source datasets into an improved YOLOv5 recognition model to output fault identification and location information, wherein the YOLOv5 recognition model achieves fault type identification and fault location location by replacing conventional convolution with variable convolution, adding a substation equipment-specific feature attention module, and modifying the lightweight backbone network; and perform collaborative diagnostic verification of equipment status and fault by combining equipment status level information and fault identification and location information.

[0044] In one embodiment, the collaborative diagnosis verification module 30 is further configured to: if the equipment status level information output by the random forest classification model matches the fault identification and location information output by the improved YOLOv5 recognition model, then directly confirm the diagnosis result and determine that the equipment has a corresponding fault; if the model output results do not match, then retrieve the equipment's historical operating data, similar fault case data, and equipment location information in the intelligent visual inspection coordinate system as supplementary data, re-input the random forest classification model and the improved YOLOv5 recognition model for iterative reasoning, and based on the iterative reasoning results, generate a comprehensive judgment result of equipment status and fault, and complete the collaborative diagnosis verification.

[0045] In one embodiment, the collaborative diagnosis and verification module 30 is further used to construct a digital twin of the substation using laser scanning and oblique photography technology. The digital twin is synchronized with the physical substation in real time with its operating status and data information. It integrates equipment working principles, historical fault cases, operation and maintenance procedures, and defect handling scheme information to construct a substation equipment knowledge graph. It also associates the real-time data in the digital twin with the substation equipment knowledge graph to achieve rule-based preliminary fault reasoning. It retrieves historical equipment operating data, similar fault case data, and equipment location information in the intelligent visual inspection coordinate system as supplementary data, and performs equipment operating status simulation in the digital twin to analyze the reasons for the mismatch in model output results. Combining the supplementary data, simulation results, and reasoning results, it re-inputs the random forest classification model and the improved YOLOv5 recognition model for iterative reasoning. Based on the iterative reasoning results, it generates a comprehensive judgment result of equipment status and fault, completing the collaborative diagnosis and verification.

[0046] In one embodiment, the collaborative diagnostic verification module 30 is further configured to generate a comprehensive judgment result of equipment status and fault based on the iterative equipment status level information output by the random forest classification model and the iterative fault identification and location information output by the improved YOLOv5 recognition model; import the comprehensive judgment result into the digital twin, compare it with the virtual operating status of the equipment, verify whether the fault type and location conform to the physical operating mechanism, and verify whether the comprehensive judgment result is effective based on similar fault case data.

[0047] In one embodiment, the task list generation module 10 is further configured to construct a three-dimensional intelligent visual inspection coordinate system with the geographical center point of the substation as the origin, and add visual constraint parameters to the intelligent visual inspection coordinate system, including the optimal inspection distance threshold and the unobstructed inspection angle domain; acquire the three-dimensional coordinates of all equipment in the substation, the equipment outline dimensions, and the spatial coordinates of key monitoring points to form a substation equipment spatial database; input the substation equipment spatial database, visual constraint parameters, equipment importance level, historical failure rate data, and inspection scenario into an improved genetic algorithm to output the optimal inspection route, inspection distance, and inspection angle of the inspection robot; and generate an inspection task list based on the output of the genetic algorithm.

[0048] In one embodiment, the task list generation module 10 is further configured to map the inspection route, inspection distance, and inspection angle to independent gene segments of the chromosome based on the three-dimensional chromosome coding rules, and construct an initial population based on the independent gene segments and visual constraint parameters; assign differentiated weights to each device according to the device importance level and historical failure rate data, and set optimization constraints for algorithm iteration based on environmental obstacles and robot movement range in the inspection scenario; construct a multi-objective fitness function that integrates inspection route length, visual constraint matching degree, and robot energy consumption, and input the initial population, differentiated weights, optimization constraints, and multi-objective fitness function into the improved genetic algorithm, and complete the iterative calculation through adaptive crossover mutation and elite retention strategies; when the algorithm iterates to the convergence condition, output the global optimal solution to obtain the optimal inspection route, inspection distance, and inspection angle of the inspection robot.

[0049] In one embodiment, the task list generation module 10 is further configured to assign differentiated weights to the corresponding evaluation index of the multi-objective fitness function, and to perform optimization constraint verification on the encoding value of each chromosome in the initial population, eliminating invalid chromosomes that have route collisions, distances, or angles exceeding the preset visual constraint range; input the processed initial population and the multi-objective fitness function into the improved genetic algorithm, dynamically adjust the crossover probability according to the iteration process, perform adaptive crossover operations on the same type of gene segments in the chromosomes for inspection routes, inspection distances, and inspection angles, and perform mutation operations on the population after crossover; and select a new generation of population through an elite retention strategy, and repeat the fitness calculation, adaptive crossover, mutation, and elite retention operations until the algorithm iterates to the convergence condition.

[0050] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for intelligent inspection of equipment status in a smart substation, characterized in that, include: Construct a three-dimensional intelligent visual inspection coordinate system, and combine equipment importance level, historical failure rate and inspection scenario to optimize the inspection robot's optimal inspection route, inspection distance and inspection angle, and generate an inspection task list. Acquire equipment image feature data and electrical parameter data synchronously collected by the inspection robot according to the inspection task list, establish a three-dimensional correlation between equipment, image features, and electrical parameters, and form a multi-source dataset; Based on the image feature data and electrical parameter data in the multi-source dataset, the random forest classification model and the improved YOLOv5 recognition model are used for collaborative analysis to output equipment status level information and fault identification and location information, respectively. The equipment status level information and the fault identification and location information are combined to perform collaborative diagnosis and verification of equipment status and fault. Based on collaborative diagnostic results, historical equipment operation data, and real-time collected data, a multi-dimensional early warning model is established, including limit warning, sudden change warning, synchronization warning, and trend warning, to achieve early warning throughout the entire equipment lifecycle.

2. The intelligent inspection method for the status of equipment in a smart substation according to claim 1, characterized in that, The step of analyzing image feature data and electrical parameter data from the multi-source dataset using a random forest classification model and an improved YOLOv5 recognition model to output equipment status level information and fault identification and location information, and then performing collaborative diagnosis and verification of equipment status and faults by combining the equipment status level information and the fault identification and location information, includes: The image feature data and electrical parameter data in the multi-source dataset are fused at the feature level to form a multi-dimensional fused feature set; The multi-dimensional fused feature set is input into the trained random forest classification model, and the device's state level information is output. The abnormal region image feature data and the corresponding associated electrical parameter data in the multi-source dataset are input into the improved YOLOv5 recognition model, and the fault identification and location information is output. The YOLOv5 recognition model realizes the identification of fault type and the location of fault by replacing conventional convolution with variable convolution, adding a substation equipment-specific feature attention module, and modifying the lightweight backbone network. The equipment status level information and the fault identification and location information are combined to perform collaborative diagnosis and verification of equipment status and faults.

3. The intelligent inspection method for the status of equipment in a smart substation according to claim 2, characterized in that, The collaborative diagnosis and verification of equipment status and faults by combining the equipment status level information and the fault identification and location information includes: If the equipment status level information output by the random forest classification model matches the fault identification and location information output by the improved YOLOv5 recognition model, the diagnosis result is directly confirmed and the equipment is determined to have a corresponding fault. If the model output results do not match, the historical operating data of the equipment, data of similar fault cases, and equipment location information in the intelligent visual inspection coordinate system are retrieved as supplementary data. The random forest classification model and the improved YOLOv5 recognition model are re-inputted for iterative reasoning. Based on the iterative reasoning results, a comprehensive judgment result of equipment status and fault is generated to complete the collaborative diagnosis verification.

4. The intelligent inspection method for the status of equipment in a smart substation according to claim 3, characterized in that, The process involves retrieving historical operating data of the equipment, data on similar fault cases, and equipment location information from the intelligent visual inspection coordinate system as supplementary data. This data is then re-input into the random forest classification model and the improved YOLOv5 recognition model for iterative inference. Based on the iterative inference results, a comprehensive judgment result of the equipment status and fault is generated, completing the collaborative diagnostic verification process, including: Using laser scanning and oblique photography technology, a digital twin of the substation is constructed, and the digital twin is synchronized with the physical substation in real time in terms of operating status and data information. By integrating information on equipment working principles, historical fault cases, operation and maintenance procedures, and defect handling solutions, a substation equipment knowledge graph is constructed. Real-time data in the digital twin is then associated with the substation equipment knowledge graph to achieve rule-based preliminary fault reasoning. Historical equipment operation data, similar fault case data, and equipment location information in the intelligent visual inspection coordinate system are retrieved as supplementary data. The equipment operation status is simulated and deduced in the digital twin to analyze the reasons for the mismatch in the model output results. Combining the supplementary data, simulation results, and inference results, the random forest classification model and the improved YOLOv5 recognition model are re-inputted for iterative inference. Based on the iterative inference results, a comprehensive judgment result of equipment status and fault is generated to complete the collaborative diagnosis verification.

5. The intelligent inspection method for the status of equipment in a smart substation according to claim 4, characterized in that, The process of generating a comprehensive judgment result of equipment status and fault based on iterative reasoning results to complete collaborative diagnostic verification includes: Based on the iterative equipment status level information output by the random forest classification model and the iterative fault identification and location information output by the improved YOLOv5 recognition model, a comprehensive judgment result of equipment status and fault is generated. The comprehensive judgment result is imported into the digital twin and compared with the virtual operating status of the device to verify whether the fault type and location conform to the physical operating mechanism. Based on the data of similar fault cases, the validity of the comprehensive judgment result is verified.

6. The intelligent inspection method for the status of equipment in a smart substation according to claim 1, characterized in that, The construction of a three-dimensional intelligent visual inspection coordinate system, combined with equipment importance level, historical failure rate, and inspection scenario, optimizes the inspection robot's optimal inspection route, inspection distance, and inspection angle, generating an inspection task list including: A three-dimensional intelligent visual inspection coordinate system is constructed with the geographical center point of the substation as the origin. Visual constraint parameters are added to the intelligent visual inspection coordinate system, including the optimal inspection distance threshold and the unobstructed inspection angle domain. Acquire the three-dimensional coordinates of all equipment in the substation, the equipment outline dimensions, and the spatial coordinates of key monitoring points to form a substation equipment spatial database; The substation equipment spatial database, the visual constraint parameters, the equipment importance level, historical failure rate data, and the inspection scenario are input into the improved genetic algorithm to output the optimal inspection route, inspection distance, and inspection angle of the inspection robot. Based on the output of the genetic algorithm, an inspection task list is generated.

7. The intelligent inspection method for the status of equipment in a smart substation according to claim 6, characterized in that, The improved genetic algorithm, which inputs the substation equipment spatial database, the visual constraint parameters, equipment importance level, historical failure rate data, and inspection scenario into the inspection scenario, outputs the optimal inspection route, inspection distance, and inspection angle for the inspection robot, including: Based on the three-dimensional chromosome coding rules, the inspection route, inspection distance, and inspection angle are mapped to independent gene segments of the chromosome, and an initial population is constructed based on the independent gene segments and the visual constraint parameters. Differentiated weights are assigned to each device based on its importance level and historical failure rate data, and optimization constraints for algorithm iteration are set based on environmental obstacles and robot movement range in the inspection scenario. A multi-objective fitness function is constructed that integrates patrol route length, visual constraint matching degree, and robot energy consumption. The initial population, the differentiated weights, the optimization constraints, and the multi-objective fitness function are input into an improved genetic algorithm, and iterative calculations are completed through adaptive crossover mutation and elite retention strategies. When the algorithm iterates to the convergence condition, it outputs the global optimal solution, obtaining the optimal inspection route, inspection distance, and inspection angle of the inspection robot.

8. The intelligent inspection method for the status of equipment in a smart substation according to claim 7, characterized in that, The step of inputting the initial population, the differential weights, the optimization constraints, and the multi-objective fitness function into the improved genetic algorithm, and completing iterative calculations through adaptive crossover mutation and elite retention strategies, includes: The differentiated weights are assigned to the corresponding evaluation index of the multi-objective fitness function, and the encoding value of each chromosome in the initial population is optimized and the constraint conditions are checked to remove invalid chromosomes that have route collisions, distances or angles that exceed the preset visual constraint range. The processed initial population and the multi-objective fitness function are input into the improved genetic algorithm. The crossover probability is dynamically adjusted according to the iteration process to perform adaptive crossover operation on the same type of gene segments in the chromosome with inspection routes, inspection distances, and inspection angles. Mutation operation is then performed on the population after crossover. After crossover and mutation, the population is screened for a new generation through an elite retention strategy. The fitness calculation, adaptive crossover, mutation, and elite retention operations are repeated until the algorithm iterates to the convergence condition.

9. A smart substation equipment status intelligent inspection device, characterized in that, include: The task list generation module is used to construct an intelligent visual inspection coordinate system based on three-dimensional space. Combining the equipment importance level, historical failure rate and inspection scenario, it optimizes the inspection robot's optimal inspection route, inspection distance and inspection angle, and generates an inspection task list. The data fusion module is used to acquire equipment image feature data and electrical parameter data synchronously collected by the inspection robot according to the inspection task list, establish a three-dimensional correlation between equipment, image features, and electrical parameters, and form a multi-source dataset; The collaborative diagnosis and verification module is used to analyze the image feature data and electrical parameter data in the multi-source dataset through a random forest classification model and an improved YOLOv5 recognition model, and output equipment status level information and fault identification and location information respectively. The module then combines the equipment status level information and the fault identification and location information to perform collaborative diagnosis and verification of equipment status and faults. The multi-dimensional early warning module is used to establish a multi-dimensional early warning model based on collaborative diagnostic results, historical equipment operation data, and real-time collected data, including limit warning, sudden change warning, synchronization warning, and trend warning, to achieve early warning throughout the entire equipment lifecycle.

Images