Substation autonomous inspection and foreign matter cleaning cooperative control system and method

Abstract

The invention discloses a substation autonomous inspection and foreign matter cleaning cooperative control system and method, and belongs to the technical field of intelligent inspection and foreign matter cleaning. The system comprises a multi-modal detection module used for acquiring images, point clouds and multi-source sensing data and identifying foreign matter coordinates and visual features; the space positioning module is used for executing space registration to obtain a three-dimensional coordinate of the foreign body; the risk decision module fuses the multi-dimensional information and evaluates the foreign matter risk level; the cleaning execution module generates a cleaning path and execution parameters and issues the cleaning path and the execution parameters; and the inspection execution module plans inspection paths and actions based on the state and constraint conditions, and completes subsequent equipment inspection tasks. According to the scheme, the system realizes clear division of labor and accurate control in multiple stages of foreign matter identification, space positioning, risk assessment, cleaning instruction generation, polling after cleaning and the like, and the expansibility and the overall response efficiency of the system are improved, so that efficient, safe and stable execution of autonomous polling and foreign matter cleaning tasks of the transformer substation is guaranteed.

Description

Technical Field

[0001] The present application belongs to the technical field of intelligent inspection and foreign matter cleaning, and specifically relates to a coordinated control system and method for autonomous inspection and foreign matter cleaning of substations. Background Art

[0002] With the development of intelligent power systems, substations, as core nodes of the power grid, are attracting significant attention for their operational safety and efficiency. Substations house a large number of high-voltage electrical equipment. The presence of foreign objects (such as animals, floating objects, and intrusive vegetation) in the surrounding environment can cause serious accidents such as short circuits, discharges, and even fires. Traditional manual inspection methods are limited by high labor costs, high operational risks, and numerous blind spots, making them unable to meet the real-time and precision requirements of modern power grids.

[0003] Nowadays, inspections and foreign object removal are done manually. First, the inspection route is planned in advance and personnel are assigned. They carry detection tools into the substation area and check the equipment status and surrounding environment through visual observation and instrument measurement. When foreign objects are found, the location, type and potential risks are recorded. Based on experience, it is determined whether immediate cleaning is required. Manual tools are used to remove foreign objects. After completion, the cleaning effect is checked again to confirm and record the data.

[0004] However, the entire process relies on manual operation, and efficiency is limited by the physical strength and experience of personnel. Fatigue or negligence can easily lead to missed inspections or incomplete cleaning. Data recording and reporting are prone to delays or errors due to human factors, and there is a lack of real-time feedback mechanism, making it difficult to quickly respond to dynamically changing environmental risks. Manual cleaning is difficult for complex or hidden foreign objects (such as conductors embedded in equipment gaps). Summary of the Invention

[0005] The embodiments of the present application provide a coordinated control system and method for autonomous inspection and foreign object cleaning of substations, which solves the problem that the existing foreign object cleaning and intelligent inspection rely entirely on manual operation, the efficiency is limited by the physical strength and experience of the personnel, and it is easy to miss inspections or incomplete cleaning due to fatigue or negligence; data recording and reporting are prone to delays or errors due to human factors, and lack a real-time feedback mechanism, making it difficult to quickly respond to dynamically changing environmental risks; for complex or hidden foreign objects (such as conductors embedded in equipment gaps), manual cleaning is difficult.

[0006] In a first aspect, an embodiment of the present application provides a substation autonomous inspection and foreign matter cleaning collaborative control system, the system comprising: a multimodal detection module, configured to acquire two-dimensional image data, lidar point cloud data, and multimodal sensor data of the substation, and employ a convolutional neural network target detection algorithm to identify the two-dimensional coordinate position of each foreign object in the two-dimensional image data and a first foreign object visual feature description; A spatial positioning module is used to perform spatial registration processing based on the laser radar point cloud data and the two-dimensional coordinate position to obtain the first three-dimensional spatial coordinates of each foreign object; a risk decision module, configured to obtain spatial layout information of the electrical equipment, input the first three-dimensional spatial coordinates, the first foreign object visual feature description, the spatial layout information, and the multimodal sensor data into a preset foreign object risk level scoring model, and obtain a first foreign object risk level score for each foreign object; a cleaning execution module, configured to determine a first foreign object operation requirement for each foreign object based on the first foreign object risk level score and a preset cleaning strategy library, obtain a first current position, a first posture angle, and a first motion state information of the inspection robot, determine a cleaning path of the inspection robot and first cleaning execution parameters for each foreign object based on the first three-dimensional space coordinates, the first current position, the first posture angle, the first motion state information, and the first foreign object operation requirement, and send the cleaning path and the first cleaning execution parameters for each foreign object to the inspection robot; The inspection execution module is used to obtain the second current position, second posture angle, second motion state information of the inspection robot upon receiving the cleaning completion information of the inspection robot, as well as the spatial topology structure and safety constraints of the substation, and determine the inspection path of the inspection robot and the inspection actions on each electrical equipment based on the second current position, second posture angle, second motion state information, spatial topology structure, spatial layout information and safety constraints, and send the inspection path and inspection actions on each electrical equipment to the inspection robot.

[0007] In a second aspect, an embodiment of the present application provides a method for coordinated control of autonomous inspection and foreign matter cleaning of a substation, the method comprising: Acquire two-dimensional image data, lidar point cloud data, and multimodal sensor data of the substation, and use a convolutional neural network target detection algorithm to identify the two-dimensional coordinate position of each foreign object in the two-dimensional image data and a first foreign object visual feature description; Performing spatial registration processing based on the laser radar point cloud data and the two-dimensional coordinate position to obtain the first three-dimensional spatial coordinates of each foreign object; Obtaining spatial layout information of the electrical equipment, inputting the first three-dimensional spatial coordinates, the first foreign object visual feature description, the spatial layout information, and the multimodal sensor data into a preset foreign object risk level scoring model to obtain a first foreign object risk level score for each foreign object; Determine a first foreign object operation requirement for each foreign object based on the first foreign object risk level score and a preset cleaning strategy library, obtain a first current position, a first posture angle, and a first motion state information of the inspection robot, determine a cleaning path of the inspection robot and a first cleaning execution parameter for each foreign object based on the first three-dimensional space coordinates, the first current position, the first posture angle, the first motion state information, and the first foreign object operation requirement, and send the cleaning path and the first cleaning execution parameter for each foreign object to the inspection robot; If the cleaning completion information of the inspection robot is received, the second current position, second posture angle, and second motion state information of the inspection robot are obtained, as well as the spatial topology structure and safety constraints of the substation are obtained, and the inspection path of the inspection robot and the inspection actions on each electrical equipment are determined according to the second current position, second posture angle, second motion state information, spatial topology structure, spatial layout information and safety constraints, and the inspection path and inspection actions on each electrical equipment are sent to the inspection robot.

[0008] In a third aspect, an embodiment of the present application provides an electronic device comprising a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the method described in the second aspect.

[0009] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the second aspect are implemented.

[0010] In the embodiment of the present application, the system achieves clear division of labor and precise control in multiple stages such as foreign object identification, spatial positioning, risk assessment, cleaning instruction generation, and post-cleaning inspection, thereby improving the system's scalability and overall response efficiency, thereby ensuring the efficient, safe, and stable execution of autonomous inspections and foreign object cleaning tasks of substations. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 This is a schematic diagram of the structure of the substation autonomous inspection and foreign matter cleaning collaborative control system provided in Example 1 of the present application; Figure 2 This is a schematic diagram of the structure of the substation autonomous inspection and foreign matter cleaning collaborative control system provided in Example 2 of the present application; Figure 3 This is a flow chart of a coordinated control method for autonomous inspection and foreign matter cleaning of a substation provided in Example 3 of the present application; Figure 4 This is a schematic diagram of the structure of the electronic device provided in Example 4 of the present application. DETAILED DESCRIPTION

[0012] To further clarify the objectives, technical solutions, and advantages of this application, specific embodiments of the present application are described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are intended only to illustrate this application and are not intended to limit it. It should also be noted that, for ease of description, the drawings only illustrate portions relevant to this application, not all of them. Before discussing the exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts depict the various operations (or steps) as sequential processes, many of the operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process may terminate upon completion of its operations, but may also include additional steps not shown in the accompanying drawings. The process may correspond to a method, function, procedure, subroutine, subprogram, or the like.

[0013] The following will be combined with the accompanying drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of this application.

[0014] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of this application can be implemented in an order other than that illustrated or described herein, and that the objects distinguished by "first," "second," and the like are generally of the same type, and do not limit the number of objects; for example, the first object can be one or more. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects connected are in an "or" relationship.

[0015] Below, in conjunction with the accompanying drawings, an RSMC chip, a chip multi-stage startup method, and a Beidou communication and navigation device provided in the embodiments of the present application are described in detail through specific embodiments and their application scenarios.

[0016] Example 1 Figure 1 This is a schematic diagram of the structure of the substation autonomous inspection and foreign matter cleaning collaborative control system provided in Example 1 of this application. Figure 1 As shown, specifically including the following: The multimodal detection module 101 is configured to acquire two-dimensional image data, lidar point cloud data, and multimodal sensor data of the substation, and to use a convolutional neural network target detection algorithm to identify the two-dimensional coordinate position of each foreign object in the two-dimensional image data and a first foreign object visual feature description; A spatial positioning module 102 is configured to perform spatial registration processing based on the laser radar point cloud data and the two-dimensional coordinate position to obtain the first three-dimensional spatial coordinates of each foreign object; The risk decision module 103 is configured to obtain spatial layout information of the electrical equipment, input the first three-dimensional spatial coordinates, the first foreign object visual feature description, the spatial layout information, and the multimodal sensor data into a preset foreign object risk level scoring model, and obtain a first foreign object risk level score for each foreign object; The cleaning execution module 104 is configured to determine a first foreign object operation requirement for each foreign object based on the first foreign object risk level score and a preset cleaning strategy library, obtain a first current position, a first posture angle, and a first motion state information of the inspection robot, determine a cleaning path for the inspection robot and first cleaning execution parameters for each foreign object based on the first three-dimensional spatial coordinates, the first current position, the first posture angle, the first motion state information, and the first foreign object operation requirement, and send the cleaning path and the first cleaning execution parameters for each foreign object to the inspection robot; The inspection execution module 105 is used to obtain the second current position, second posture angle, second motion state information of the inspection robot upon receiving the cleaning completion information of the inspection robot, as well as the spatial topology structure and safety constraints of the substation, and determine the inspection path of the inspection robot and the inspection actions on each electrical equipment based on the second current position, second posture angle, second motion state information, spatial topology structure, spatial layout information and safety constraints, and send the inspection path and the inspection actions on each electrical equipment to the inspection robot.

[0017] In this embodiment, a substation refers to a key facility in the power system that performs voltage conversion, energy distribution, and control and protection functions. In this scenario, the substation is the target area for inspection and cleaning operations, and contains multiple electrical equipment (such as transformers, switchgear, busbars, lightning arresters, etc.), as well as auxiliary facilities such as access paths, fences, and ground surfaces.

[0018] Two-dimensional image data is captured by an image sensor (such as a high-definition camera) and is expressed as a pixel matrix. It typically includes RGB or grayscale images. In this solution, 2D image data is used to perform tasks such as object detection, image change recognition, and regional feature extraction, and is one of the primary data sources for visual analysis.

[0019] LiDAR point cloud data can refer to the three-dimensional point set data obtained by the LiDAR device through laser scanning. Each point contains attributes such as spatial coordinates and possible reflection intensity.

[0020] Multimodal sensor data can refer to perception data collected by different types of sensors (such as infrared, temperature and humidity, ultrasonic, gas detection, vibration sensors, etc.) to describe the state of the environment or objects.

[0021] Convolutional neural network object detection algorithms are deep learning models based on convolutional neural networks (CNNs) and are used to detect the location and category of target objects in images. Typical examples include YOLO, Faster R-CNN, and SSD. In this solution, they are used to detect foreign objects in substations from two-dimensional image data, including their 2D coordinate location and feature extraction.

[0022] Foreign matter may refer to abnormal objects that do not belong to the substation equipment or its auxiliary structures, and may enter the station due to wind, animal intrusion, falling, etc., such as plastic bags, bird nests, tools, animals, branches, etc.

[0023] The 2D coordinate position refers to the pixel-level location of a foreign object in the image coordinate system, typically expressed as the center point coordinates or the coordinates of the top-left and bottom-right corners of a bounding box. This is one of the outputs of the object detection algorithm and is subsequently used for spatial registration and 3D positioning calculations.

[0024] The first foreign object visual feature description may refer to the feature vector or semantic description information extracted by the convolutional neural network when identifying foreign objects, which can characterize the appearance attributes of the foreign object, such as color distribution, shape contour, texture pattern, etc.

[0025] High-definition visible light industrial cameras deployed on inspection robots or fixed monitoring devices can capture the substation scene in real time at a set frame rate (e.g., 1–5 frames / second), acquiring image data covering the entire substation area to generate the two-dimensional image data. This image data is in RGB format, with a resolution of 1920×1080 or higher, and is timestamped to facilitate subsequent synchronization with point cloud data and sensor data. A three-dimensional lidar (e.g., Velodyne, Livox, etc.) mounted on a robot or fixed platform scans the substation's spatial environment using multiple beams and 360° rotation to acquire the lidar point cloud data. Each sampling point in the point cloud data records its three-dimensional coordinates (X, Y, Z) and reflection intensity value, which can be used to reconstruct the geometry of substation equipment and the spatial distribution of foreign objects. Other sensor components deployed in the inspection system (e.g., infrared thermal imagers, ultrasonic rangefinders, thermo-hygrometers, air quality sensors, etc.) simultaneously acquire non-image data reflecting the substation environment and equipment operating status to generate the multimodal sensor data. This data can be used to enhance the accuracy of foreign object detection and subsequent risk level determination. The two-dimensional image data is fed into a trained convolutional neural network (CNN) object detection model, such as YOLOv5, Faster R-CNN, or RetinaNet. The model structure includes: Backbone network (such as CSPDarknet, ResNet), used to extract image features; Neck networks (such as FPN and PAN) perform multi-scale feature fusion; The head network outputs the bounding box location, category label, and confidence of the foreign object.

[0026] The network output includes the rectangular bounding box and center point coordinates of each detected foreign object, forming the two-dimensional coordinate position of the foreign object.

[0027] Based on the bounding box output by the detection model, region alignment techniques such as RoI Align are used to extract the deep feature vector of each foreign object region in the CNN's intermediate feature map. This vector represents the foreign object's color distribution, texture details, and edge contours, forming the first foreign object visual feature description. This description is used in subsequent steps such as foreign object classification, risk scoring, and cleaning strategy matching.

[0028] The first three-dimensional spatial coordinate can be the three-dimensional coordinate representing the actual position of the foreign object in the substation scene calculated by spatially aligning the two-dimensional coordinates of the foreign object on the image plane with the three-dimensional point cloud data collected by the lidar to establish a pixel-point cloud mapping relationship.

[0029] First, use methods such as the Zhang calibration method or the LiDAR-Camera Calibration Toolkit (such as Kalibr) to obtain the extrinsic parameter matrix between the camera and the lidar. This is the rotation matrix R (3×3) and translation vector T (3×1) between the lidar coordinate system and the image coordinate system. These extrinsic parameters are used to transform the radar points from the radar coordinate system to the camera coordinate system. Obtain the camera's intrinsic parameter matrix K, including the focal length fx, fy and the principal point coordinates cx, cy, which are used to describe the projection process from three-dimensional space to the two-dimensional image plane. This matrix is ​​often obtained through the checkerboard image calibration method. For each point X, Y, Z in the lidar point cloud data (in the lidar coordinate system), use the following transformation steps to map it to the corresponding pixel point (u, v) in the image coordinate system: First, transform the point cloud data into the camera coordinate system through rotation and translation:

[0030] Then project the 3D points onto the image plane:

[0031] Among them, (X, Y, Z) are the coordinates of the lidar point cloud data; R is the rotation matrix (3×3), which represents the direction transformation of rotating the point in the radar coordinate system to the camera coordinate system; T is the translation vector (3×1), which represents the offset position of the origin of the radar coordinate system relative to the camera coordinate system; ( , , ) is the three-dimensional coordinate in the camera coordinate system (that is, the point cloud point is converted to the camera's perspective).

[0032] A one-to-one mapping is then established between the projected pixel positions (u, v) of all laser points in the image and their corresponding 3D coordinates (X, Y, Z), forming an image pixel index table. This index table can be used to quickly find the 3D point corresponding to any image point. For each foreign object identified using the convolutional neural network object detection algorithm, the corresponding 3D point of the closest (u, v) point is searched in the index table. To enhance robustness, a valid point cloud projection is searched within a 3×3 or 5×5 neighborhood window around the pixel coordinate to avoid matching failures caused by occlusion or sparse point clouds. Based on the matching results, the 3D position (X, Y, Z) of each foreign object in the actual substation environment is extracted as its first 3D spatial coordinate.

[0033] Electrical equipment can refer to various high-voltage or low-voltage devices in substations that undertake functions such as electric energy transmission, conversion, control, measurement and protection, mainly including transformers: used for voltage conversion between high and low voltage; circuit breakers / switch devices: used to control the on and off of circuits; cable trays, busbars, and buses: for transmitting current; voltage transformers and current transformers: for measurement and protection; capacitors and reactors: for reactive power compensation; lightning arresters and grounding devices: for protection; communication and monitoring devices: for transmitting status information; ring network cabinets, switch cabinets, complete distribution cabinets, etc.

[0034] Spatial layout information refers to the three-dimensional spatial arrangement structure information of electrical equipment inside the substation, including the equipment coordinates and the installation position of each electrical equipment in three-dimensional space, such as (X, Y, Z) in the world coordinate system; equipment size and bounding box information are used to determine whether foreign objects are close to or blocking the equipment; equipment categories such as circuit breakers, transformers, busbars, etc. may correspond to different sensitive areas and risk levels; installation relationships and topological structures, such as some equipment is connected in groups, and multiple points need to be paid attention to at the same time during cleaning; maintenance / channel restriction information, which areas are high-voltage danger zones, whether robots are allowed to enter, and how many meters from the boundary they need to slow down, etc.

[0035] The preset foreign object risk level scoring model may be a risk assessment model designed based on historical experience or expert rules, and is used to assign a risk level score to each foreign object target detected in the substation.

[0036] The first foreign object risk level score may refer to a risk score result obtained by evaluating a certain foreign object target through the above model during the current identification cycle, which reflects the potential threat level of the foreign object to the safety of substation operation.

[0037] First, the spatial layout information of the electrical equipment is exported from the BIM model (Building Information Modeling) or GIS system of the substation. This information is usually expressed in a standard three-dimensional model (such as IFC format) or JSON / XML structured format. The content includes the spatial position coordinates (Xi, Yi, Zi), physical dimensions (length, width and height), installation direction, type number, operating voltage level and logical connection relationship with surrounding equipment (topological constraints) of various types of equipment (such as transformers, circuit breakers, switchgear, etc.). In order to match the three-dimensional spatial coordinate system actually used by the cleaning robot, a unified spatial coordinate transformation matrix (such as quaternion rotation and Euclidean translation based on homogeneous transformation) is used to register the layout data to the world coordinate system used by the robot. Then the first three-dimensional spatial coordinates, the first foreign body visual feature description, the spatial layout information and the multimodal sensor data are combined into a multidimensional feature vector. in, is the first three-dimensional space coordinate, Description of the visual characteristics of the first foreign body; It is the spatial layout information; The multimodal sensor data is then fed into a pre-trained foreign object risk rating model. This model can be a lightweight neural network (e.g., an MLP or Transformer structure), a tree model (e.g., XGBoost or LightGBM), or a weighted model that combines rules with a linear combination of scoring factors. The model output is the first foreign object risk rating score for the foreign object.

[0038] In order to build a generalizable foreign body risk rating model, we first need to construct a high-quality historical training dataset. Each sample in the training dataset consists of the following four input elements and a label: The first three-dimensional space coordinate (i.e. the three-dimensional position of the historical foreign body ( )); First, the foreign object visual feature description (usually the image semantic vector extracted by pre-trained deep convolutional neural network such as ResNet50 or YOLOv5) , indicating the type, texture, size and other characteristics of the foreign matter); Spatial layout information (including the spatial relationship between foreign objects and key electrical equipment, such as distance , direction angle, whether it crosses the safety boundary, etc.); Multimodal sensor data (such as temperature distribution of infrared thermal imaging, wind speed direction, electromagnetic interference intensity and other environmental parameters, uniformly represented as ); The training label is the risk level label of each historical foreign body sample after being scored by experts or calibrated by accident data retrospectively, usually a continuous risk score (such as ∈[0,1]).

[0039] For each historical sample, the above four types of data are encoded into an input feature vector in a unified format: . All vector features are normalized, such as using Min-Max normalization or Z-score standardization, to ensure that different dimensions have consistent weights in the model input. Different model structures are selected according to the type of scoring label: if the label is a continuous risk score (regression task), a multi-layer perceptron (MLP) can be used: a fully connected neural network or a gradient boosting decision tree such as XGBoost and LightGBM, which is suitable for small and medium-sized data; if the label is a risk level (classification task), a classification network with a cross-entropy loss function can be used; or an SVM + PCA dimensionality reduction method. The optimization method used in training can be Adam or SGD, and the loss functions are: for regression tasks: minimizing the mean squared error (MSE); for classification tasks: minimizing the cross entropy loss (Cross Entropy); through multiple rounds of training and cross-validation, the generalization performance of the model in different scenarios is ensured.

[0040] After training is complete, an independent test set is used to evaluate the model's prediction accuracy and robustness, using metrics including regression metrics: , MAE, and RMSE; classification metrics: accuracy, recall, F1-score, and confusion matrix. If the model performs poorly in certain scenarios, further optimization can be achieved through oversampling / undersampling; data augmentation (such as image rotation and brightness perturbation); and model ensembles (such as bagging or boosting). Ultimately, the trained foreign object risk rating model is deployed to edge devices or backend servers to quickly assign risk scores to newly identified foreign objects, which is then used to drive subsequent cleaning strategies and task scheduling.

[0041] The preset cleaning strategy library may refer to a set of parameter rules that are defined and stored in advance and used to guide the inspection robot to perform foreign matter cleaning operations.

[0042] The first foreign object handling requirement refers to a specific set of cleaning parameters and control requirements for each foreign object, selected from a cleaning strategy library based on the current identification information and risk level score. This includes the specified cleaning method (e.g., air jet / adsorption / brush); the required angular adjustment range of the end effector; the safe approach distance; the cleaning force and time; the robot's posture stability requirements (e.g., whether deceleration and stable docking are required); and whether cleaning can be completed in stages (e.g., dust removal followed by adsorption).

[0043] The inspection robot in this system refers to a mobile intelligent robot platform with automatic navigation, visual recognition, foreign object cleaning and path planning capabilities.

[0044] The first current position may refer to the current spatial position coordinates measured by the inspection robot based on a navigation and positioning system (such as GPS, SLAM, or odometer + lidar fusion positioning) when the inspection robot receives a cleaning task instruction.

[0045] The first attitude angle may refer to the current orientation or attitude information of the robot, such as pitch, yaw, and roll, which are used to describe the current orientation of the robot.

[0046] The first motion state information may refer to the current dynamic parameter information of the inspection robot, including the current speed (linear speed and angular speed); the current acceleration; whether the robot is currently in a turning, accelerating, or decelerating state; and the current path curvature or driving stability.

[0047] A clearing path can refer to a three-dimensional path trajectory from the current robot position (the first current position) to the location of the foreign object (the first three-dimensional coordinate), planned based on requirements such as obstacle avoidance, shortest path, and safe distance. This path is calculated by the navigation system using path planning algorithms such as A*, D*, RRT, PRM, and Bezier curve fitting.

[0048] The first cleaning execution parameters refer to the specific cleaning action parameters that the inspection robot must perform before and after reaching each foreign object, controlling its cleaning actuator to complete the cleaning task. These parameters may include: cleaning method (jet, gripper, suction, etc.); cleaning trigger distance and angle; cleaning execution force (torque, suction, pressure, etc.); action duration (e.g., how long the cleaning state is maintained); posture adjustment angle; and action stability parameters (e.g., speed change curve, buffering action, etc.).

[0049] After completing a first-level foreign object risk level score for each foreign object, the system matches the score results with a pre-set cleaning strategy library to perform a cleaning decision matching process. This cleaning strategy library is a structured knowledge base, typically organized as a rule table, decision tree, or conditional mapping function. It defines first-level foreign object handling requirements corresponding to different risk levels and foreign object types. The system first extracts the corresponding first-level foreign object risk level score for each foreign object and searches for matching rules in the strategy library based on the object's type identifier (such as plastic film, tree branch, or metal cable). During the strategy matching process, the system uses a multi-conditional logic rule matching mechanism: first, a coarse classification based on foreign object risk level (high / medium / low) is performed, followed by a detailed classification based on the foreign object's visual and spatial location features. To improve decision accuracy, the system also incorporates a fuzzy logic scoring mechanism to calculate the membership between the first-level foreign object risk level score and the risk domain in the strategy library, ensuring decision flexibility within boundary conditions. Ultimately, each foreign object is assigned a complete first-level foreign object handling requirement. To obtain the inspection robot's first current position, first attitude angle, and first motion state information, the system employs a fusion positioning perception approach, combining an inertial measurement unit (IMU), a high-precision GNSS positioning device, and an odometer to obtain preliminary motion and position data. In static areas or indoor environments where GNSS signals are unavailable, the system further collects the robot's onboard LiDAR point cloud data and combines it with a map of the substation's internal environment for laser SLAM positioning. By matching the distribution of feature points between the current laser scan frame and the constructed map, an accurate estimation of the first current position is achieved. Simultaneously, the IMU's three-axis gyroscope and accelerometer record the robot's rotational angular velocity and linear acceleration in real time. Using a Kalman filter or extended Kalman filter algorithm, the IMU data is fused with visual / radar positioning results to calculate the robot's attitude Euler angles or quaternion information, i.e., the first attitude angles. Furthermore, based on the velocity and acceleration data continuously recorded by the IMU and wheel encoders, the system extracts the robot's linear velocity and acceleration vectors at its current position, forming a complete first motion state information. Based on the first current position and the first three-dimensional spatial coordinates of the target foreign object, a path planning algorithm such as A* (A-Star) or RRT* (Rapidly-exploring Random TreeStar) is used to generate a cleaning path for the inspection robot to reach the target foreign object. The path generation process not only considers obstacle avoidance and path smoothness, but also reserves posture adjustment nodes to meet the end-point posture accuracy requirements required for cleaning operations. The system then calculates the posture deviation based on the current first posture angle and the spatial orientation of the target foreign object. Using an inverse kinematics algorithm, combined with the structural parameters of the robot's end effector, it infers the posture adjustment angle required for the cleaning task, ensuring that the robot can operate on the foreign object in the correct direction.On this basis, the system further uses a fifth-order polynomial trajectory interpolation method or a speed-acceleration constraint planning algorithm based on the speed and acceleration data contained in the first motion state information and the speed limit and operation stability requirements in the first foreign object operation requirements to generate a first speed control parameter that meets the dynamic constraints to ensure that the robot has stability and safety when approaching the target foreign object. Finally, based on the obtained posture adjustment angle and the first speed control parameter, combined with the force control, range of action and trigger mechanism in the first foreign object operation requirements, the system uses an impedance control or force-position hybrid control strategy to generate detailed cleaning action instructions, including control parameters such as the motion trajectory, action force, and execution duration of the end cleaner. The system packages the above path planning results, posture adjustment sequence and cleaning action into the first cleaning execution parameter, and sends it together with the cleaning path to the inspection robot, which then performs the cleaning task.

[0050] The cleaning completion information may refer to a status signal or data packet that the inspection robot feeds back to the control system after completing the task of cleaning a certain area or specific foreign matter.

[0051] The second current position can be the robot's current 3D spatial coordinates or 2D plane coordinates, which refers to the specific location of the robot when it completes the cleaning task. It is usually calculated by integrating sensors such as GNSS, LiDAR SLAM, and inertial navigation systems.

[0052] The second attitude angle can be the robot's orientation or posture at that location, typically expressed as Euler angles (pitch, yaw, and roll) or quaternions. It reflects the robot's tilt and orientation relative to the ground or equipment, helping to assess the safety and feasibility of the next action.

[0053] The second motion state information can include dynamic information such as the robot's linear velocity, angular velocity, and acceleration, describing the robot's current motion trend. This is crucial for planning the next path and action execution, ensuring smooth and safe movement.

[0054] Spatial topology is an abstract representation of the spatial relationships and connections between various devices, facilities, passageways, and obstacles within a substation. It includes information such as distances between devices, navigable paths, and node connectivity, and is typically stored as a topological graph or network structure. Spatial topology provides constraints for path planning algorithms, preventing robots from entering inaccessible areas or causing collisions.

[0055] Safety constraints can refer to safety regulations and restrictions that must be adhered to by the substation environment and robot operations. Examples include electrical safety distance requirements (minimum safe distance between the robot and high-voltage equipment), speed limits (limiting maximum speed within a specific area), load and operating force limits (preventing damage to equipment or the robot itself), restricted access areas or time windows (access to certain areas is limited to specific times), and environmental state restrictions (e.g., restricting movement in rainy or snowy weather).

[0056] An inspection path is a specific route planned for a robot within a substation, including its starting point, end point, waypoint coordinates, and movement method. This path is typically generated based on spatial topology and safety constraints, aiming to efficiently cover all electrical equipment areas requiring inspection.

[0057] Inspection actions can be specific detection or operation tasks performed by the robot on the inspection target equipment, which may include visual photography and image acquisition, infrared thermal imaging detection, vibration, temperature, current, voltage and other sensor data acquisition, sound or gas leak detection, abnormal alarm and data upload.

[0058] Upon receiving the inspection robot's cleaning completion message, the system first utilizes the robot's integrated navigation system (including an inertial navigation unit (IMU), lidar SLAM, and visual odometry) to obtain its latest state data. This system calculates and obtains the robot's second current position (i.e., its absolute position in three-dimensional space after cleaning), second attitude angles (including pitch, yaw, and roll), and second motion state information (including linear velocity, acceleration, and angular velocity) in real time. This data constitutes the robot's kinematic state and serves as the initial constraint boundary for path planning and motion control. The system then loads and analyzes the substation's spatial topology, which is represented as a directed graph. Nodes represent electrical equipment, walls, and passageways, and edges represent traversable paths. Edges are weighted based on factors such as path length and obstacle risk level, forming a weighted graph model. Based on this, the system uses an improved heuristic A or RRT algorithm combined with a heuristic function to rapidly calculate multiple candidate paths, ensuring that path planning meets the shortest path requirements while also considering travel safety and risk minimization.

[0059] At the same time, the system extracts spatial layout information from the CAD or 3D BIM model, detailing the geometric dimensions, position coordinates, spatial volume occupied, and surface normal direction of each device and obstacle. Using voxel gridding, the space is divided into traversable and impassable cells, forming a 3D voxel map. Combined with the robot's current second posture angle, a coordinate transformation is performed using a rotation matrix to calculate the robot's sensor field of view and direction of motion. This ensures that the robot's geometric dimensions and steering constraints are fully considered during path planning, preventing collisions with equipment or obstacles caused by path design.

[0060] In addition, the system analyzes the robot's current motion stage (acceleration, constant speed, or deceleration) based on the second motion state information provided by the robot. If the acceleration exceeds the preset threshold, the system automatically inserts a buffer path segment during the planning process to reduce motion risks, and uses fifth-order polynomial trajectory planning to smoothly adjust the speed to ensure that the robot maintains a low-speed and stable state when performing fine inspection actions, thereby improving data collection quality.

[0061] The system further loads and applies safety constraints based on substation electrical safety regulations, including minimum safety distances in high-voltage areas, restricted area markings (such as areas around knife switches), heat sources, and electromagnetic interference areas. Through spatial Boolean operations, these areas are mapped into inaccessible and high-risk areas in three-dimensional space, which are strictly avoided during path planning and incorporated into the path search algorithm as hard constraints.

[0062] Based on all of the above information, the system generates the final path according to the following steps: Starting from the second current position and second posture angle, the system utilizes a heuristic A* algorithm, combining spatial topology and safety constraints, to generate a preliminary set of paths that meet safety constraints. Based on the device geometry and space occupancy data in the spatial layout information and the robot's posture, the system adjusts the field of view and range of motion to eliminate candidate paths that could cause collisions. Dynamic constraints from the second motion state information are incorporated to adjust the velocity profile between path points, ensuring smooth robot motion and suitability for subsequent actions. Through hierarchical risk filtering, paths with the lowest risk and compliance with operating procedures are prioritized, taking into account both path length and inspection efficiency. The resulting final inspection path ensures that the robot reaches the target electrical equipment locations in the optimal and safe manner. The system then matches corresponding inspection action templates from the action database based on the equipment type and inspection standards, such as infrared imaging, image recognition, sound detection, and temperature and humidity data collection. By combining the robot's second posture angle with the equipment's orientation, the system accurately calculates sensor adjustment angles and working distances to ensure effective action execution and data quality.

[0063] Ultimately, the system encodes the determined inspection path and corresponding inspection actions into structured control instructions, which are then transmitted to the inspection robot via the industrial wireless communication network, achieving closed-loop control of path navigation and action acquisition. The robot completes its autonomous inspection tasks according to the instructions and provides real-time feedback on its execution status, enabling dynamic scheduling and safety assurance.

[0064] In the embodiment of the present application, the system achieves clear division of labor and precise control in multiple stages such as foreign object identification, spatial positioning, risk assessment, cleaning instruction generation, and post-cleaning inspection, thereby improving the system's scalability and overall response efficiency, thereby ensuring the efficient, safe, and stable execution of autonomous inspections and foreign object cleaning tasks of substations.

[0065] Based on the above technical solution, optionally, the cleaning execution module is further used to: Calculate, based on the first current position and the first three-dimensional spatial coordinates, a sub-cleaning path of the inspection robot from the first current position to each foreign object using a path planning algorithm, and determine a cleaning path of the inspection robot based on the sub-cleaning path of the inspection robot from the first current position to each foreign object; Calculating a first end posture adjustment angle of the inspection robot for cleaning foreign objects according to the first posture angle and the first three-dimensional space coordinate; Determining a first speed control parameter of the inspection robot in each foreign object approach section according to the first motion state information and the first foreign object operation requirement; The first cleaning execution parameter of each foreign object is determined according to the first end posture adjustment angle, the first speed control parameter and the first foreign object operation requirement.

[0066] In this solution, the path planning algorithm refers to an algorithm used to calculate the path that the inspection robot takes from its current position, through a series of spatial points (such as multiple foreign object locations), and finally forms a travel path that meets the requirements (such as shortest, safe, avoids obstacles, meets constraints, etc.). Including: A* (A-star) algorithm: combines heuristic function and cost function to calculate the optimal path from the starting point to the target point. Dijkstra algorithm: Based on the graph structure, it calculates the shortest path from the starting point to all points, which is suitable for non-heuristic global optimal path search. RRT: Suitable for path exploration in unstructured spaces in dynamic environments. TSP variant algorithm: For path optimization of multiple foreign object points (i.e., multi-target path), genetic algorithm and ant colony algorithm can be used for optimization in combination with constraints.

[0067] The sub-cleaning path may refer to a minimum cost path from the current position (first current position) of the inspection robot to the three-dimensional space coordinates (first three-dimensional space coordinates) of a target foreign object, and is a local cleaning path for each foreign object.

[0068] The first end-position adjustment angle may be an angle value that the robot's execution arm or end-effector needs to adjust before approaching the foreign object, so as to accurately align the spatial posture of the foreign object.

[0069] The first speed control parameter may refer to the speed and acceleration parameter set that the inspection robot needs to follow when approaching a foreign object, which may include a uniform approach speed (linear speed), a maximum acceleration, a deceleration threshold (triggered according to the approach distance), and a jitter threshold (for posture stabilization).

[0070] The foreign object approach segment may refer to a short path area from the last navigation point on the path of the inspection robot to the actual location of the foreign object, which is usually located at the end of the cleaning path.

[0071] Based on the inspection robot's current first position and the first three-dimensional spatial coordinates of each foreign object, the system uses a multi-objective path optimization strategy to calculate sub-cleaning paths from the current point to multiple foreign object locations. During the path generation process, the system prioritizes the use of an improved A* path planning algorithm, which combines the spatial topology structure with the robot's kinematic model to calculate the shortest distance and dynamic constraints of each path segment, while also introducing path risk level and safety constraint information as judgment criteria. In addition, for the sequential optimization of multiple foreign objects, the system combines multiple sub-cleaning paths with the Traveling Salesman Problem (TSP) heuristic algorithm to generate a global cleaning path that meets the requirements of shortest path length, minimum energy consumption, and reasonable cleaning priority, ensuring the overall efficient execution of the cleaning task.

[0072] The system then uses forward kinematics to determine the current actuator state based on the inspection robot's first posture angle and the first three-dimensional coordinates of the target foreign object. Combined with inverse kinematics, the system calculates the first end-effector posture adjustment angle required for the robot's end-effector or cleaning tool to reach each foreign object cleaning point. This angle is expressed as Euler angles or quaternions. The calculation fully considers the robot's multi-degree-of-freedom redundant solution space and selects the feasible solution that provides the most stable operating posture and avoids mechanical interference, ensuring proper contact between the cleaning tool and the foreign object surface and smooth operation.

[0073] Next, the system integrates the inspection robot's current first motion state information, including linear velocity, acceleration, and acceleration rate of change, as well as the specific operational requirements of each foreign object, such as cleaning intensity, approach distance, and posture accuracy. It then uses a dynamic speed adjustment algorithm (such as a trajectory tracking PID control algorithm or speed planning based on model predictive control (MPC)) to generate the first speed control parameter for each foreign object approach segment. This parameter ensures that the robot achieves smooth deceleration when approaching the cleaning target, meeting the requirement that the terminal speed approaches zero, while also ensuring that the spatial accuracy and safety distance constraints in path planning are not violated. If specific cleaning stability or speed limitations exist, the system further uses an S-shaped speed curve to adjust the speed gradient to reduce robot vibration and operational impact.

[0074] The system then inputs the calculated first end-effector posture adjustment angle into the inverse kinematics model to calculate the specific posture trajectory of the inspection robot's end-effector during the cleaning operation, ensuring that the end-effector's orientation is aligned with the object's surface normal as closely as possible, thereby improving the accuracy and stability of the cleaning operation. Combined with the first velocity control parameter, the system then plans the end-effector's velocity and acceleration profiles to ensure a smooth and safe motion. The inverse kinematics model training process typically involves the following steps: First, a large amount of end-effector spatial position and posture data is collected from the robot's manipulator at various joint angle configurations to construct a high-quality dataset mapping joint angles to end-effector positions. Next, a suitable machine learning model (such as a neural network, support vector machine, or gradient descent-based deep learning model) is selected to train on this dataset. By minimizing the prediction error of the end-effector position and posture, the model parameters are continuously adjusted to improve prediction accuracy. During training, constraint optimization is performed incorporating physical constraints (such as joint angle limits and the geometric relationships of the manipulator structure) to ensure that the output inverse solution meets the manipulator's kinematic feasibility. Finally, cross-validation and actual robot motion testing verify the model's generalization and stability, resulting in an efficient prediction model that can be used for real-time inverse kinematics solutions.

[0075] The system also analyzes the first foreign object operation requirements for each foreign object in detail, covering multiple parameters such as the cleaning method type (such as mechanical scrubbing, air blowing, or suction), the effective distance, the cleaning force, the action trigger conditions, and the action duration. Through logical combination and condition matching, the system integrates the posture adjustment and speed control results to form a complete set of first cleaning execution parameters. Specifically, this includes the cleaning instruction sequence, the terminal posture trajectory planning, the speed and force control range, the action start and end conditions, and the execution time, ensuring that the robot performs foreign object cleaning tasks accurately, stably, and efficiently.

[0076] In this solution, by designing the inspection robot's path planning and cleaning action parameters in steps, the system can adaptively generate the optimal cleaning path and execution parameters based on the spatial position of the foreign object and the current state of the robot, thereby improving the cleaning accuracy and stability in complex substation environments, while also having stronger robustness and dynamic adjustment capabilities.

[0077] Based on the above technical solution, optionally, the inspection execution module is further used to: Determining an inspection path of the inspection robot according to the second current position and the spatial topological structure; Determine, based on the second posture angle and the spatial layout information, the perception angle setting parameters for the inspection robot to reach the location of each electrical device in the inspection path; Determine the path geometric characteristics based on the spatial topological structure and spatial layout information, and determine the second speed control parameters of the inspection robot in the approach section of each electrical equipment based on the second motion state information, the path geometric characteristics, and the safety constraint conditions; The inspection action of the inspection robot on each electrical device is determined according to the perception angle setting parameter and the second speed control parameter.

[0078] In this solution, the perception angle setting parameters may refer to the control parameters that require the inspection robot to adjust the orientation angle (such as pitch angle, yaw angle, and roll angle) of its sensor or camera when performing the perception or detection task of a certain electrical equipment to ensure the optimal viewing angle and obtain effective images, thermal imaging, infrared or other sensor data.

[0079] Path geometric features may refer to the geometric properties of each segment in the inspection path, including but not limited to path curvature, slope, corner radius, path width, obstacle clearance, turning radius, terrain height difference and other geometric data.

[0080] The second speed control parameter may be a speed control value set during the inspection process to ensure that the robot approaches various electrical equipment stably, safely, and efficiently, and typically includes dimensions such as forward speed, turning speed, and deceleration.

[0081] The inspection robot's second current position, i.e., its current three-dimensional coordinates (x, y, z), is obtained. Combined with the substation's spatial topology, this structure is typically represented as a directed graph with device nodes as vertices and traversable paths as edges. Based on this graph structure, a path search is performed using the A* (A-star) algorithm or the Dijkstra algorithm. Using the second current position as the starting point and the locations of the electrical equipment to be inspected as the target points in the path, the robot calculates the shortest inspection path that meets the traversable constraints. This path is represented as a sequence of continuous coordinate points, where each point represents a location reached by the robot in sequence.

[0082] Subsequently, an angular transformation is calculated based on the second posture angles (such as pitch, yaw, and roll) and the spatial layout of each target device. A coordinate transformation matrix (rotation matrix and transformation matrix) is used to determine the relative posture angle difference between the robot's current posture and the target device's observation direction. Based on this, Euler angle interpolation fitting or quaternion interpolation methods (such as SLERP) are used to calculate the robot's perception angle setting parameters for each device. This is the angle combination required to ensure that the sensor is facing the device and capturing the view.

[0083] Next, the generated inspection route's geometric structure is analyzed for each segment. Geometric features are extracted based on data such as curvature, turning radius, channel width, obstacle distribution, and height difference. This step is typically accomplished by performing curvature fitting (such as spline interpolation) on the path point set and extracting curve features using a sliding window method. Furthermore, spatial layout information (indicating equipment space occupancy, occlusion relationships, and equipment size) is integrated with topological constraints to determine whether there are any obstructions or access conflicts between devices.

[0084] Next, the path geometry is combined with the second motion state information (representing the robot’s current speed, acceleration, and angular velocity) and the substation’s preset safety constraints (such as the minimum safe distance from high-voltage equipment, the maximum allowable speed, and the safe obstacle avoidance radius). A speed planning algorithm (e.g., based on model predictive control (MPC) or the dynamic windowing algorithm (DWA)) is used to calculate the optimal travel speed for each section of the equipment approach path. The corresponding second speed control parameters are then obtained to ensure that the robot can safely and stably approach the equipment without collision or observation deviation.

[0085] During the execution of the inspection path, the system sets parameters based on the perception angle corresponding to the position of each target device, calculates the orientation adjustment angle of the robot's onboard sensors (such as cameras and infrared thermal imagers), and uses an attitude solution algorithm (such as a rotation matrix based on Euler angles or quaternion interpolation) to control the rotation of the robot's pan-tilt head or robotic arm so that its sensors are aimed at the key monitoring areas of the target electrical equipment (such as terminals, connection points, nameplates, etc.).

[0086] At the same time, based on the second speed control parameter corresponding to the path segment, the system plans the robot's movement speed as it approaches the electrical equipment, ensuring stable, jitter-free image acquisition or infrared measurement. If the equipment risk level is high or the surrounding environment is highly disturbed, the controller will decelerate, park, or dynamically slow down according to the speed control parameter to ensure data acquisition quality.

[0087] Ultimately, the system integrates the perception angle setting parameters and speed control parameters to generate a specific inspection action sequence, which includes posture adjustment actions (such as rotating the pan-tilt head to the set angles α, β, γ); sensor activation actions (such as starting image acquisition, infrared imaging, sound recording, etc.); motion control actions (such as slowing down to a distance L and stopping, and remaining still for T seconds); abnormality judgment actions (such as image clarity assessment and temperature stability judgment); and encapsulating these action instructions into control instruction packets and sending them to the robot execution controller.

[0088] In this solution, path planning, posture adjustment, speed control and action decision-making are finely decoupled and optimized in a coordinated manner to ensure that the inspection robot can complete accurate perception and data collection of electrical equipment at the optimal angle and speed under the premise of safety, efficiency and stability, thereby improving the coverage, accuracy and robustness of the inspection.

[0089] On the basis of the above technical solution, optionally, the system further includes a cleaning and retrying module, which is used to: If a cleaning failure message is received from the inspection robot, the second three-dimensional spatial coordinates of the failed foreign object and the abnormality type are determined; Determine whether the second three-dimensional space coordinates and the abnormality type meet a preset cleaning and retry condition, and if the preset cleaning and retry condition is met, update the first posture angle and the first motion state information of the inspection robot; Update the first end posture adjustment angle of the inspection robot that cleans the failed foreign object according to the updated first posture angle and the second three-dimensional space coordinates of the failed foreign object; updating a first speed control parameter of the inspection robot in a foreign object approaching section of a failed foreign object according to the updated first motion state information and the first foreign object operation requirement; updating the first cleaning execution parameter of the failed foreign object according to the updated first end posture adjustment angle, the updated first speed control parameter, and the first foreign object operation requirement, and sending the updated first cleaning execution parameter of the failed foreign object to the inspection robot; Accordingly, the system further includes a cleaning success module, which is configured to: If a cleaning success message is received from the inspection robot, a continue cleaning instruction is sent to the inspection robot, so that the inspection robot can clean the next foreign object according to the continue cleaning instruction; Accordingly, the system further includes a cleaning failure module, which is configured to: If the cleaning failure information sent by the inspection robot is received, the second three-dimensional spatial coordinates of the failed foreign object will be sent to the control center.

[0090] In this solution, the cleaning failure information may refer to the feedback data sent back by the inspection robot due to execution failure (such as abnormal mechanical movement, positioning deviation, substandard cleaning effect, etc.) during the process of performing a cleaning task of a foreign object.

[0091] The second three-dimensional spatial coordinate may refer to the three-dimensional position coordinate (x, y, z) of the foreign object obtained by repositioning the system after the cleaning fails, or by back-calculating the foreign object using the current posture of the robot.

[0092] The exception type refers to the system's classification of the cause of cleaning failure, which is used for subsequent decision support. This may include mechanical execution anomalies (such as a stuck robot arm or a grasping failure), target positioning deviations (such as point cloud drift), environmental interference (such as occlusion or strong reflections), and foreign object changes (such as displacement or fragmentation).

[0093] The preset cleaning retry conditions may refer to a set of rules defined in advance in the system that can trigger a cleaning retry, including that the foreign object risk level is higher than a threshold, the current failure cause is recoverable (such as angle deviation, insufficient cleaning force), the number of attempts is within the limit, and the surrounding environment allows re-action (no new obstacles).

[0094] Cleaning success information refers to the successful feedback information sent back by the inspection robot after completing the cleaning operation of a foreign object. It includes a success mark, the comparison results of the image / point cloud features after cleaning, the cleaning time / energy consumption data, and the success timestamp.

[0095] The continue cleaning instruction may refer to a task scheduling instruction issued by the system to the inspection robot based on the current execution status (such as a foreign object has been successfully cleaned), which is used to execute the cleaning path and action of the next foreign object.

[0096] The control center can refer to the central control system or remote management platform in the system.

[0097] Upon receiving a cleaning failure notification from the inspection robot, the system immediately extracts various sensor data from the failure record, including the target foreign object identifier, cleaning action execution feedback, execution timestamp, end-effector status, the robot's current posture data, and image feedback. By combining posture inversion at the time of failure with point cloud coordinate projection reconstruction, the system relocates the target foreign object's second 3D spatial coordinates within the current environment. Based on operational state analysis from the anomaly feedback (such as gripper closure status, cleaning arm force sensor response, and motor torque feedback), combined with image / point cloud feature changes (e.g., presence of foreign object edge features, contour integrity, and rebound in point cloud density), the system uses a set of anomaly type discrimination rules (e.g., insufficient cleaning force, positional offset, or occlusion) to determine the current anomaly type. A set of preset cleaning retry conditions is introduced; if these conditions are met, a recovery strategy is executed. Based on the current environmental state and the spatial relationship between the robot's end-effector posture and the target foreign object, the system uses forward kinematics and inverse solution algorithms to adjust the initial posture angle to calculate a new posture vector appropriate for the current obstacle environment. Combining the current motion state of the inspection robot (including joint angular velocity, vehicle body speed, etc.) with the first foreign object operation requirements (such as cleaning force requirements, contact stability, and angle range), the system uses a Bayesian optimization strategy or a rule-based speed constraint function to recalculate the robot's first speed control parameters in the failed foreign object approach segment to ensure its motion smoothness and positioning accuracy within the obstacle avoidance range. Based on the updated first end posture adjustment angle and first speed control parameters, combined with the settings of the cleaning action trajectory, cleaning method (such as wiping, blowing, grabbing), and execution duration in the first foreign object operation requirements, the first cleaning execution parameters of the target foreign object are reconstructed to form an action package containing the posture target, speed trajectory, action mode, and fault tolerance threshold. The cleaning execution parameters are then sent to the inspection robot to retry cleaning the target foreign object.

[0098] If the subsequent system receives a successful cleaning message, it will take out the next foreign object task from the cleaning task scheduling queue and generate a new continue cleaning instruction based on its spatial position, operation level and the current position of the robot. The instruction contains the target number, path information, estimated execution time and other content for the robot to complete the next foreign object cleaning process.

[0099] If the cleaning failure information is received again and the abnormality type or spatial status does not meet the cleaning retry conditions, the system will package the second three-dimensional spatial coordinates of the failed foreign object and the relevant abnormality cause, image data and failure level into an abnormality report, and send it to the control center via TCP communication or MQTT message push to trigger manual review or remote intervention process.

[0100] In this solution, the success rate of the robot's cleaning tasks in complex scenarios is improved, and task interruptions caused by small-scale deviations or minor disturbances are avoided; the tasks are ensured to be executed safely and reliably within the automated closed loop, thereby enhancing the system's robustness and autonomous decision-making capabilities.

[0101] Example 2 Figure 2 This is a schematic diagram of the structure of the substation autonomous inspection and foreign matter cleaning collaborative control system provided in the second embodiment of this application. Figure 2 As shown, specifically including the following: The system further includes a residue identification module 106, which is configured to: Reacquiring the substation's 2D image data, LiDAR point cloud data, and multimodal sensor data, calculating a difference map between the 2D image data before and after cleaning using a structural similarity algorithm, and performing connected region analysis on the difference map to determine image disturbance areas. Perform spatial registration processing on the LiDAR point cloud data before and after cleaning to determine the point cloud disturbance area; Verifying the spatial consistency of the image disturbance region and the point cloud disturbance region; if they are consistent, confirming the image disturbance region as a candidate residual foreign matter region; Perform target detection on the candidate residual foreign matter area to identify whether there is residual foreign matter.

[0102] In this embodiment, the structural similarity algorithm can be an image quality assessment method used to measure the similarity between two images in terms of structure, brightness, and contrast. Compared with simple pixel differences, SSIM is closer to the human eye's perception of image changes.

[0103] A difference map may be a representation of image differences generated by comparing two-dimensional image data before and after cleaning, and used to highlight locations in the image where changes have occurred.

[0104] The image disturbance region may refer to a continuously changing region extracted by connected region analysis in the difference map, representing a location in the image that is significantly disturbed.

[0105] The point cloud disturbance area can be the location area where the point cloud structure is found to have changed significantly after the lidar point cloud data before and after cleaning is compared through spatial registration (such as ICP algorithm).

[0106] The candidate residual foreign body region can be the intersection area where the image disturbance region and the point cloud disturbance region are highly consistent in space, indicating that the position has changed in both the image and physical space.

[0107] Residual foreign matter may refer to target foreign matter that has not been completely removed after a cleaning task has been performed.

[0108] The substation's 2D image data, LiDAR point cloud data, and multimodal sensor data can be re-acquired. Image comparison uses a structural similarity algorithm, which compares local window regions of the image block by block based on brightness, contrast, and structure, outputting a similarity score map. This similarity score map is then pixel-inverted and thresholded to generate a difference map, indicating locations with significant structural changes. The system then clusters pixels in this difference map using a connected component analysis algorithm to identify contiguous regions of high difference. These regions are then extracted as candidate image disturbance regions, representing areas of potential cleanup or physical structural changes. The system also performs registration on the LiDAR point cloud data collected before and after cleaning, using algorithms such as ICP or its improved versions (such as NDT or Go-ICP) to align the two point clouds in 3D space. The registration results are compared for features such as local point density changes and geometric structure differences, identifying regions with significant changes as point cloud disturbance regions. Spatial consistency verification is then performed: coordinate transformation is used to map the image disturbance regions into 3D space, and spatial overlap is checked with the point cloud disturbance regions. If there is a high degree of spatial overlap between the two (for example, the center distance is less than a preset threshold, and the IoU is greater than a certain threshold), the area is considered a reliable candidate for residual foreign matter. The system performs target detection on these candidate areas using deep learning methods, such as YOLOv5, Mask R-CNN, or RetinaNet. The input is a cropped fragment of the original image within the candidate area, and the output is the detection result of the presence of residual foreign matter, as well as the location and confidence of the target box. If the detection result indicates the presence of residual foreign matter, this information will be annotated and reported to the system control center, triggering subsequent compensation cleaning or manual review.

[0109] In this embodiment, the accuracy and robustness of residual foreign matter detection are improved; by combining the structural similarity algorithm with the spatial registration technology, sensitive capture of small disturbances is achieved, ensuring timely detection of cleaning omissions or environmental anomalies, and enhancing the system's autonomous monitoring capabilities and safety assurance level.

[0110] On the basis of the above technical solution, optionally, the system further includes a local cleaning module, which is used to: If there is residual foreign matter, obtaining the third three-dimensional spatial coordinates and the second foreign matter visual feature description of the residual foreign matter, inputting the third three-dimensional spatial coordinates, the second foreign matter visual feature description, the spatial layout information, and the re-acquired multimodal sensor data into a preset foreign matter risk level scoring model to obtain a second foreign matter risk level score for the residual foreign matter; Determine the second foreign object operation requirements for each foreign object based on the second foreign object risk level score and the preset cleaning strategy library; Obtain the third current position, third posture angle and third motion state information of the inspection robot, determine the local cleaning path of the inspection robot and the second cleaning execution parameters for residual foreign objects based on the third three-dimensional space coordinates, the third current position, the third posture angle, the third motion state information and the second foreign object operation requirements, and send the local cleaning path and the second cleaning execution parameters for residual foreign objects to the inspection robot.

[0111] In this solution, the third three-dimensional spatial coordinate may refer to the residual foreign matter re-identified after cleaning, and the three-dimensional coordinate of the residual foreign matter in the substation space obtained by image and point cloud fusion registration and positioning.

[0112] The second foreign object visual feature description can refer to the visual feature information extracted from the two-dimensional image collected after cleaning, such as color histogram, texture features (such as LBP), depth features (such as embedding vectors extracted by CNN), etc. This description is used for comparison with historical foreign object features, risk analysis, and decision-making and scheduling.

[0113] The second foreign object risk level score can refer to the risk level of the residual foreign object, predicted using a preset foreign object risk level scoring model based on the third 3D spatial coordinates, the second foreign object visual feature description, spatial layout information, and multimodal sensor data. This is used to determine whether the object poses a safety hazard or is a priority for emergency cleanup.

[0114] The second foreign matter operation requirement may be an operation specification required for residual foreign matter obtained by mapping the second foreign matter risk level score with a preset cleaning strategy library.

[0115] The third current position may be the position of the robot in the substation coordinate system.

[0116] The third posture angle can be, for example, an Euler angle or a quaternion, representing the overall orientation of the robot.

[0117] The third motion state information may be, for example, a velocity vector, acceleration information, etc., reflecting the dynamic state of the robot.

[0118] The local cleaning path may be a short-distance path planned from the third current position to the third three-dimensional space coordinate, taking into account obstacle avoidance, safety margins, etc.

[0119] The second cleaning execution parameter may be a cleaning action parameter generated according to the third posture angle, the third motion state information and the second foreign object operation requirement, such as the end effector trajectory, speed, force control parameters, etc.

[0120] When the system detects spatial consistency between the image disturbance area and the point cloud disturbance area, and further confirms the presence of residual foreign matter through the target detection algorithm, the system first uses a spatial reconstruction algorithm (such as triangulation based on multi-view geometry or a point cloud segmentation and positioning algorithm) to locate the residual foreign matter. Combining the correspondence between the center point of the visual area extracted from the cleaned image and the local clustering area of ​​the lidar point cloud, the system estimates and calculates the third 3D spatial coordinates of the foreign matter in the substation's 3D coordinate system. Simultaneously, the system extracts visual description information of the foreign matter area from the cleaned image to construct a second visual feature description of the foreign matter. This feature vector can be represented by the intermediate layer embedding vector output by a pre-trained convolutional neural network (such as ResNet or EfficientNet), capturing color, edge, texture, and shape information. Utilizing the previously constructed foreign object risk rating model, the third 3D spatial coordinates, the second foreign object visual feature description, the spatial layout information of the current equipment area (including coordinates of surrounding high-voltage equipment, cable routing, and cleanable airspace), and newly acquired multimodal sensor data (such as temperature and humidity, wind speed, infrared radiation, and electromagnetic intensity) are integrated to form a multidimensional input feature vector. This vector is then input into the foreign object risk rating model, which uses ensemble learning (such as XGBoost) or a neural network regression architecture to predict and output a second foreign object risk rating for the residual foreign object. The value range can be normalized to the [0, 1] range. Based on the mapping relationship between this risk score and the system's pre-set cleaning strategy library, the system automatically searches for and generates the second foreign object action requirements corresponding to the current risk level. These requirements include the required cleaning mode (such as mechanical wiping, suction, or air jet), force setting, safety margin limits, and actuator configuration. The system acquires the inspection robot's current third position (obtained by a positioning system such as RTK-GPS or UWB+IMU fusion), third attitude angle (derived from Euler angles or quaternions using gyroscopes or visual inertial navigation), and third kinematic state information (such as acceleration vectors read from accelerometers and wheel encoder velocities). The system combines the third current position with the target's third 3D spatial coordinates and executes a heuristic path planning algorithm (such as A*, RRT*, or DWA local planner) within the obstacle avoidance grid map to plan a shortest or optimal local cleaning path. This path planning considers known obstacle data and safety constraints in the spatial layout, such as maintaining a certain distance from high-voltage equipment and safe operating area restrictions. Furthermore, the planned path geometry, third attitude angle, and second foreign object handling requirements are combined with the kinematic model and velocity mapping function of the end effector in the robot controller to calculate the required trajectory velocity and attitude transition method for the cleaning end effector as it approaches the remaining foreign object. This ultimately generates the second cleaning execution parameters used to control the robot, including path segment velocity, end attitude adjustment angle, action sequence, and execution duration.The system packages the generated local cleaning path and the corresponding second cleaning execution parameters in the form of a control instruction sequence, and sends it to the task control unit of the inspection robot, driving it to return to the residual foreign matter area according to the path and action requirements and complete the secondary cleaning operation.

[0121] In this solution, the compensation robustness, task closure and system intelligence level of the cleaning operation are improved, effectively ensuring the continuous cleanliness and operational safety of the substation environment.

[0122] On the basis of the above technical solution, optionally, the system further includes an abnormality reporting module, and the abnormality reporting module is used to: If there is no residual foreign matter, edge change detection is performed on the image disturbance area to determine whether there is any image structural abnormality; If there are abnormal image structures, density change analysis methods are used to identify whether there are abnormal point cloud distributions based on the LiDAR point cloud data before and after cleaning. If there is an abnormal point cloud distribution, an abnormality reporting information package is generated based on the re-acquired two-dimensional image data, lidar point cloud data and multimodal sensor data of the substation, and the abnormality reporting information package is transmitted to the control center.

[0123] In this solution, the abnormality reporting information package may refer to a data set automatically generated when no residual foreign matter is detected but structural and spatial anomalies are found in the image or point cloud data, which contains abnormal information, positioning information, and auxiliary analysis content. It is used to report abnormal events to the control center to facilitate subsequent manual review or automatic intervention.

[0124] If no residual foreign matter is detected, the system will conduct further structural anomaly analysis on the disturbed area. First, using 2D image data of the substation collected before and after cleaning, an edge detection algorithm (such as the Canny algorithm or edge gradient detection based on the Sobel operator) is used to extract edges from the disturbed area. By comparing the contour morphology changes of the image edges before and after cleaning, the system detects any significant image structural anomalies such as contour breakage, edge displacement, and changes in closure. If the magnitude of the edge change exceeds a preset threshold, the system determines that the disturbed area has a structural anomaly. The system further performs spatial alignment processing on the LiDAR point cloud data acquired before and after cleaning, using spatial registration technology based on the ICP (Iterative Closest Point) algorithm to align the point clouds before and after cleaning to the same spatial coordinate system. After registration, the point cloud voxel grid downsampling and statistical density analysis methods are used to spatially compare the point density within each voxel. If the point cloud density in a spatial area changes significantly, such as the appearance of holes in a previously dense area or the sudden increase in density in a previously sparse area, the area is considered to have a point cloud distribution anomaly. If both image structure anomalies and point cloud distribution anomalies are confirmed, the system identifies the area as an abnormal area that may pose a potential safety risk or structural disturbance. The system then integrates the newly collected 2D substation image data, LiDAR point cloud data, and currently collected multimodal sensor data (such as ambient temperature and humidity, vibration, and electromagnetic interference levels) to automatically construct a complete anomaly reporting package. This package includes the location coordinates of the image disturbance area, an image structure anomaly edge map, a point cloud density change heat map, a summary of the current sensor data, and a label for the identified anomaly type (such as increased occlusion, ground uplift, equipment tilt). Combined with the electrical equipment distribution map, a risk level score is generated and ultimately transmitted to the control center in real time via the system communication protocol.

[0125] In this solution, closed-loop verification of cleaning effects and risk compensation reporting are achieved, thereby significantly improving the reliability and safety assurance capabilities of the inspection system.

[0126] Example 3 Figure 3 This is a flow chart of the coordinated control method for substation autonomous inspection and foreign matter cleaning provided in the third embodiment of the present application. Figure 3 As shown, the specific steps include: S301, obtaining two-dimensional image data, lidar point cloud data, and multimodal sensor data of the substation, and using a convolutional neural network target detection algorithm to identify the two-dimensional coordinate position of each foreign object in the two-dimensional image data and a first foreign object visual feature description; S302, performing spatial registration processing based on the laser radar point cloud data and the two-dimensional coordinate position to obtain first three-dimensional spatial coordinates of each foreign object; S303, obtaining spatial layout information of the electrical equipment, inputting the first three-dimensional spatial coordinates, the first foreign object visual feature description, the spatial layout information, and the multimodal sensor data into a preset foreign object risk level scoring model to obtain a first foreign object risk level score for each foreign object; S304: Determine a first foreign object operation requirement for each foreign object based on the first foreign object risk level score and a preset cleaning strategy library, obtain a first current position, a first posture angle, and a first motion state information of the inspection robot, determine a cleaning path for the inspection robot and first cleaning execution parameters for each foreign object based on the first three-dimensional space coordinates, the first current position, the first posture angle, the first motion state information, and the first foreign object operation requirement, and send the cleaning path and the first cleaning execution parameters for each foreign object to the inspection robot; S305. If the cleaning completion information of the inspection robot is received, the second current position, second posture angle, and second motion state information of the inspection robot are obtained, as well as the spatial topology structure and safety constraints of the substation are obtained. The inspection path of the inspection robot and the inspection actions on each electrical equipment are determined according to the second current position, second posture angle, second motion state information, spatial topology structure, spatial layout information and safety constraints, and the inspection path and the inspection actions on each electrical equipment are sent to the inspection robot.

[0127] The embodiment of the present application provides a method for coordinated control of autonomous inspection and foreign matter cleaning of substations, which corresponds to the systems provided in the above embodiments and has corresponding execution processes and beneficial effects, and will not be repeated here.

[0128] Example 4 like Figure 4 As shown, an embodiment of the present application also provides an electronic device 400, including a processor 401, a memory 402, and a program or instruction stored in the memory 402 and executable on the processor 401. When the program or instruction is executed by the processor 401, each process of the above-mentioned embodiment of the method for the coordinated control system of autonomous inspection and foreign matter cleaning of substations is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be described here.

[0129] It should be noted that the electronic devices in the embodiments of the present application include the mobile electronic devices and non-mobile electronic devices mentioned above.

[0130] Example 5 An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, each process of the above-mentioned cable installation process based on the tension adaptive control system embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0131] The processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes a computer-readable storage medium, such as a computer read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk.

[0132] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or system comprising the element. In addition, it should be noted that the scope of the methods and systems in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved. For example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0133] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better embodiment. Based on this understanding, the technical solution of this application, or the part that contributes to the existing technology, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of this application.

[0134] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are within the protection of this application.

[0135] The above are only preferred embodiments of the present application and the technical principles employed. The present application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that are possible for those skilled in the art will not depart from the scope of protection of the present application. Therefore, although the present application has been described in more detail through the above embodiments, the present application is not limited to the above embodiments and may include more other equivalent embodiments without departing from the concept of the present application. The scope of the present application is determined by the scope of the claims.

Claims

1. A substation autonomous inspection and foreign matter cleaning collaborative control system, characterized in that: The system comprises: a multimodal detection module, configured to acquire two-dimensional image data, lidar point cloud data, and multimodal sensor data of the substation, and employ a convolutional neural network target detection algorithm to identify the two-dimensional coordinate position of each foreign object in the two-dimensional image data and a first foreign object visual feature description; A spatial positioning module is used to perform spatial registration processing based on the laser radar point cloud data and the two-dimensional coordinate position to obtain the first three-dimensional spatial coordinates of each foreign object; a risk decision module, configured to obtain spatial layout information of the electrical equipment, input the first three-dimensional spatial coordinates, the first foreign object visual feature description, the spatial layout information, and the multimodal sensor data into a preset foreign object risk level scoring model, and obtain a first foreign object risk level score for each foreign object; a cleaning execution module, configured to determine a first foreign object operation requirement for each foreign object based on the first foreign object risk level score and a preset cleaning strategy library, obtain a first current position, a first posture angle, and a first motion state information of the inspection robot, determine a cleaning path of the inspection robot and first cleaning execution parameters for each foreign object based on the first three-dimensional space coordinates, the first current position, the first posture angle, the first motion state information, and the first foreign object operation requirement, and send the cleaning path and the first cleaning execution parameters for each foreign object to the inspection robot; The inspection execution module is used to obtain the second current position, second posture angle, second motion state information of the inspection robot upon receiving the cleaning completion information of the inspection robot, as well as the spatial topology structure and safety constraints of the substation, and determine the inspection path of the inspection robot and the inspection actions on each electrical equipment based on the second current position, second posture angle, second motion state information, spatial topology structure, spatial layout information and safety constraints, and send the inspection path and inspection actions on each electrical equipment to the inspection robot.

2. The substation autonomous inspection and foreign matter cleaning coordinated control system according to claim 1 is characterized in that: The cleaning execution module is further used to: Calculate, based on the first current position and the first three-dimensional spatial coordinates, a sub-cleaning path of the inspection robot from the first current position to each foreign object using a path planning algorithm, and determine a cleaning path of the inspection robot based on the sub-cleaning path of the inspection robot from the first current position to each foreign object; Calculating a first end posture adjustment angle of the inspection robot for cleaning foreign objects according to the first posture angle and the first three-dimensional space coordinate; Determining a first speed control parameter of the inspection robot in each foreign object approach section according to the first motion state information and the first foreign object operation requirement; The first cleaning execution parameter of each foreign object is determined according to the first end posture adjustment angle, the first speed control parameter and the first foreign object operation requirement.

3. The substation autonomous inspection and foreign matter cleaning coordinated control system according to claim 1 is characterized in that: The inspection execution module is further used to: Determining an inspection path of the inspection robot according to the second current position and the spatial topological structure; Determine, based on the second posture angle and the spatial layout information, the perception angle setting parameters for the inspection robot to reach the location of each electrical device in the inspection path; Determine the path geometric characteristics based on the spatial topological structure and spatial layout information, and determine the second speed control parameters of the inspection robot in the approach section of each electrical equipment based on the second motion state information, the path geometric characteristics, and the safety constraint conditions; The inspection action of the inspection robot on each electrical device is determined according to the perception angle setting parameter and the second speed control parameter.

4. The substation autonomous inspection and foreign matter cleaning coordinated control system according to claim 1 is characterized in that: The system further includes a cleaning and retrying module, which is configured to: If a cleaning failure message is received from the inspection robot, the second three-dimensional spatial coordinates of the failed foreign object and the abnormality type are determined; Determine whether the second three-dimensional space coordinates and the abnormality type meet a preset cleaning and retry condition, and if the preset cleaning and retry condition is met, update the first posture angle and the first motion state information of the inspection robot; Update the first end posture adjustment angle of the inspection robot that cleans the failed foreign object according to the updated first posture angle and the second three-dimensional space coordinates of the failed foreign object; updating a first speed control parameter of the inspection robot in a foreign object approaching section of a failed foreign object according to the updated first motion state information and the first foreign object operation requirement; updating the first cleaning execution parameter of the failed foreign object according to the updated first end posture adjustment angle, the updated first speed control parameter, and the first foreign object operation requirement, and sending the updated first cleaning execution parameter of the failed foreign object to the inspection robot; Accordingly, the system further includes a cleaning success module, which is configured to: If a cleaning success message is received from the inspection robot, a continue cleaning instruction is sent to the inspection robot, so that the inspection robot can clean the next foreign object according to the continue cleaning instruction; Accordingly, the system further includes a cleaning failure module, which is configured to: If the cleaning failure information sent by the inspection robot is received, the second three-dimensional spatial coordinates of the failed foreign object will be sent to the control center.

5. The substation autonomous inspection and foreign matter cleaning coordinated control system according to claim 1 is characterized in that: The system further includes a residue identification module, wherein the residue identification module is configured to: Reacquiring the substation's 2D image data, LiDAR point cloud data, and multimodal sensor data, calculating a difference map between the 2D image data before and after cleaning using a structural similarity algorithm, and performing connected region analysis on the difference map to determine image disturbance areas. Perform spatial registration processing on the LiDAR point cloud data before and after cleaning to determine the point cloud disturbance area; Verifying the spatial consistency of the image disturbance region and the point cloud disturbance region; if they are consistent, confirming the image disturbance region as a candidate residual foreign matter region; Perform target detection on the candidate residual foreign matter area to identify whether there is residual foreign matter.

6. The substation autonomous inspection and foreign matter cleaning coordinated control system according to claim 5 is characterized in that: The system further includes a local cleaning module, which is configured to: If there is residual foreign matter, obtaining the third three-dimensional spatial coordinates and the second foreign matter visual feature description of the residual foreign matter, inputting the third three-dimensional spatial coordinates, the second foreign matter visual feature description, the spatial layout information, and the re-acquired multimodal sensor data into a preset foreign matter risk level scoring model to obtain a second foreign matter risk level score for the residual foreign matter; Determine the second foreign object operation requirements for each foreign object based on the second foreign object risk level score and the preset cleaning strategy library; Obtain the third current position, third posture angle and third motion state information of the inspection robot, determine the local cleaning path of the inspection robot and the second cleaning execution parameters for residual foreign objects based on the third three-dimensional space coordinates, the third current position, the third posture angle, the third motion state information and the second foreign object operation requirements, and send the local cleaning path and the second cleaning execution parameters for residual foreign objects to the inspection robot.

7. The substation autonomous inspection and foreign matter cleaning coordinated control system according to claim 5 is characterized in that: The system further includes an abnormality reporting module, which is configured to: If there is no residual foreign matter, edge change detection is performed on the image disturbance area to determine whether there is any image structural abnormality; If there are abnormal image structures, density change analysis methods are used to identify whether there are abnormal point cloud distributions based on the LiDAR point cloud data before and after cleaning. If there is an abnormal point cloud distribution, an abnormality reporting information package is generated based on the re-acquired two-dimensional image data, lidar point cloud data and multimodal sensor data of the substation, and the abnormality reporting information package is transmitted to the control center.

8. A method for coordinated control of autonomous inspection and foreign matter cleaning of a substation, characterized in that: The method comprises: Acquire two-dimensional image data, lidar point cloud data, and multimodal sensor data of the substation, and use a convolutional neural network target detection algorithm to identify the two-dimensional coordinate position of each foreign object in the two-dimensional image data and a first foreign object visual feature description; Performing spatial registration processing based on the laser radar point cloud data and the two-dimensional coordinate position to obtain the first three-dimensional spatial coordinates of each foreign object; Obtaining spatial layout information of the electrical equipment, inputting the first three-dimensional spatial coordinates, the first foreign object visual feature description, the spatial layout information, and the multimodal sensor data into a preset foreign object risk level scoring model to obtain a first foreign object risk level score for each foreign object; Determine a first foreign object operation requirement for each foreign object based on the first foreign object risk level score and a preset cleaning strategy library, obtain a first current position, a first posture angle, and a first motion state information of the inspection robot, determine a cleaning path of the inspection robot and a first cleaning execution parameter for each foreign object based on the first three-dimensional space coordinates, the first current position, the first posture angle, the first motion state information, and the first foreign object operation requirement, and send the cleaning path and the first cleaning execution parameter for each foreign object to the inspection robot; If the cleaning completion information of the inspection robot is received, the second current position, second posture angle, and second motion state information of the inspection robot are obtained, as well as the spatial topology structure and safety constraints of the substation are obtained, and the inspection path of the inspection robot and the inspection actions on each electrical equipment are determined according to the second current position, second posture angle, second motion state information, spatial topology structure, spatial layout information and safety constraints, and the inspection path and inspection actions on each electrical equipment are sent to the inspection robot.

9. An electronic device, characterized in that: It includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, the steps of the coordinated control method for autonomous inspection and foreign matter cleaning of substations as described in claim 8 are implemented.

10. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the steps of the substation autonomous inspection and foreign matter cleaning coordinated control method as claimed in claim 8 are implemented.

Images