Self-adaptive inspection control method and system for electric inspection robot

Abstract

The invention relates to the technical field of inspection control, and provides a self-adaptive inspection control method and system for an electric inspection robot. Space-time alignment and fusion modeling are carried out on the multi-modal environment data to obtain a three-dimensional environment model, and dynamic path planning is carried out on the three-dimensional environment model according to the position information of each target robot and a target inspection point list to generate an anti-interference elastic path. Performing motion parameter dynamic optimization by combining the real-time dynamic parameters of each target robot to generate an anti-interference control instruction, and calculating the detection distance between each target robot and the to-be-detected target according to the position information; and when the detection distance is smaller than a distance threshold, performing anomaly detection and classification on the to-be-detected target to obtain a corresponding fault level label, and performing multi-machine cooperative task allocation on the fault level label according to the global scheduling information to obtain an optimal inspection queue. Through real-time planning, online optimization and multi-machine cooperation, the safety, reliability and inspection efficiency of the electric power inspection robot are improved.

Description

Technical Field

[0001] The present application relates to the field of patrol control technology, and in particular to an adaptive patrol control method and system for a power patrol robot. Background Art

[0002] As the lifeline of the national economy, the safe and stable operation of power lines is directly related to the continuity of social life and industrial production. Traditional inspection methods rely heavily on manual labor or simple semi-automated equipment. Limited by personnel vision, climatic conditions, and inspection frequency, they are prone to missed inspections and misjudgments, making it difficult to promptly detect hidden dangers such as aging conductors, contaminated insulators, and hot spots. As the scale and complexity of power grids expand, establishing an all-weather, full-area, intelligent inspection system can not only reduce personal risks and operation and maintenance costs, but also improve early warning efficiency and processing accuracy, ensuring the long-term stable operation of the power system.

[0003] Existing control methods for power inspection robots are mostly based on pre-set waypoints or fixed patrol paths, combined with traditional PID or model predictive control to regulate robot motion. Some solutions incorporate vision or laser sensors for local obstacle avoidance, but these typically only detect obstacles on a two-dimensional plane, react slowly to environmental changes, and often rely on offline algorithms for path replanning. Overall, these methods emphasize the stability and accuracy of single-machine motion, but neglect the deep integration of multimodal environmental information and the scheduling optimization of collaborative multi-machine collaboration. Summary of the Invention

[0004] In view of this, the present application provides an adaptive inspection control method and system for an electric power inspection robot to solve the problem of low inspection efficiency and reliability caused by single-dimensional or low-dimensional perception and control methods.

[0005] A first aspect of the present application provides an adaptive inspection control method for a power inspection robot, the method comprising: Perform spatiotemporal alignment and fusion modeling on the multimodal environmental data in the preset area to obtain a three-dimensional environmental model; Based on the real-time collected position information of each target robot and a preset target inspection point list, dynamic path planning is performed in the three-dimensional environment model to generate an anti-interference elastic path; Dynamically optimize motion parameters based on the collected real-time dynamic parameters of each target robot and the anti-interference elastic path to generate anti-interference control instructions; Calculating the detection distance between each target robot and the target to be detected on the anti-interference elastic path based on the position information collected in real time; When the detection distance is less than a preset distance threshold, performing abnormality detection and classification on the target to be detected to obtain a fault level label corresponding to the target to be detected; The fault level labels are assigned to multi-machine collaborative tasks according to the global scheduling information of all target robots to obtain the optimal inspection queue.

[0006] In an optional embodiment, performing spatiotemporal alignment and fusion modeling on the multimodal environment data in the preset area to obtain a three-dimensional environment model includes: Performing time domain alignment on the multimodal environmental data within a preset area according to the timestamps of each data to obtain synchronized environmental data; Performing power equipment identification and preliminary anomaly detection on the synchronized environmental data through a preset CNN network to obtain obstacle labels and obstacle probabilities; spatially fusing the synchronized environmental data according to an extrinsic calibration matrix and a coordinate system of the multimodal environmental data to obtain a three-dimensional grid map; The obstacle labels and the obstacle probabilities are marked on the three-dimensional grid map to obtain the three-dimensional environment model.

[0007] In an optional embodiment, the target inspection point list includes position information of the points to be inspected, and performing dynamic path planning in the three-dimensional environment model based on the real-time collected position information of each target robot and the preset target inspection point list to generate an anti-interference elastic path includes: Marking a path node set in a three-dimensional environment model based on the real-time collected position information of each target robot and the position information of the inspection point; Using a preset improved A* algorithm, paths between adjacent path nodes are evaluated according to a preset total interference force and the obstacle probability to obtain each path and a corresponding cost value; Selecting paths between adjacent path nodes according to the cost value to obtain a global inspection path framework; Calculating target attraction and obstacle repulsion based on the coordinate points in the global inspection path framework, the location information of the points to be inspected, and the centers of the grid cells with high obstacle probability along the way; The global inspection path framework is smoothly corrected according to the target attraction and the obstacle repulsion by a preset artificial potential field model to generate the anti-interference elastic path.

[0008] In an optional embodiment, the dynamically optimizing motion parameters based on the collected real-time dynamic parameters of each target robot and the anti-interference elastic path to generate the anti-interference control instruction includes: Real-time collection of each target robot's real-time speed, current acceleration, current environment 3D point cloud data and current position information; Determining an expected acceleration and a target path node based on the current position information and the anti-interference elastic path, and performing interference force estimation and path deviation calculation based on the expected acceleration, the target path node, the current acceleration, and the current environmental three-dimensional point cloud data to update the total interference force and obtain a path deviation value; Performing real-time adjustment on PID control parameters according to the total disturbance force and the path deviation value to update the proportional coefficient, the integral coefficient, and the differential coefficient; The motor speed and steering angle are calculated according to the proportional coefficient, the integral coefficient, the differential coefficient, the real-time speed, and the path deviation value to generate continuous anti-interference control instructions.

[0009] In an optional embodiment, the step of calculating the detection distance between each target robot and the target to be detected on the anti-interference elastic path based on the real-time collected position information includes: Collecting the current position information of each target robot in real time, and locating the coordinate information of the corresponding target to be detected on the anti-interference elastic path according to the position information; Calculating the detection distance between the current position of each target robot and the target to be detected based on the position information and the coordinate information using a preset three-dimensional space distance measurement method; The detection distance is compared with the distance threshold to determine whether the target robot enters a preset device detection area.

[0010] In an optional embodiment, performing abnormality detection and classification on the target to be detected to obtain a fault level label corresponding to the target to be detected includes: Performing device detection image acquisition on the target to be detected to obtain an infrared thermal image temperature matrix and a visible light characteristic image; Performing fault feature recognition and extraction on the infrared thermal image temperature matrix and the visible light feature image through a preset CNN model to obtain a fault feature set; The fault feature set is classified into fault levels using a preset IAOA-BP anomaly detection model to obtain a fault level label corresponding to the target to be detected.

[0011] In an optional embodiment, performing multi-machine collaborative task allocation on the fault level labels according to the global scheduling information of all target robots to obtain the optimal inspection queue includes: Summarize the current task status, remaining task list, location distribution and resource load of all target robots in the preset area to obtain the global scheduling information; Constructing a scheduling optimization model according to the fault level label, the target distribution density in the three-dimensional environment model, and the global scheduling information; The scheduling optimization model is dynamically solved according to a preset branch and bound algorithm to obtain the optimal inspection queue.

[0012] A second aspect of the present application provides an adaptive inspection control device for an electric power inspection robot, the device comprising: The environmental perception module is used to perform spatiotemporal alignment and fusion modeling of multimodal environmental data in a preset area to obtain a three-dimensional environmental model; A path planning module is used to perform dynamic path planning in the three-dimensional environment model based on the real-time collected position information of each target robot and a preset target inspection point list to generate an anti-interference elastic path; A motion correction module is used to dynamically optimize the motion parameters based on the collected real-time dynamic parameters of each target robot and the anti-interference elastic path to generate anti-interference control instructions; A distance monitoring module is used to calculate the detection distance between each target robot and the target to be detected on the anti-interference elastic path based on the position information collected in real time; a fault detection module, configured to perform abnormality detection and classification on the target to be detected when the detection distance is less than a preset distance threshold, so as to obtain a fault level label corresponding to the target to be detected; The multi-machine coordination module is used to perform multi-machine collaborative task allocation on the fault level labels according to the global scheduling information of all target robots to obtain the optimal inspection queue.

[0013] The third aspect of the present application provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the computer program, the steps of the adaptive inspection control method of the power inspection robot as described above are implemented.

[0014] A fourth aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the adaptive inspection control method of the power inspection robot as described above.

[0015] In summary, this application has at least the following beneficial technical effects: 1. Dynamic path planning is performed on the 3D environment model based on real-time coordinates and a preset inspection point list. The path can be quickly replanned when the environment changes or external interference occurs, maintaining the continuity and safety of the inspection task.

[0016] 2. Anti-interference control instructions can actively compensate for external disturbances, making the robot movement more stable, thereby improving the data quality of the detection equipment.

[0017] 3. Incorporating the fault level labels of each target to be inspected into global scheduling not only prioritizes high-risk or high-level faults, but also dynamically allocates inspection tasks based on the robot's current load, distance, remaining power, and other resource conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 This is a flow chart of an adaptive inspection control method for a power inspection robot provided in an embodiment of the present application; Figure 2 This is a functional module diagram of an adaptive inspection control device for an electric power inspection robot provided in an embodiment of the present application; Figure 3 It is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0021] like Figure 1 FIG2 is a flowchart of an adaptive inspection control method for an electric power inspection robot according to an embodiment of the present application. The adaptive inspection control method for an electric power inspection robot according to an embodiment of the present application includes the following steps.

[0022] Step S1: Perform spatiotemporal alignment and fusion modeling on the multimodal environmental data in a preset area to obtain a three-dimensional environmental model.

[0023] First, to obtain a high-precision 3D environment model, the multimodal raw data from the visible light camera, infrared thermal imager, lidar, and inertial measurement unit must first be strictly time-aligned according to the timestamps, ensuring that all sensors observe the same environmental state under the same time reference. Specifically, by reading the microsecond timestamp marked by the hardware clock in each data packet, the data from each sensor is interpolated to the reference time t0. This step ensures that the visible light image, infrared thermal image, and point cloud reflect the same spatial scene at the same time. For example, if the camera frame is collected at t0 and the radar frame is collected at t0+Δt, the point cloud data at time t0 is generated through linear or high-order interpolation, and the interpolation error is controlled within 1ms to avoid registration deviations caused by dynamic objects or the mobile robot itself.

[0024] Next, the synchronized environmental data after time domain alignment is input into a pre-trained convolutional neural network (CNN). The network is used to identify the geometric and texture features of the power equipment. At the same time, the pixel temperature values ​​in the infrared temperature matrix are combined to complete the preliminary high-temperature area detection, thereby generating temperature anomaly labels and equipment type labels. The obstacle probability is obtained by calculating the projection probability of the point cloud density at the same viewing angle. This not only automatically distinguishes equipment types such as transmission towers, insulators, and conductors, but also marks possible hidden danger areas in the environmental model. For example, local areas where the infrared image temperature exceeds 80°C are marked as "high temperature risk areas", and pixel areas where significant cracks are detected by CNN are marked as "crack risk areas". Subsequently, the multimodal data needs to be spatially fused. First, the camera coordinate system, infrared instrument coordinate system, and LiDAR coordinate system are aligned to the global map coordinate system based on the external parameter matrix obtained during the calibration process. Then, the interpolated and aligned point cloud is mapped to the voxel grid to form a hollow three-dimensional grid, and the corresponding obstacle probability value P is stored in each voxel unit. occ 、Temperature abnormality label L T With device type label L D. In order to balance the freshness and spatial reliability of the observation, the embodiment of the present application calculates a spatiotemporal consistency weight C(v) for each voxel grid point v. Specifically, a "time attenuation factor" is calculated based on the difference Δt between the observation time and the reference time of each grid point. That is, Δt is divided by the pre-selected time bandwidth constant Γ, and then the square is taken and substituted into the negative exponential function of the natural exponential. In this way, the closer the observation is to the reference time (that is, the smaller Δt is), the closer its exponential value is to 1, and the farther away the observation is, the faster its exponential value decays to approach 0. Secondly, the same treatment is performed on each grid point based on its spatial distance Δd from the sensor position. That is, Δd is first divided by the spatial bandwidth constant d0 and squared, and then substituted into the same form of negative exponential function to obtain a "spatial attenuation factor". This factor gives a larger value for points that are closer to the sensor in space, and decays faster for points that are farther away. Finally, the "time attenuation factor" is used as the numerator, and the sum of the "time attenuation factor" and the "space attenuation factor" is used as the denominator. The ratio obtained by dividing the two is the weight C(v).

[0025] The temporal freshness and spatial proximity are fused in a nonlinear ratio through exponential decay, ensuring that the latest observation information is taken into account while avoiding the high-frequency noise of distant observations, thereby generating a comprehensive spatiotemporal confidence weight. Finally, the confidence weight C(v) is combined with the previously calculated P occ 、L T and L D A weighted combination is performed in each voxel to obtain the final 3D environmental model. This model not only contains accurate geometric structure, but also reflects the device temperature risk and type information in each grid. For example, in the same grid point, when C(v)>0.8 and L T When marked as a "high temperature risk area", the grid point will be colored red and provide warning information to the subsequent path planning module, thereby ensuring that the path planning automatically avoids high-risk areas while taking into account the system's ability to respond quickly to real-time environmental changes.

[0026] Step S2: Perform dynamic path planning in the three-dimensional environment model based on the real-time collected position information of each target robot and a preset target inspection point list to generate an anti-interference elastic path.

[0027] Based on the real-time collected information about the current position of each target robot and the preset locations of inspection points, a node cluster is first generated in the three-dimensional environment model through a three-dimensional isovoxel gridding method. The center of each voxel is a potential path node, and its spatial spacing is adapted to the model resolution to ensure both navigation accuracy and computational complexity. Subsequently, the cost of the connected paths between adjacent nodes is evaluated based on the improved A* pathfinding framework. The cost value is determined by the Euclidean distance, obstacle risk, and interference intensity. Among them, the cost evaluation part must first measure the geometric length of the path from one node to the next node. This length reflects the basic energy consumption and time cost of the robot's movement. At the same time, it is necessary to consider the obstacle risk of the path location, that is, the higher the probability of a certain area being marked as an obstacle in the environmental model constructed by the point cloud, the greater the possibility of potential collision or restriction, and the stronger the penalty for the path cost. Therefore, based on the Euclidean distance, it is weighted according to the nonlinear amplification method of the risk probability to ensure that the high-risk area will be significantly increased in cost even if the length is very short; in addition, it is also necessary to combine the real-time environmental interference intensity estimation, such as wind speed fluctuations, sensor noise or electromagnetic interference. When the interference level rises, the cost is additionally increased in the form of quadratic growth, so that the path planning algorithm gives priority to avoiding areas with severe interference. Cost evaluation can dynamically reflect that the shorter the distance, the lower the risk, and the less interference the better the path. When applying the improved A* algorithm, the value G in the formula is replaced ij As the cumulative cost value from starting point i to node j, combined with the heuristic function based on the node and the final inspection target (for example, the weighted sum of the three-dimensional compensated Euclidean distance and the residual risk estimate), the most likely path branch is prioritized to quickly approach the global optimal framework.

[0028] After the preliminary path framework is generated, according to each framework node P k The corresponding next inspection target coordinate t k Calculate the target attraction and the grid center O of the nearby high probability obstacle mCalculating obstacle repulsion. "Target attraction" ensures the robot always moves toward the next inspection point. This calculation is based on the vector difference between the current position and the next target point. This vector is normalized and multiplied by a fixed attraction coefficient, which specifies both the numerical value and the direction in which the robot should move. The attraction coefficient determines the force of the "pulling" force toward the target point: a larger coefficient forces the robot to approach the target in a straight line; a smaller coefficient allows the robot more leeway to avoid obstacles or path adjustments. The "obstacle repulsion" calculation focuses on repelling surrounding high-risk areas. First, one or more grid centers that are close to the current position and assigned a high obstacle probability in the environment model are identified. For each such grid, the vector difference between the current position and the grid center is calculated, scaled by the cube of the distance, and multiplied by a fixed repulsion coefficient. The cubic distance decay design ensures that as the robot approaches a danger zone, the repulsion strength increases rapidly, preventing the robot from entering too close a danger radius. However, as the distance increases slightly, the repulsion rapidly decreases, preventing excessive behavioral deviations in distant risk areas. All these individual repulsive force vectors are accumulated to form a total repulsive force vector, which, together with the target attractive force vector, participates in the final path fine-tuning. Next, the two force vectors are superimposed to obtain a resultant force, and a smooth interpolation is performed based on the resultant force direction and the preliminary node distribution. The node coordinates are adjusted to ensure that the path is more aligned with the inspection target direction while avoiding high-risk areas. For example, when a certain section of the frame guides the robot to approach the housing of a high-temperature transformer, the repulsive force component automatically deflects the trajectory to a safe distance, while the attractive force ensures that the overall path does not deviate from the task point. Ultimately, the interference-resistant elastic path generated through potential field fine-tuning retains the efficient short-range advantage of A* in shape, while leveraging the dynamic adjustment of the artificial potential field to achieve adaptive avoidance of sudden obstacles and environmental changes. The path is then output to the downstream control and task execution modules in the form of a node sequence.

[0029] Step S3: Dynamically optimize the motion parameters according to the collected real-time dynamic parameters of each target robot and the anti-interference elastic path to generate an anti-interference control instruction.

[0030] Real-time control begins with a complete understanding of the robot's motion state, including obtaining the current linear velocity and acceleration from the Inertial Measurement Unit (IMU), extracting environmental deformation information from the LiDAR point cloud, and the three-dimensional spatial coordinates returned by the positioning system. These dynamic parameters can not only determine the robot's motion trend along the elastic path, but also detect deviations from the motion trajectory caused by external interference. After perception is completed, the "target path node" is determined by comparing the Euclidean distance between the current position and the nearest path node. The path curvature and current velocity are used to estimate the expected acceleration, providing a benchmark for subsequent interference force assessment.

[0031] The difference between the expected acceleration and the measured acceleration reflects the force of the motion deviation, while the terrain undulations or obstacle movement revealed by the point cloud increment lead to additional interference. The two are combined to obtain the total interference force - that is, the sum of mass multiplied by the acceleration difference plus the terrain friction compensation force - and the path deviation value binary pair. The core goal of this stage is to quantify the external interference and explicitly feed it back to the controller so that it can make immediate adjustments to the "unexpected" impact. For example, if a section of the line causes the robot to have a 0.3m / s 2 If the acceleration changes suddenly, the change will be mapped as a high interference force, causing the path deviation value to increase rapidly, thereby triggering a more aggressive correction action.

[0032] On this basis, the three major control coefficients of proportional-integral-differential need to be adaptively optimized to meet the dual requirements of "fast response" and "smooth convergence". The adaptive algorithm based on the "control urgency index" U in this embodiment of the application takes into account the changing environment. The control urgency index can be expressed by the following formula: The control urgency index is used to measure the current path deviation value ΔS and the total interference force F dist The relative urgency in the overall control objective. Parameter γ is the sensitivity adjustment constant, F ref is the reference disturbance threshold. If the deviation or disturbance increases significantly, U approaches 1, indicating that a significant increase in control gain is required. Conversely, when U approaches 0, moderate deviation correction is required.

[0033] Then, the proportionality coefficient K p according to Perform a linear mapping, where K p0 With K p max They are the initial and strongest correction gains respectively. Integral coefficient K i Then use To suppress the excessive oscillation caused by long-term deviation accumulation, λ is the integral decay rate. Differential coefficient K d Similarly, the deviation change rate d(ΔS) / dt can be dynamically increased or decreased slightly to ensure controlled overshoot. This adaptive strategy allows the robot to quickly correct its trajectory under large deviations or strong disturbances while avoiding unnecessary oscillations during stable tracking.

[0034] Finally, using the updated three sets of coefficients to collect speed and deviation signals, the controller calculates drive commands—the speed adjustments for the left and right wheel motors and the yaw angle for the vehicle's servo—using the conventional formula for speed difference and steering angle. The controller continuously outputs a series of anti-interference control commands. When the next path node is detected or the deviation is reduced to a critical range (e.g., less than 0.1m), the process is repeated, maintaining high-precision tracking in complex and changing environments. This continuous iteration ensures that the robot can quickly return to its intended path despite wind, terrain disturbances, or external forces, enabling stable and efficient autonomous inspections.

[0035] Step S4: Calculate the detection distance between each target robot and the target to be detected on the anti-interference elastic path based on the position information collected in real time.

[0036] During the adaptive inspection control process of the power inspection robot, ensuring that each target robot can accurately identify the spatial proximity between itself and the required inspection object is the key to ensuring the accuracy of the inspection task triggering. Therefore, it is necessary to build a real-time distance calculation and judgment mechanism based on high-precision position perception and path target distribution information. First, the three-dimensional position information of all current target robots is obtained through the high-frequency positioning module. This position information can be obtained by fusing multi-source data such as inertial navigation system (INS), differential GPS and visual odometry (VO), and finally expressed in the form of a vector in the global coordinate system, that is, the position of each robot is P r =[x r ,y r ,z r ] T .

[0037] In order to determine which detection target is associated with, it is necessary to combine the task planning data on the anti-interference elastic path and extract the spatial position of the preset detection point near the current position of the robot. The anti-interference elastic path is a dynamically generated set of spatial paths on which several detection targets are distributed. The three-dimensional coordinates P of each target are d =[x d ,y d ,z d ] T Already marked in the path planning phase above. Based on the target robot's current position and the path progress index, it can quickly locate the nearest detection target it is heading towards. For example, if the robot has passed the previous node and the next detection point is 5 meters ahead on the path, that point is the detection target for this phase.

[0038] After obtaining the target coordinates, the three-dimensional space distance calculation is required. In order to eliminate the short-term nonlinear drift error caused by the track disturbance, this method proposes a three-dimensional Euclidean distance measurement function based on robust weighting. Assume that the current position of the robot is Pr , the detection target position is P d , then the Euclidean distance can be obtained as the detection distance D through the three-dimensional space distance calculation formula. In the embodiment of the present application, when performing the Z-axis difference calculation, the weighted coefficient w in the height direction is introduced z It is used to reflect the significant impact of vertical deviation on detection accuracy, especially in scenes with significant height differences such as power towers. Even small vertical deviations may lead to detection failure. z It can be set to be much larger than 1 to enhance the influence of vertical deviation on the total distance; on the contrary, in ground inspection or low-altitude short-range environment, w z Can be set to 1.

[0039] After completing the detection distance calculation, the system compares the distance with a preset distance threshold D thresh Compare. If D<D thresh , the robot is considered to have entered the detection area, and the next stage of visual acquisition and anomaly identification can be initiated. This threshold is determined based on the target device's perception requirements, the sensor's field of view, and the robot's movement accuracy. For example, in infrared detection, the infrared imaging module has an effective recognition range of 3 to 5 meters, so setting a threshold of 4 meters can ensure detection reliability.

[0040] Through this mechanism, each robot dynamically assesses the changing distance trend between itself and the device as it approaches its target, implementing a position-driven detection triggering strategy. This, combined with flexible paths, allows the robot to adjust its current position to the target, thus avoiding duplicate detection, missed detection, or interference with detection. This process strengthens the system's autonomous inspection and local judgment capabilities, and is a key component in achieving an efficient and robust intelligent power inspection system.

[0041] Step S5: When the detection distance is less than a preset distance threshold, abnormality detection and classification are performed on the target to be detected to obtain a fault level label corresponding to the target to be detected.

[0042] When D < D thresh Finally, high-resolution image acquisition of the target device is required to obtain the temperature matrix output by the infrared thermal imager and the feature image provided by the visible light camera. The infrared thermal image temperature matrix records the absolute temperature value of the target surface at each pixel, while the visible light image contains the device surface texture information. Synchronous acquisition ensures accurate spatial correspondence in subsequent analysis. For example, when the robot is less than 4 meters from the transformer winding, the infrared module can complete a 320×240 pixel temperature matrix frame in 0.1 seconds, while the visible light sensor captures RGB images at the same rate, ensuring that details of cracks and stains are not lost due to motion blur.

[0043] After the acquisition is completed, the two types of images are input into a deep convolutional neural network to achieve accurate identification and extraction of fault features. The network consists of a feature extraction layer, a residual block, and a multi-scale fusion module. It can automatically locate semantic elements such as temperature mutation areas, crack edges, and stain contours. For example, by performing pixel-level segmentation on infrared images, pixels with a mean temperature of 95°C can be marked as "high temperature risk areas." By applying edge detection and morphological closing operations to visible light images, the crack length L can be calculated. crack The extracted feature set includes the average temperature T avg , maximum temperature difference ΔT and crack length and other multi-dimensional indicators to form the fault feature set F=[T avg ,ΔT,L crack ].

[0044] To measure the degree of anomaly of multidimensional features, the embodiment of the present application adopts the following Anomaly Confidence Index (AIC) calculation formula based on weighted geometric mean: in, 、 as well as are the reference safety thresholds. W1+W2+W3=1 is the normalized weight. By comparing the ratio of a feature to its safety threshold, then using a logarithmic operation to eliminate dimensional differences, and finally restoring the scale using an exponential operation, we can highlight the abnormal weight when a single indicator far exceeds the safety threshold, while also taking into account the combined impact of multiple indicators to generate a scalar confidence score that can be directly input into the classification model.

[0045] Subsequently, the fault feature set F and the corresponding ACI are fed into a back-propagation neural network optimized with an improved operator (i.e., the IAOA-BP anomaly detection model). The improved operator optimizes the weight and threshold search process by introducing a dynamic migration strategy during initial population generation, fitness evaluation, and iterative updates, enabling the BP network to converge to the global optimum more quickly. The network outputs a three-dimensional vector [p0, p1, p2] corresponding to the probability distribution of "normal," "normal fault," and "serious fault," with the entry with the highest probability being the final fault level label. For example, when the model outputs [0.05, 0.15, 0.80], the device is judged to have a serious fault, triggering the prioritized maintenance process.

[0046] Step S6: performing multi-machine collaborative task allocation on the fault level labels according to the global scheduling information of all target robots to obtain an optimal inspection queue.

[0047] During the multi-machine collaborative task allocation phase, the current task execution status, list of remaining tasks to be inspected, spatial location distribution, and resource load (such as remaining power, available tool types, master control computing resources, etc.) of all robots in the inspection system must first be summarized in real time to form global scheduling information. This information is equivalent to a digital portrait of each robot's "battle situation" in a spherical coordinate system, enabling the scheduling model to perceive each robot's available service capabilities and potential proximity. For example, when a robot's battery level is less than 20%, it has been assigned two high-priority faults, and is far away from the fourth fault point, its priority in the scheduling decision will drop significantly. A robot that has just completed a low-level fault and is located in the middle of the inspection area will become the first choice for the next task allocation.

[0048] Next, we need to combine the fault level labels of each target to be inspected, the spatial distribution density of each fault point in the 3D environment model, and the aforementioned global scheduling information to jointly build a scheduling optimization model that minimizes the overall system inspection time and energy consumption while ensuring rapid response to high-priority faults. The core of this model is the objective function: Where M represents the total number of robots. i is the set of fault points assigned to the i-th robot. ij represents the estimated travel time required for robot i to move from its current position to fault point j. j ) is the fault distribution density within the unit voxel around the location of the fault point j in the three-dimensional environment model (reflecting the degree of fault concentration in the area), and μ is the density penalty coefficient to control the parallel scheduling conflict in high-density areas. j is the fault level of point j. Function w(L j ) maps fault levels to priority weights (e.g., normal, general, and severe are mapped to 1, 5, and 10, respectively). The objective function ensures rapid response to high-priority faults while balancing the spatial and energy allocation of multiple robots through a combination of density penalties and travel time.

[0049] After the optimization model is established, the model is dynamically solved using a pre-set branch-and-bound algorithm. This algorithm constructs a branching tree, treating each possible task assignment as a node in the tree. When the node is expanded, it prunes the subtree based on the lower bound of the objective function corresponding to the current partial assignment solution. When the lower bound is higher than the known optimal solution, the subtree is no longer searched. This "bound" is estimated from the minimum possible value of the objective function under partial task assignment. During the scheduling process, the algorithm can respond in real time to changes in robot status or fault levels. For example, if a robot cannot continue its task due to low battery or mechanical failure, the algorithm immediately re-evaluates the corresponding subtree and, after pruning it, reallocates the remaining unfinished tasks of the robot to other available robots.

[0050] The entire solution process is output as an optimal inspection queue, which provides the sequence of fault points that each robot must visit in sequence and their estimated arrival times. For example, using three robots and five fault points, if the optimal assignment obtained through branch and bound is: Robot A → {Severe Fault 1 → General Fault 4}, Robot B → {Severe Fault 2 → Normal Fault 5}, Robot C → {General Fault 3}, this queue not only ensures the fastest response to high-priority faults but also avoids local overcrowding in spatial distribution while minimizing energy consumption. This optimal inspection queue can be distributed by the dispatch center to each robot's execution unit, and in conjunction with the navigation and control modules, achieves efficient and robust fully autonomous collaborative inspections.

[0051] This application is applied to the field of patrol control technology. It obtains a three-dimensional environmental model by performing spatiotemporal alignment and fusion modeling on multimodal environmental data. It performs dynamic path planning on the three-dimensional environmental model based on the position information of each target robot and the target patrol point list to generate an anti-interference elastic path. It dynamically optimizes the motion parameters in combination with the real-time dynamic parameters of each target robot to generate anti-interference control instructions. It calculates the detection distance between each target robot and the target to be detected based on the position information. When the detection distance is less than the distance threshold, it performs abnormal detection and classification on the target to be detected to obtain the corresponding fault level label. It performs multi-machine collaborative task allocation on the fault level label based on the global scheduling information to obtain the optimal patrol queue. This application comprehensively improves the safety, reliability and patrol efficiency of power patrol robots through precise perception, real-time planning, online optimization and multi-machine collaboration, and provides strong technical support for the intelligent maintenance and fault prevention of power equipment.

[0052] like Figure 2 , which is a functional module diagram of an adaptive inspection control device for an electric power inspection robot provided in an embodiment of the present application.

[0053] In some embodiments, the adaptive inspection control device 2 of the power inspection robot may include a plurality of functional modules composed of computer program segments. The computer program of each program segment in the adaptive inspection control device 2 of the power inspection robot may be stored in a memory of a server and executed by at least one processor to execute (see Figure 1 (Describe) the functions of the adaptive inspection control method of the power inspection robot.

[0054] In this embodiment, the adaptive inspection control device 2 of the power inspection robot can be divided into multiple functional modules based on the functions they perform. These functional modules may include an environmental perception module 21, a path planning module 22, a motion correction module 23, a distance monitoring module 24, a fault detection module 25, and a multi-machine coordination module 26. As used herein, a module refers to a series of computer program segments that can be executed by at least one processor and perform a fixed function, and are stored in a memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0055] The environment perception module 21 is used to perform spatiotemporal alignment and fusion modeling on the multimodal environment data in a preset area to obtain a three-dimensional environment model.

[0056] In an optional embodiment, the environment perception module 21 is specifically configured to: Performing time domain alignment on the multimodal environmental data within a preset area according to the timestamps of each data to obtain synchronized environmental data; Performing power equipment identification and preliminary anomaly detection on the synchronized environmental data through a preset CNN network to obtain obstacle labels and obstacle probabilities; spatially fusing the synchronized environmental data according to an extrinsic calibration matrix and a coordinate system of the multimodal environmental data to obtain a three-dimensional grid map; The obstacle labels and the obstacle probabilities are marked on the three-dimensional grid map to obtain the three-dimensional environment model.

[0057] The path planning module 22 is used to perform dynamic path planning in the three-dimensional environment model based on the real-time collected position information of each target robot and a preset target inspection point list to generate an anti-interference elastic path.

[0058] In an optional embodiment, the path planning module 22 is specifically configured to: Marking a path node set in a three-dimensional environment model based on the real-time collected position information of each target robot and the position information of the inspection point; Using a preset improved A* algorithm, paths between adjacent path nodes are evaluated according to a preset total interference force and the obstacle probability to obtain each path and a corresponding cost value; Selecting paths between adjacent path nodes according to the cost value to obtain a global inspection path framework; Calculating target attraction and obstacle repulsion based on the coordinate points in the global inspection path framework, the location information of the points to be inspected, and the centers of the grid cells with high obstacle probability along the way; The global inspection path framework is smoothly corrected according to the target attraction and the obstacle repulsion by a preset artificial potential field model to generate the anti-interference elastic path.

[0059] The motion correction module 23 is used to dynamically optimize the motion parameters according to the collected real-time dynamic parameters of each target robot and the anti-interference elastic path to generate an anti-interference control instruction.

[0060] In an optional embodiment, the motion correction module 23 is specifically configured to: Real-time collection of each target robot's real-time speed, current acceleration, current environment 3D point cloud data and current position information; Determining an expected acceleration and a target path node based on the current position information and the anti-interference elastic path, and performing interference force estimation and path deviation calculation based on the expected acceleration, the target path node, the current acceleration, and the current environmental three-dimensional point cloud data to update the total interference force and obtain a path deviation value; Performing real-time adjustment on PID control parameters according to the total disturbance force and the path deviation value to update the proportional coefficient, the integral coefficient, and the differential coefficient; The motor speed and steering angle are calculated according to the proportional coefficient, the integral coefficient, the differential coefficient, the real-time speed, and the path deviation value to generate continuous anti-interference control instructions.

[0061] The distance monitoring module 24 is used to calculate the detection distance between each target robot and the target to be detected on the anti-interference elastic path based on the position information collected in real time.

[0062] In an optional embodiment, the distance monitoring module 24 is specifically configured to: Collecting the current position information of each target robot in real time, and locating the coordinate information of the corresponding target to be detected on the anti-interference elastic path according to the position information; Calculating the detection distance between the current position of each target robot and the target to be detected based on the position information and the coordinate information using a preset three-dimensional space distance measurement method; The detection distance is compared with the distance threshold to determine whether the target robot enters a preset device detection area.

[0063] The fault detection module 25 is configured to perform abnormality detection and classification on the target to be detected when the detection distance is less than a preset distance threshold, so as to obtain a fault level label corresponding to the target to be detected.

[0064] In an optional embodiment, the fault detection module 25 is specifically configured to: Performing device detection image acquisition on the target to be detected to obtain an infrared thermal image temperature matrix and a visible light characteristic image; Performing fault feature recognition and extraction on the infrared thermal image temperature matrix and the visible light feature image through a preset CNN model to obtain a fault feature set; The fault feature set is classified into fault levels using a preset IAOA-BP anomaly detection model to obtain a fault level label corresponding to the target to be detected.

[0065] The multi-machine coordination module 26 is used to perform multi-machine collaborative task allocation on the fault level labels according to the global scheduling information of all target robots to obtain the optimal inspection queue.

[0066] In an optional embodiment, the multi-machine coordination module 26 is specifically configured to: Summarize the current task status, remaining task list, location distribution and resource load of all target robots in the preset area to obtain the global scheduling information; Constructing a scheduling optimization model according to the fault level label, the target distribution density in the three-dimensional environment model, and the global scheduling information; The scheduling optimization model is dynamically solved according to a preset branch and bound algorithm to obtain the optimal inspection queue.

[0067] It should be understood that the various variations and specific embodiments of the methods provided in the above embodiments are also applicable to the adaptive patrol control device of the electric power inspection robot in this embodiment. Through the above detailed description of the adaptive patrol control method of the electric power inspection robot, those skilled in the art can clearly know the implementation method of the adaptive patrol control device of the electric power inspection robot in this embodiment. For the sake of brevity of the specification, it will not be described in detail here.

[0068] like Figure 3 , which is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0069] In a preferred embodiment of the present invention, the electronic device 3 may include, but is not limited to: a memory 31 , at least one processor 32 and at least one communication bus 33 .

[0070] Those skilled in the art should understand that Figure 3 The structure of the electronic device 3 shown does not constitute a limitation of the embodiment of the present invention. The electronic device 3 may also include more or less other hardware or software than shown in the figure, or a different component arrangement.

[0071] In some embodiments, the electronic device 3 is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application-specific integrated circuits, programmable gate arrays, digital processors and embedded devices.

[0072] It should be noted that the electronic device 3 is only an example. Other existing or future electronic products that are suitable for this application should also be included in the scope of protection of this application and included here by reference.

[0073] In some embodiments, the memory 31 stores a computer program that, when executed by the at least one processor 32, implements all or part of the steps in the adaptive inspection control method for a power inspection robot. The memory 31 includes a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM), or other optical disk storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data. Furthermore, the computer-readable storage medium may primarily include a program storage area and a data storage area, wherein the program storage area may store an operating system, at least one application required for a function, and the like.

[0074] In some embodiments, the at least one processor 32 is the control core (Control Unit) of the electronic device 3. It connects the various components of the electronic device 3 using various interfaces and circuits. It executes programs or modules stored in the memory 31 and accesses data stored in the memory 31 to perform various functions and process data in the electronic device 3. For example, when executing the computer program stored in the memory 31, the at least one processor 32 implements all or part of the steps of the adaptive inspection control method for the power inspection robot described in the embodiments of the present application; or implements all or part of the functions of the adaptive inspection control device for the power inspection robot. The at least one processor 32 can be composed of an integrated circuit, such as a single packaged integrated circuit or multiple packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and a combination of various control chips.

[0075] In some embodiments, the at least one communication bus 33 is configured to enable communication between the memory 31 and the at least one processor 32. Although not shown, the electronic device 3 may also include a power supply (e.g., a battery) to power various components. Preferably, the power supply can be logically connected to the at least one processor 32 via a power management device, thereby enabling the power management device to manage charging, discharging, and power consumption. The power supply may also include one or more DC or AC power supplies, a recharging device, a power failure detection circuit, a power converter or inverter, a power status indicator, and other components. The electronic device 3 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be detailed here.

[0076] The above-mentioned integrated unit implemented in the form of a software functional module can be stored in a computer-readable storage medium. The above-mentioned software functional module stored in a storage medium includes a number of instructions for causing an electronic device (which can be a personal computer, electronic device, or network device, etc.) or a processor to execute portions of the methods described in various embodiments of the present application.

[0077] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the module division is only a logical function division, and other division methods may be used in actual implementation.

[0078] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, and may be located in one place or distributed across multiple network elements. Some or all of the modules may be selected to achieve the purpose of this embodiment based on actual needs.

[0079] The above are all preferred embodiments of the present application, and are not intended to limit the scope of protection of the present application. Therefore, any equivalent changes made based on the structure, shape, and principle of the present application should be included in the scope of protection of the present application.

Claims

1. An adaptive inspection control method for an electric power inspection robot, characterized in that: The method comprises: Perform spatiotemporal alignment and fusion modeling on the multimodal environmental data in the preset area to obtain a three-dimensional environmental model; Based on the real-time collected position information of each target robot and a preset target inspection point list, dynamic path planning is performed in the three-dimensional environment model to generate an anti-interference elastic path; Dynamically optimize motion parameters based on the collected real-time dynamic parameters of each target robot and the anti-interference elastic path to generate anti-interference control instructions; Calculating the detection distance between each target robot and the target to be detected on the anti-interference elastic path based on the position information collected in real time; When the detection distance is less than a preset distance threshold, performing abnormality detection and classification on the target to be detected to obtain a fault level label corresponding to the target to be detected; The fault level labels are assigned to multi-machine collaborative tasks according to the global scheduling information of all target robots to obtain the optimal inspection queue.

2. The adaptive inspection control method of the power inspection robot according to claim 1, characterized in that: The performing spatiotemporal alignment and fusion modeling of the multimodal environmental data in the preset area to obtain a three-dimensional environmental model includes: Performing time domain alignment on the multimodal environmental data within a preset area according to the timestamps of each data to obtain synchronized environmental data; Performing power equipment identification and preliminary anomaly detection on the synchronized environmental data through a preset CNN network to obtain obstacle labels and obstacle probabilities; spatially fusing the synchronized environmental data according to an extrinsic calibration matrix and a coordinate system of the multimodal environmental data to obtain a three-dimensional grid map; The obstacle labels and the obstacle probabilities are marked on the three-dimensional grid map to obtain the three-dimensional environment model.

3. The adaptive inspection control method of the power inspection robot according to claim 2, characterized in that: The target inspection point list includes position information of the points to be inspected, and performing dynamic path planning in the three-dimensional environment model based on the real-time collected position information of each target robot and the preset target inspection point list to generate an anti-interference elastic path includes: Marking a path node set in a three-dimensional environment model based on the real-time collected position information of each target robot and the position information of the inspection point; Using a preset improved A* algorithm, paths between adjacent path nodes are evaluated according to a preset total interference force and the obstacle probability to obtain each path and a corresponding cost value; Selecting paths between adjacent path nodes according to the cost value to obtain a global inspection path framework; Calculating target attraction and obstacle repulsion based on the coordinate points in the global inspection path framework, the location information of the points to be inspected, and the centers of the grid cells with high obstacle probability along the way; The global inspection path framework is smoothly corrected according to the target attraction and the obstacle repulsion by a preset artificial potential field model to generate the anti-interference elastic path.

4. The adaptive inspection control method of the power inspection robot according to claim 3, characterized in that: The step of dynamically optimizing motion parameters based on the collected real-time dynamic parameters of each target robot and the anti-interference elastic path to generate anti-interference control instructions includes: Real-time collection of each target robot's real-time speed, current acceleration, current environment 3D point cloud data and current position information; Determining an expected acceleration and a target path node based on the current position information and the anti-interference elastic path, and performing interference force estimation and path deviation calculation based on the expected acceleration, the target path node, the current acceleration, and the current environmental three-dimensional point cloud data to update the total interference force and obtain a path deviation value; Performing real-time adjustment on PID control parameters according to the total disturbance force and the path deviation value to update the proportional coefficient, the integral coefficient, and the differential coefficient; The motor speed and steering angle are calculated according to the proportional coefficient, the integral coefficient, the differential coefficient, the real-time speed, and the path deviation value to generate continuous anti-interference control instructions.

5. The adaptive inspection control method of the power inspection robot according to claim 1, characterized in that: Calculating the detection distance between each target robot and the target to be detected on the anti-interference elastic path based on the real-time collected position information includes: Collecting the current position information of each target robot in real time, and locating the coordinate information of the corresponding target to be detected on the anti-interference elastic path according to the position information; Calculating the detection distance between the current position of each target robot and the target to be detected based on the position information and the coordinate information using a preset three-dimensional space distance measurement method; The detection distance is compared with the distance threshold to determine whether the target robot enters a preset device detection area.

6. The adaptive inspection control method of the power inspection robot according to claim 1, characterized in that: The performing abnormality detection and classification on the target to be detected to obtain a fault level label corresponding to the target to be detected includes: Performing device detection image acquisition on the target to be detected to obtain an infrared thermal image temperature matrix and a visible light characteristic image; Performing fault feature recognition and extraction on the infrared thermal image temperature matrix and the visible light feature image through a preset CNN model to obtain a fault feature set; The fault feature set is classified into fault levels using a preset IAOA-BP anomaly detection model to obtain a fault level label corresponding to the target to be detected.

7. The adaptive inspection control method of the power inspection robot according to claim 1, characterized in that: The multi-machine collaborative task allocation for the fault level labels according to the global scheduling information of all target robots to obtain the optimal inspection queue includes: Summarize the current task status, remaining task list, location distribution and resource load of all target robots in the preset area to obtain the global scheduling information; Constructing a scheduling optimization model according to the fault level label, the target distribution density in the three-dimensional environment model, and the global scheduling information; The scheduling optimization model is dynamically solved according to a preset branch and bound algorithm to obtain the optimal inspection queue.

8. An adaptive inspection control device for an electric power inspection robot, characterized in that: The device comprises: The environmental perception module is used to perform spatiotemporal alignment and fusion modeling of multimodal environmental data in a preset area to obtain a three-dimensional environmental model; A path planning module is used to perform dynamic path planning in the three-dimensional environment model based on the real-time collected position information of each target robot and a preset target inspection point list to generate an anti-interference elastic path; A motion correction module is used to dynamically optimize the motion parameters based on the collected real-time dynamic parameters of each target robot and the anti-interference elastic path to generate anti-interference control instructions; A distance monitoring module is used to calculate the detection distance between each target robot and the target to be detected on the anti-interference elastic path based on the position information collected in real time; a fault detection module, configured to perform abnormality detection and classification on the target to be detected when the detection distance is less than a preset distance threshold, so as to obtain a fault level label corresponding to the target to be detected; The multi-machine coordination module is used to perform multi-machine collaborative task allocation on the fault level labels according to the global scheduling information of all target robots to obtain the optimal inspection queue.

9. An electronic device, characterized in that: The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the adaptive inspection control method of the power inspection robot according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the adaptive inspection control method of the power inspection robot according to any one of claims 1 to 7 are implemented.

Images