High-voltage power transmission and distribution line all-terrain autonomous inspection robot path planning method

Abstract

The invention discloses a high-voltage power transmission and distribution line all-terrain autonomous inspection robot path planning method, and the method comprises the steps: constructing a comprehensive risk map which reflects the dynamic risk distribution in a distribution line through a mode of combining exponential decay superposition with Markov prediction, and providing accurate risk information for subsequent planning; then, a heuristic greedy-genetic algorithm hybrid strategy is adopted to carry out iterative optimization on the task sequence, an optimal inspection scheduling sequence is output, and finally, the optimal inspection scheduling sequence is continuously and dynamically corrected based on a multi-mode perception fusion model and a self-adaptive task feedback mechanism of Bayesian state estimation. The inspection efficiency is greatly improved, and the inspection precision and reliability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of inspection robots, and particularly to a method for path planning of an all-terrain autonomous inspection robot for high-voltage power transmission and distribution lines. Background Art

[0002] Distribution lines are one of the key electrical facilities in the urban power distribution system, undertaking important functions such as voltage transformation, power distribution, and protection. To ensure the stable operation of the power system, regular inspections need to be carried out on the equipment within the distribution lines. The conventional inspection items include temperature, voltage, current, equipment operation status indicators, instrument readings, etc.

[0003] The traditional inspection task of distribution lines is manual inspection. Due to the narrow space, complex channels, and dense arrangement of equipment in the distribution lines, the efficiency of manual inspection is low, and there are certain safety risks when operating under special working conditions such as night, high temperature, and thunderstorms. In addition, there are problems such as strong subjectivity, high omission rate, and difficulty in tracing data when manually recording equipment information, which are difficult to meet the requirements of refined, data-based, and automated management.

[0004] In view of this, intelligent inspection robots are gradually applied to the field of distribution line inspection. Currently, there are mainly the following two technical routes for intelligent inspection robots: (1) Fixed-path navigation based on laser SLAM or QR code positioning. Laser SLAM navigation is to scan through a lidar and construct a map of the internal environment of the distribution line for self-positioning and navigation. The robot usually conducts inspections along a preset fixed path, and sequentially reaches the target equipment points along the way for image acquisition or status detection. For QR code positioning, the robot locates and navigates by scanning the pre-laid QR code points, and conducts equipment detection and image acquisition according to the preset route. The above navigation methods complete various tasks in a fixed order and cannot be flexibly adjusted according to the fault priority of the equipment, the urgency of the inspection task, or environmental changes. This scheduling method results in low inspection efficiency and cannot respond to emergencies or conduct key inspections on certain equipment in a short time.

[0005] (2) Dynamic path navigation integrating visual SLAM and multi-sensor fusion. The robot obtains environmental data of the distribution line through multiple sensors such as lidar, camera, and IMU, and uses visual SLAM technology for map construction and self-positioning. In this process, visual SLAM captures environmental image information through a camera and estimates the pose through image feature matching; at the same time, the lidar provides distance data, and the IMU helps to eliminate the errors caused by attitude changes. However, this dynamic path navigation scheme requires a large amount of computational resources and has poor real-time performance. This technology relies on the data fusion of multiple sensors such as visual SLAM, lidar, and IMU, and strong computing power is required to process these data. Summary of the Invention

[0006] In view of the above problems, the present invention proposes a path planning method for an all-terrain autonomous inspection robot for high-voltage power transmission and distribution lines, mainly solving at least one of the problems raised in the background art.

[0007] To solve the above technical problems, the technical solution of the present invention is as follows: A path planning method for an all-terrain autonomous inspection robot for high-voltage power transmission and distribution lines, comprising: Based on a static prior map, an immediate risk assessment is performed on static obstacles inside the target distribution line, and a Markov chain is used to predict the risk of dynamic obstacles, which are respectively defined as immediate risk and predicted risk. The immediate risk and the predicted risk are superimposed by exponential decay to generate a comprehensive risk map; Generate a model map according to the target distribution line. Based on the model map, the A* search algorithm with risk perception is used as the main path for global path planning, and then a deep Q network is introduced to fine-tune and correct the local path in real time. Finally, a heuristic greedy-genetic algorithm hybrid strategy is used to iteratively optimize the task order, comprehensively considering task priorities and time window constraints, and output the optimal inspection scheduling order; The robot performs inspection tasks according to the constraints of the comprehensive risk map and the optimal inspection scheduling order. During the inspection tasks, an adaptive task feedback mechanism based on a multi-modal perception fusion model and Bayesian state estimation continuously and dynamically corrects the optimal inspection scheduling order.

[0008] In some embodiments, the construction of the static prior risk map includes: Use indoor SLAM to perform offline mapping of the interior of the target distribution line to obtain a grid map of the clear space of the distribution line; On the grid map, mark the walls, cabinet edges, and fixed equipment as high-risk areas to form the static prior risk map.

[0009] In some embodiments, the process of the immediate risk assessment includes: Call the lidar and camera to collect obstacle distance information in each sensing cycle, map the obstacle distance information to the grid points of the grid map for dynamic risk perception, obtain the current immediate risk value of the robot, and insert the immediate risk value into the corresponding position in the static prior risk map to form a time-varying risk map.

[0010] In some embodiments, the calculation method of the immediate risk value is: ; In the formula, d is the obstacle distance information, x and y are the current abscissa and ordinate of the robot respectively, and λ dp is the instantaneous decay rate corresponding to the obstacle distance information d. When d approaches 0, p approaches 1, and as the distance increases, the risk decays rapidly in an exponential form.

[0011] In some embodiments, the process of Markov chain prediction includes: For each grid point in the time-varying risk map, a first-order Markov chain is established, and a transition matrix is used to describe the risk transition probability and risk distribution vector between adjacent grids of the robot. The predicted risk value is calculated according to the risk transition probability and the risk distribution vector.

[0012] In some embodiments, the scheduling objective function of the heuristic greedy-genetic algorithm hybrid strategy is: ; In the formula, is the delay penalty weight coefficient of the k-th sensor cycle, is the delay of the k-th sensor cycle, d(v ti , v ti+1 ) is the path length between adjacent task points, θ is the path length weight, and m is the total number of tasks.

[0013] In some embodiments, the construction process of the multi-modal perception fusion model includes: Let the observation of the i-th sensor at time t be z i (t), then the total observation vector is: ; Construct the conditional probability of each state S k in the observation space: ; Among them, P(Z∣S k ) is obtained by fitting a Bayesian network through training data, k is the index number of the state, and j is the index number of traversal.

[0014] The beneficial effects of the present invention are as follows: By combining exponential decay superposition and Markov prediction, a comprehensive risk map reflecting the "dynamic risk" distribution inside the distribution line is constructed to provide accurate risk information for subsequent planning. Then, a heuristic greedy-genetic algorithm hybrid strategy is used to iteratively optimize the task order, and the optimal inspection scheduling order is output. Finally, based on the multi-modal perception fusion model and the adaptive task feedback mechanism of Bayesian state estimation, the optimal inspection scheduling order is continuously and dynamically corrected, greatly improving the inspection efficiency, inspection accuracy and reliability. Description of the Drawings

[0015] Figure 1Schematic flow chart of the path planning method for a high-voltage power transmission and distribution line all-terrain autonomous inspection robot disclosed in an embodiment of the present invention. Detailed implementation manners

[0016] To make the objectives, technical solutions and advantages of the present invention clearer and more definite, the content of the present invention will be further described in detail below with reference to the accompanying drawings and specific implementation manners. It can be understood that the specific embodiments described herein are only used to explain the present invention, rather than limiting the present invention. Additionally, it should be noted that for the sake of description, only parts related to the present invention are shown in the accompanying drawings, rather than all the content.

[0017] This embodiment proposes a path planning method for a high-voltage power transmission and distribution line all-terrain autonomous inspection robot, as Figure 1 shown, including: Step 1: Based on a static prior map, perform an immediate risk assessment on static obstacles inside the target distribution line, and use a Markov chain to predict the risk of dynamic obstacles, which are respectively defined as immediate risk and predicted risk. By exponentially decaying and superimposing the immediate risk and the predicted risk, a comprehensive risk map is generated.

[0018] Step 101: The construction of the static prior risk map includes: Use indoor SLAM to perform offline mapping on the inside of the target distribution line to obtain a grid map of the clear space area of the distribution line; On the grid map, mark the walls, cabinet edges, and fixed equipment as high-risk areas to form a static prior risk map, which is defined as: .

[0019] Step 102: The process of immediate risk assessment includes: Call the lidar and camera to collect obstacle distance information in each sensing cycle, map the obstacle distance information to the grid points of the grid map for dynamic risk perception, obtain the current immediate risk value of the robot, and insert the immediate risk value into the corresponding position in the static prior risk map to form a time-varying risk map. In this solution, the calculation method of the immediate risk value is: ; In the formula, d is the obstacle distance information, x and y are respectively the current abscissa and ordinate of the robot, and λ d is the immediate decay rate corresponding to the obstacle distance information d. When d approaches 0, p approaches 1, and as the distance increases, the risk decays rapidly in an exponential form. λ d can be adjusted according to the laser ranging accuracy and environmental density.

[0020] Step 103: The process of Markov chain prediction includes: For each grid point in the time-varying risk map, a first-order Markov chain is established, and a transition matrix is used to describe the risk transition probability and risk distribution vector of the robot between adjacent grids. The predicted risk value is calculated based on the risk transition probability and risk distribution vector.

[0021] Specifically: (1) Transition probability estimation The number of times observed from grid point j→i in the past T time steps is n ij , then .

[0022] (2) Multi-step prediction Use matrix power operation to predict the risk distribution vector p after τ steps t : .

[0023] (3) Unfolding of single-step prediction For a single grid i: .

[0024] Step 104, Exponential decay superposition update: Superpose the immediate risk value and the predicted risk value obtained in Steps 102 and 103 respectively in proportion to balance "current observation" and "historical trend": ; α is the forgetting factor, and when it is closer to 1, it depends more on the current observation; the cumulative prediction term reflects the short-term dynamic trend.

[0025] Step 105, Formation of the comprehensive risk map: The final comprehensive risk map M(x, y, t) takes the point-by-point maximum value of the static prior risk map and the dynamic risk after superposition update in Step 104: ; Thus, it is ensured that the prior high-risk area is always retained, and at the same time, new risks brought by dynamic obstacles are introduced; M(x, y, t) ∈ [0, 1] can be directly used as the risk penalty weight in the planning cost.

[0026] The total number of grids M of the static prior risk map is usually in the thousands to tens of thousands. Through the transition matrix and neighborhood risk reduction pruning, the efficiency can be greatly improved, ensuring that the full-map risk update is completed within every 100 ms.

[0027] In the above steps 101-105, by exponentially decaying and superimposing the real-time obstacle distance information of the lidar / camera and combining with the short-term prediction of the first-order Markov chain, a time-varying risk weight M(x, y, t) is generated, which can accurately reflect the "dynamic risk" distribution inside the distribution line. Compared with the scheme that only relies on the static map, the risk exposure rate is reduced by about 20%.

[0028] Step 2: Generate a model diagram for the target distribution line. Based on the model diagram, use the risk-aware A* search algorithm as the backbone path for global path planning. Then, introduce a deep Q-network to fine-tune and correct the local path in real time. Finally, adopt a hybrid strategy of heuristic greedy-genetic algorithm to iteratively optimize the task order, comprehensively consider the task priority and time window constraints, and output the optimal inspection scheduling order.

[0029] Step 201: Model diagram construction: Discretize the distribution line space into a topological graph G=(V, E), where: V={v1, v2, …, vn}: represents the center of each grid or the point of interest; E={(vi, vj)}: If there is no obstacle and it is passable between two points, an edge is built, and the edge weight is defined as: ; In the formula, d ij is the Euclidean distance, is the average risk based on the dynamic risk map in the previous section, and λ is the risk weight factor.

[0030] Step 202: Shortest path search: Improved A* search algorithm: Use the risk-aware A* search algorithm as the backbone for global path planning: Heuristic function definition: ; In the formula, β is the spatial heuristic term coefficient, γ is the risk heuristic term coefficient, and M(v) is the comprehensive risk value of the current point.

[0031] Compared with the traditional A* that only uses distance as a reference, in step 202, the environmental risk is introduced into the path cost estimation, which can effectively avoid high-risk obstacle areas.

[0032] Step 203: Local obstacle avoidance and path correction: Deep Q-network (DQN): During the process of the AGV executing the path, when encountering sudden obstacles (such as personnel movement, the door closing suddenly), the global path may become unavailable. Therefore, introduce DQN to fine-tune the local path in real time: (1) State space S: includes the current position of the AGV, the target direction, and the laser point cloud scan; (2) Action space A = {Move left, Move forward, Move right, Stop}; (3) Reward function design: ; (4) Through the Q - value update formula: .

[0033] The AGV learns the optimal action strategy in real - time, avoids obstacles and keeps the overall path direction stable.

[0034] The above steps 201 - 203 are based on the comprehensive risk map obtained in step 1, introducing the A* global search optimized by risk weights and the local fine - tuning of the deep Q - network, making the inspection path not only the shortest but also able to avoid high - risk areas. In actual tests, compared with the pure distance - shortest algorithm, the obstacle - avoidance success rate is increased by 25%.

[0035] Step 204, Multi - task scheduling and priority optimization: Let the inspection task set be T = {t1, t2, …, tm}, and each task contains the following parameters: location node v tk , priority π k , feasible time window [e k , l k .

[0036] Finally, use the heuristic greedy - genetic algorithm hybrid strategy to iteratively optimize the task order, comprehensively considering the priority and time - window constraints, and output the optimal inspection scheduling order.

[0037] Among them, the scheduling objective function of the heuristic greedy - genetic algorithm hybrid strategy is: ; In the formula, is the delay penalty weight coefficient of the k - th sensor cycle, is the delay amount of the k - th sensor cycle, where, ( represents the actual arrival time of task k, represents the expected completion time threshold of task k), d(v ti , v ti+1 ) is the path length between adjacent task points (the i - th task point and the (i + 1) - th task point, where i represents the position in the task execution sequence), θ is the path - length weight, and m is the total number of tasks.

[0038] In step 204 above, a heuristic greedy-genetic algorithm hybrid strategy is adopted to construct a scheduling objective function. Considering both the task priority π_k and the time window [e_k, l_k], the inspection point order is iteratively optimized. Experiments show that the total travel load is reduced by 15% on average, and the task response delay is reduced by 18%.

[0039] Step 3: The robot performs inspection tasks according to the constraints of the comprehensive risk map and the optimal inspection scheduling order. During the inspection tasks, an adaptive task feedback mechanism based on the multi-modal perception fusion model and Bayesian state estimation continuously and dynamically corrects the optimal inspection scheduling order.

[0040] Step 301: The construction process of the multi-modal perception fusion model includes:[[]] The AGV is equipped with multiple sensor modules (IMU, lidar, vision camera, environmental temperature and humidity sensor, etc.). Therefore, let the observation of the i-th sensor at time t be z i (t), then the total observation vector is:[[]] ; Construct the conditional probability of each state S k (such as: normal, deviation, jamming, completion and other task states) in the observation space:[[]] ; Among them, is obtained by fitting the Bayesian network through training data (or using the Gaussian mixture model (GMM)), represents the conditional probability of observing Z(t) at time t in state S k and is obtained by fitting through the Bayesian network with training samples, where is the network parameter vector, k is the index number of the state, and j is the traversed index number.

[0041] Step 302: State estimation and self-correction mechanism (improved Kalman filter): State transition uses linear prediction: ; Observation update: ; Among them, is the Kalman gain, x^t is the state estimate value (position / deviation / whether there is an error), zt is the current sensor observation, and R is the measurement noise covariance.

[0042] Through this method, continuous dynamic prediction and error correction of the AGV state can be performed to achieve self-adjustment during task interruption or incorrect inspection.

[0043] Step 303: Task completion degree and information entropy determination mechanism: Introduce the information entropy H to judge the "credibility" of the current task completion degree. The lower the entropy, the clearer the state. The formula is as follows: ; If the confidence level of a certain state (such as "task completed") and the entropy value is less than the set threshold , it is considered that the task is completed: ; Otherwise, feedback the abnormality or continue to monitor.

[0044] Step 304, abnormal feedback and scheduling linkage: When the system determines that the state is abnormal (such as deviation from the path, task failure, sensing failure, etc.), immediately perform the following operations: Report the state flag Sk and the current observation Z(t) to the main control scheduling system; The scheduling system re-plans the task path according to the task priority table and the current available path; If there are continuous abnormalities within a short period of time, the system will mark this area as a "high-risk area", and subsequent tasks will avoid it or schedule manual re-inspection.

[0045] The above steps 301 - 304 are based on the multi-modal observation vector Z(t) and Bayesian state estimation, combined with the information entropy H(Z) determination mechanism, which can trigger re-planning in real time when abnormalities such as deviation from the path and sensing failure occur. The measured task completion rate is increased by 10% compared with the no-feedback scheme, and the system robustness is significantly enhanced.

[0046] Step 305, abnormal log recording and visualization report generation: In order to achieve full-cycle recording and traceability of the system operation state, the system of this solution designs an abnormal event log recording mechanism and a periodic inspection report automatic generation module.

[0047] The structure of each abnormal event record is as follows: ; Among them, t i is the timestamp when the abnormality occurs; Type i is the type of abnormality (such as path deviation, sensing failure, jamming, etc.); Z i is the observation vector when the abnormality occurs; (x i , y i ) is the grid coordinate of the abnormality occurrence location; Risk i is the comprehensive risk value corresponding to the abnormality.

[0048] Step 306, high-risk area management and dynamic threshold adjustment: When an anomaly continuously occurs at a certain location, the system marks it as a "high-risk area". However, to avoid misjudgment and information redundancy, the following mechanism is introduced for automatic clearing and dynamic risk adjustment.

[0049] (1) High-risk marking revocation mechanism If a high-risk grid (x, y) does not have any more anomalies in consecutive Mclear cycles, the risk marking at this point is automatically revoked: ; where M0(x, y) is the initial static risk.

[0050] (2) Dynamic threshold adjustment for risk determination In scenarios with drastic environmental changes or frequent false alarms during inspections, the risk confidence threshold δ and the information entropy threshold ε can be dynamically adjusted as follows: ; where σ t is the intensity of anomaly fluctuations in the recent period; η1, η2 are adjustment factors; δ0, ε0 are the initial confidence and entropy thresholds.

[0051] Step 307, Manual re-inspection task scheduling and queue-jumping mechanism: (1) Rules for generating manual re-inspection tasks If a grid (x, y) has more than N abn anomalies within the time window [t, t + T], a manual re-inspection task is triggered: ; (2) Feedback mechanism for re-inspection results The operation and maintenance personnel can enter the re-inspection results through the platform: Normal: The risk marking is cleared, and the scheduling system resumes the original plan; Anomaly confirmed: Update the risk value at this point to the highest level M(x, y) = 1, and restrict the robot from entering.

[0052] (3) Queue-jumping mechanism The manual re-inspection task has the highest priority, denoted as πmanual = 1. The system inserts it at the head of the task queue and executes as follows: The original queue task set T = {t1, t2,..., tn}; After insertion: T′ = {manual_review, t1,..., t n}.

[0053] Step 308, Anomaly type extension and system interface opening: (1) Customized anomaly type configuration The system supports extending anomaly types through configuration files, such as: The abnormal types include deviation from the path, lidar failure, image recognition failure, communication interruption, and equipment jamming, which are automatically classified and matched by the perception module after loading.

[0054] (2) Interface protocol definition Custom abnormal types: In addition to standard abnormalities such as deviation and sensing failure, new abnormal types such as jamming, collision, and communication interruption can be added through a configuration file and automatically loaded during system initialization.

[0055] Open interface: Provide a RESTful API or message bus interface to push abnormal data to a third-party operation and maintenance system, realizing multi-system linkage and secondary development.

[0056] The above embodiments are only used to illustrate the technical concept and features of the present invention, and the purpose is to enable ordinary technicians in the art to understand the content of the present invention and implement it accordingly. However, the protection scope of the present invention cannot be limited thereby. Any equivalent changes or modifications made according to the essence of the content of the present invention should be covered within the protection scope of the present invention.

Claims

1. A path planning method for an all-terrain autonomous inspection robot of high-voltage transmission and distribution lines, characterized in that Including: Based on a static prior map, conduct real-time risk assessment of static obstacles inside the target distribution line, and use a Markov chain to predict the risk of dynamic obstacles, which are defined as real-time risk and predicted risk respectively. Generate a comprehensive risk map by exponentially decaying and superimposing the real-time risk and the predicted risk; Generate a model map according to the target distribution line. Based on the model map, use the A* search algorithm with risk perception as the main path for global path planning, then introduce a deep Q network to fine-tune and correct the local path in real time. Finally, use a heuristic greedy-genetic algorithm hybrid strategy to iteratively optimize the task order, comprehensively consider task priorities and time window constraints, and output the optimal inspection scheduling order; The robot performs inspection tasks according to the constraints of the comprehensive risk map and the optimal inspection scheduling order. During the inspection tasks, an adaptive task feedback mechanism based on a multi-modal perception fusion model and Bayesian state estimation continuously and dynamically corrects the optimal inspection scheduling order.

2. The path planning method for the all-terrain autonomous inspection robot of high-voltage power transmission and distribution lines according to claim 1, wherein, The construction of the static prior risk map includes: Use indoor SLAM to perform offline mapping of the inside of the target distribution line to obtain a grid map of the clearance area of the distribution line; On the grid map, mark the walls, cabinet edges, and fixed equipment as high-risk areas to form the static prior risk map.

3. The path planning method for the all-terrain autonomous inspection robot of high-voltage power transmission and distribution lines according to claim 2, wherein, The process of the real-time risk assessment includes: Call the lidar and camera to collect obstacle distance information in each sensing cycle, map the obstacle distance information to the grid points of the grid map for dynamic risk perception, obtain the current real-time risk value of the robot, and insert the real-time risk value into the corresponding position in the static prior risk map to form a time-varying risk map.

4. The path planning method for the all-terrain autonomous inspection robot of high-voltage power transmission and distribution lines according to claim 3, characterized in that, The calculation method of the real-time risk value is: ； where d is the obstacle distance information, x and y are the current abscissa and ordinate of the robot, respectively, and λ d is the instantaneous decay rate corresponding to the obstacle distance information d. When d approaches 0, p approaches 1, and as the distance increases, the risk decays rapidly in an exponential form.

5. The path planning method for the all-terrain autonomous inspection robot of high-voltage transmission and distribution lines according to claim 3, characterized in that The process of the Markov chain prediction includes: For each grid point in the time-varying risk map, establish a first-order Markov chain, use a transition matrix to describe the risk transfer probability and risk distribution vector of the robot between adjacent grids, and calculate the predicted risk value according to the risk transfer probability and the risk distribution vector.

6. The path planning method for the all-terrain autonomous inspection robot of high-voltage transmission and distribution lines according to claim 1, characterized in that, The scheduling objective function of the heuristic greedy-genetic algorithm hybrid strategy is: ； Wherein, is the delay penalty weight coefficient for the k-th sensor cycle, is the delay amount for the k-th sensor cycle, d(v ti , v ti+1 ) is the path length between adjacent task points, θ is the path length weight, and m is the total number of tasks.

7. The path planning method for the all-terrain autonomous inspection robot of high-voltage power transmission and distribution lines according to claim 1, characterized in that, The construction process of the multi-modal perception fusion model includes: Let the observation of the \(i\)-th sensor at time \(t\) be \(z\) i (t), then the total observation vector is: ； Construct each state S k Conditional probability in the observation space: ； Among them, P(Z∣S k ) is obtained by fitting a Bayesian network with training data, where k is the number of state indices and j is the number of traversed indices.

Images