Dynamic path planning control method and system of photovoltaic panel cleaning unmanned aerial vehicle and storage medium

Abstract

The invention discloses a dynamic path planning control method and system for a photovoltaic panel cleaning unmanned aerial vehicle, and a storage medium, and relates to the field of unmanned aerial vehicle path planning. Collecting environment information of the photovoltaic power station and parameter information of the unmanned aerial vehicle in real time; an undirected topological graph is constructed based on photovoltaic power station environment information discretization, nodes of the undirected topological graph are temporary obstacle positions and photovoltaic panel surface pollution distribution conditions, and edges between the nodes represent unmanned aerial vehicle cleaning paths; by taking cleaning maximization as a target and taking parameter information of the unmanned aerial vehicle as a constraint condition of an unmanned aerial vehicle cleaning path, constructing a dynamic path planning model; the dynamic path planning model adopts an improved A * algorithm to adjust a preset initial path in real time according to the environment information and the equipment state parameters which are updated in real time, and an optimal cleaning path is generated. The invalid flight mileage of the unmanned aerial vehicle is reduced, and energy consumption and battery loss are directly reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of unmanned aerial vehicle path planning, in particular to a dynamic path planning control method and system for photovoltaic panel cleaning unmanned aerial vehicle and a storage medium. BACKGROUND

[0002] The existing photovoltaic cleaning unmanned aerial vehicle adopts a preset fixed path planning mode, that is, the trajectory is planned in advance according to the initial layout of the power station, and then the cleaning task is performed according to the trajectory. However, the actual power station operating environment is complex and dynamic, and this fixed path mode has significant defects. The core problem is concentrated in the difficulty in coping with multi-dimensional dynamic factor interference and the inability to realize the dynamic adaptation of cleaning demand and equipment state. On the one hand, the power station is often faced with dynamic interference such as gust, sand, temporary obstacles, etc. The fixed path cannot be adjusted in real time, which easily causes problems such as unmanned aerial vehicle collision, deviation from the trajectory, etc., resulting in cleaning missed cleaning, repeated cleaning, and reducing efficiency and safety. For example, desert sand will block the positioning signal, causing the unmanned aerial vehicle to lose, and strong gusts along the coast will affect the flight attitude. At the same time, the pollution distribution of photovoltaic panels is uneven, and the "one-size-fits-all" cleaning strategy of the fixed path not only causes resource waste, but also may not clean the key pollution area sufficiently. On the other hand, when multiple machines are working together, the fixed path lacks dynamic adaptation ability, and problems such as path overlap and conflict are easily caused. In addition, some unmanned aerial vehicles with basic self-path planning ability lack a multi-dimensional real-time data comprehensive decision-making mechanism, and the timeliness and accuracy of path adjustment are insufficient, and the cleaning efficiency and energy consumption are not balanced. In summary, the existing path planning technology has become a bottleneck restricting the development of photovoltaic cleaning unmanned aerial vehicles. Therefore, it is of urgent need and important value to develop a control method that can real-time integrate multi-source dynamic information and adaptively adjust the cleaning path to improve the operation efficiency, safety and economy, and promote the upgrading of power station automation operation. SUMMARY

[0003] Therefore, the present application provides a dynamic path planning control method and system for photovoltaic panel cleaning unmanned aerial vehicle and a storage medium to solve the problems in the background art.

[0004] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0005] A dynamic path planning control method for photovoltaic panel cleaning unmanned aerial vehicle, comprising the following steps:

[0006] Real-time collection of photovoltaic power station environment information and unmanned aerial vehicle self-parameter information, the photovoltaic power station environment information including temporary obstacle position and photovoltaic panel surface pollution distribution;

[0007] Discrete construction of an undirected topological graph based on photovoltaic power station environment information, the nodes of the undirected topological graph being temporary obstacle positions and photovoltaic panel surface pollution distribution, and the edges between the nodes representing unmanned aerial vehicle cleaning paths;

[0008] A dynamic path planning model is constructed by taking the self-parameter information of the unmanned aerial vehicle as a constraint condition of the unmanned aerial vehicle cleaning path, aiming at maximizing cleaning.

[0009] The dynamic path planning model generates an optimal cleaning path by using an improved A* algorithm to adjust the preset initial path in real time according to real-time updated environmental information and equipment state parameters.

[0010] Optionally, the improved A* algorithm realizes path planning through the following formula:

[0011] The path evaluation function is:

[0012] f(n)=g(n)+h(n);

[0013] The calculation formula of the cost function g(n) is:

[0014] g(n)=g0+λ1·L(n,n0)+λ1·E_fly(n)+λ3·E_clean(n);

[0015] In the formula, g0 is the cumulative cost from the starting point to the parent node n0, L(n,n0) is the path length between the node n and the parent node n0, E_fly(n) is the flight energy consumption of the path, E_clean(n) is the cleaning energy consumption of the path, λ1, λ2, and λ3 are dynamic adjustment coefficients.

[0016] The calculation formula of the heuristic function h(n) is:

[0017] h(n)=μ1·D(n,T)+μ2·W(n)+μ3·P(n);

[0018] In the formula, D(n,T) is the straight-line distance from the node n to the target point T, W(n) is the environmental interference weight of the region where the node n is located, P(n) is the pollution priority weight of the photovoltaic panel region corresponding to the node n, μ1, μ2, and μ3 are weight factors that are dynamically updated based on real-time monitoring data.

[0019] The above functions are used to realize the optimization of the path under multiple constraint conditions.

[0020] Optionally, the step of constructing a non-directional topological graph based on the photovoltaic power station environment information discretization: discretizing the photovoltaic power station operation area into grid cells according to a preset precision, collecting temporary obstacle information and photovoltaic panel pollution level data of each grid cell; taking the obstacle grid center as the obstacle node, the grid center with a pollution level exceeding a preset threshold as the high-pollution target node, and constructing the node set of the non-directional topological graph; when the connection between two nodes does not pass through the obstacle grid, a non-directional edge is established, and the edge weight is obtained by weighted calculation of the physical distance, the path area average pollution level and the environmental interference coefficient; the dynamic path planning model performs path search based on the non-directional topological graph, and updates the connection relationship and weight value of the edge in real time when the node attribute changes.

[0021] Optionally, the dynamic path planning model is constructed, specifically including the following steps:

[0022] Based on the adjacency matrix of the graph structure constructed by the dynamic path planning model, a multi-dimensional position encoding vector reflecting the characteristics of the graph structure is generated by matrix operation, and the position encoding matrix representing the structural position of the node in the graph is formed by splicing the position encoding vectors in each dimension according to a preset manner;

[0023] Based on the graph structure, different views are constructed, two sets of graph encoders with consistent parameter structures are used to process each view to generate node representation vectors, and a loss function is constructed by comparing the node representation vectors and introducing a temperature coefficient. After a preset number of training, the node initial representation for subsequent modeling is output;

[0024] Based on the node initial representation, the graph neural network is used to obtain the local context representation of the node, the local context representation and the position encoding matrix are fused to form the input information, and the input information is processed by the attention mechanism to interact and adjust the weight of the global information, and the updated node representation fused with local and global features is generated;

[0025] An improved path search algorithm is used to construct path evaluation criteria based on the target scene related features in the updated node representation and the execution device parameter features to perform initial path search, generate alternative paths by perturbing the initial path, select the optimal path according to the preset target function and device parameter constraints, and monitor the changes of the scene and the device parameters to trigger path re-planning.

[0026] Optionally, the dynamic path planning model has self-learning ability, which continuously optimizes the path planning algorithm parameters by analyzing historical cleaning data, to improve the efficiency and accuracy of path planning.

[0027] Optionally, the optimal cleaning path satisfies the following conditions: the overlap rate of adjacent cleaning trajectories is not more than 10%, and at least 2 repeated cleaning is performed on the photovoltaic panel area with a pollution level higher than a preset value.

[0028] A dynamic path planning control system of a photovoltaic panel cleaning unmanned aerial vehicle, comprising:

[0029] An environmental information and unmanned aerial vehicle information acquisition module: used for collecting photovoltaic power station environmental information and unmanned aerial vehicle self parameter information in real time, the photovoltaic power station environmental information including temporary obstacle position and photovoltaic panel surface pollution distribution situation;

[0030] An unmanned aerial vehicle cleaning path generation module: used for constructing an undirected topological graph based on photovoltaic power station environmental information discretization, the nodes of the undirected topological graph being temporary obstacle position and photovoltaic panel surface pollution distribution situation, and the edges between the nodes representing unmanned aerial vehicle cleaning paths;

[0031] A dynamic path planning model construction module: used for constructing a dynamic path planning model with cleaning maximization as the target and the self parameter information of the unmanned aerial vehicle as the constraint condition of the unmanned aerial vehicle cleaning path;

[0032] An optimal cleaning path generation module: used for generating an optimal cleaning path by using an improved A* algorithm to adjust a preset initial path in real time according to real-time updated environmental information and equipment state parameters.

[0033] A computer storage medium, the computer storage medium storing a computer program, the computer program being executed by a processor to implement the steps of the dynamic path planning control method of the photovoltaic panel cleaning unmanned aerial vehicle.

[0034] According to the above technical solution, compared with the prior art, the application provides a dynamic path planning control method, system and storage medium of a photovoltaic panel cleaning unmanned aerial vehicle, which has the following beneficial effects:

[0035] 1. The model is constructed with "cleaning maximization" as the target, the path is planned in combination with the photovoltaic panel surface pollution distribution, the seriously polluted area can be preferentially covered, and invalid paths can be avoided;Meanwhile, the path is dynamically adjusted based on real-time pollution conditions, so that repeated cleaning or omission is reduced, and the effective cleaning area per unit time is improved.

[0036] 2. The information of temporary obstacles (such as sudden shielding objects and other equipment) is collected in real time, and then the information is discretely processed through an undirected topological graph, and then the path is dynamically adjusted by using an improved A* algorithm, so that the temporary obstacles can be quickly avoided, the problem of "stuck" of the traditional fixed path in the face of sudden environmental changes is solved, and the cleaning task can be continuously carried out.

[0037] 3. The self parameters (such as endurance, load and maneuverability) of the unmanned aerial vehicle are used as the path constraint conditions, so that the planned path can match the actual performance of the unmanned aerial vehicle, the risks such as crash and failure caused by the path exceeding the equipment capacity (such as insufficient endurance and mismatched turning radius) are avoided, and the service life of the equipment is prolonged.

[0038] 4. Dynamic optimal path reduces the invalid flight mileage of drones, directly reducing energy consumption and battery wear; at the same time, it eliminates the need for frequent manual intervention to adjust the path, reducing manual operation and maintenance costs. Furthermore, due to more precise cleaning and a lower equipment failure rate, it indirectly reduces power generation losses caused by contamination of photovoltaic panels, thereby improving the overall economic benefits of the power station. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0040] Figure 1 This is a schematic diagram of the method flow provided by the present invention;

[0041] Figure 2 This is a schematic diagram of the system structure provided by the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] This invention discloses a dynamic path planning and control method for a photovoltaic panel cleaning drone, such as... Figure 1 As shown, it includes the following steps:

[0044] Real-time collection of environmental information from photovoltaic power plants and the drone's own parameter information. The environmental information from photovoltaic power plants includes the location of temporary obstacles and the distribution of pollution on the surface of photovoltaic panels.

[0045] An undirected topological graph is constructed based on the discretization of environmental information of photovoltaic power plants. The nodes of the undirected topological graph represent the locations of temporary obstacles and the distribution of pollution on the surface of photovoltaic panels, and the edges between nodes represent the cleaning paths of drones.

[0046] With the goal of maximizing cleaning, the drone's own parameter information is used as a constraint on the drone's cleaning path to construct a dynamic path planning model.

[0047] The dynamic path planning model uses an improved A* algorithm to adjust the preset initial path in real time based on the real-time updated environmental information and equipment status parameters, thereby generating the optimal cleaning path.

[0048] Further, the improved A* algorithm realizes path planning through the following formula:

[0049] The path evaluation function is:

[0050] f(n) = g(n) + h(n);

[0051] Wherein, the calculation formula of the cost function g(n) is:

[0052] g(n) = g0+ λ1·L(n, n0) + λ1·E_fly(n) + λ3·E_clean(n);

[0053] In the formula, g0 is the cumulative cost from the starting point to the parent node n0, L(n, n0) is the path length between the node n and the parent node n0, E_fly(n) is the flight energy consumption of the path, E_clean(n) is the cleaning energy consumption of the path, λ1, λ2, λ3 are dynamic adjustment coefficients;

[0054] The calculation formula of the heuristic function h(n) is:

[0055] h(n) = μ1·D(n, T) + μ2·W(n) + μ3·P(n);

[0056] In the formula, D(n, T) is the straight-line distance from the node n to the target point T, W(n) is the environmental interference weight of the region where the node n is located, P(n) is the pollution priority weight of the photovoltaic panel region corresponding to the node n, μ1, μ2, μ3 are weight factors dynamically updated based on real-time monitoring data;

[0057] The above functions are calculated to realize the optimization of the path under multiple constraint conditions.

[0058] Further, the improved A algorithm is a path planning algorithm based on the traditional A algorithm and optimized for the particularity of the photovoltaic panel cleaning unmanned aerial vehicle operation scene. Its core is to solve the shortcomings of the traditional A* algorithm in dynamic environment adaptability, energy consumption optimization and cleaning task specificity by introducing a dynamic weight adjustment mechanism and a multi-constraint condition fusion strategy. The specific improvements are as follows:

[0059] The heuristic function h(n) of the traditional A* algorithm usually adopts Euclidean distance or Manhattan distance, only considering the shortest geometric path. The improved algorithm designs the heuristic function as a multi-factor weighted form:

[0060] h(n) = α·d(n, goal) + β·w_wind(n) + γ·w_pollution(n)

[0061] Wherein, d(n, goal) is the straight-line distance from node n to the target point, w_wind(n) is the influence weight of wind speed on the path (the greater the wind speed, the higher the weight, guiding the path to avoid strong wind area), w_pollution(n) is the pollution degree weight (the more serious the pollution, the higher the weight, preferentially planning the path through the high pollution area). Alpha, beta, gamma are dynamic adjustment coefficients, which are self-adaptively adjusted according to real-time environmental data (such as wind speed mutation, pollution distribution change), so that the path planning takes into account efficiency and task demand.

[0062] The energy consumption model is introduced in the cost function g(n), which associates flight energy consumption, cleaning energy consumption and path length:

[0063] g(n) = g(parent(n)) + k·L(n, parent(n)) + δ·C_wash(n)

[0064] Wherein L(n, parent(n)) is the path length between nodes, k is the unit distance flight energy consumption coefficient, C_wash(n) is the cleaning energy consumption of the section (positively correlated with pollution degree), and δ is the cleaning energy consumption weight. Through the model, the algorithm can preferentially select the path with lower energy consumption (such as reducing repeated round trips in low pollution areas), and prolong the endurance of the unmanned aerial vehicle for single operation.

[0065] For temporary obstacles (such as maintenance equipment, birds), the algorithm sets a "dynamic safety buffer zone": when the sensor detects an obstacle, the obstacle coordinates and movement trend are updated in real time, and a no-fly zone with a radius dynamically changing with the movement speed of the obstacle is generated in the path search. At the same time, the "segment re-planning" strategy is adopted, only the local path within the influence range of the obstacle is recalculated, avoiding the efficiency loss caused by global path reconstruction, and ensuring that the obstacle avoidance response time is controlled within 0.5 seconds.

[0066] In the multi-airport scenario, the algorithm assigns a unique "path priority" label to each unmanned aerial vehicle, and triggers conflict detection at the path intersection point through real-time interaction of position information. When potential collision risk is detected, the local path is dynamically adjusted based on priority: the low-priority unmanned aerial vehicle temporarily deviates from the original trajectory, generating a circumnavigation path (deviation distance not more than 1 meter, ensuring uninterrupted cleaning coverage), and then returning to the planned path after the high-priority unmanned aerial vehicle passes, realizing collision-free cooperative operation.

[0067] For the regular layout of photovoltaic panel array (mostly rectangular arrangement), when the algorithm gridding the path search space, it forces the distance between adjacent trajectories to match the width of the photovoltaic panel (error ≤ 5 cm), avoiding missed cleaning. At the same time, a "secondary cleaning marker" is set for the high pollution area, which automatically inserts a turning node when generating the path, ensuring that the area is covered at least twice, and the two cleaning trajectories intersect at 90°, improving the cleaning effect.

[0068] Further, the step of constructing the undirected topological graph based on the photovoltaic power station environmental information discretization: discretize the photovoltaic power station operation area into grid cells according to a preset precision, collect temporary obstacle information and photovoltaic panel pollution level data of each grid cell; take the obstacle grid center as the obstacle node, the grid center with a pollution level exceeding a preset threshold as the high-pollution target node, and construct the node set of the undirected topological graph; when the connection between two nodes does not pass through the obstacle grid, an undirected edge is established, and the edge weight is obtained by weighted calculation of the physical distance, the average pollution level of the path area, and the environmental interference coefficient; the dynamic path planning model performs path search based on the undirected topological graph, and updates the connection relationship and weight value of the edge in real time when the node attribute changes.

[0069] Further, the dynamic path planning model is constructed, specifically including the following steps:

[0070] (1) Generate a position encoding matrix: construct a degree matrix based on the adjacency matrix of the undirected topological graph, calculate the Laplacian matrix by the adjacency matrix and the degree matrix and perform eigenvalue decomposition, select the first several low-frequency eigenvectors as the Laplacian position encoding vector; calculate the shortest path distance between any two nodes and construct a path distance matrix, and generate a path position encoding vector by dimension reduction through a nonlinear mapping function; calculate the node transition probability based on the adjacency matrix and construct a transition probability matrix, and generate a random walk position encoding vector by statistically analyzing the node access frequency through multiple rounds of random walk; concatenate the three types of encoding vectors according to the column dimension to form a position encoding matrix;

[0071] (2) Perform self-supervised graph pre-training: construct an original view and a perturbed view based on the initial graph structure, the perturbed view is obtained by mask processing of part of the feature dimensions of the node feature matrix, random discarding of some edge connection relationships of the adjacency matrix, and random subgraph sampling of the graph structure; use two sets of graph attention networks with the same parameter structure as the main encoder and the target encoder, update the parameters of the main encoder through back propagation of the gradient, and update the parameters of the target encoder through exponential moving average of the parameters of the main encoder; take the cosine similarity of the representation vectors of the same node in the two views as positive samples, and the cosine similarity of the representation vectors between different nodes as negative samples, introduce a temperature coefficient to construct a self-supervised contrast loss function, and output the main encoder output as the initial representation of the communication node after a preset number of training;

[0072] (3) Perform local message passing and global attention modeling: based on the initial representation of the communication node, perform local message passing operation using graph convolution network, and obtain the node local context representation through formula (H (l+1) = σ (D -1 / 2 A -1 / 2 D (l) W (l) H (l) is the node representation of the lth layer, D is the degree matrix, A is the adjacency matrix, W(l) is the nonlinear activation function; the local context representation and the position encoding matrix are spliced according to the feature dimension to form a fusion input matrix, and the fusion input matrix is subjected to information interaction and weight adjustment through a multi-head attention mechanism to generate an updated node representation of fused local and global information;

[0073] (4) Dynamic path search and optimization is implemented: an improved A* algorithm is used to integrate the pollution level of the photovoltaic panel and the self-parameter adaptation degree feature of the unmanned aerial vehicle in the heuristic function and the cost function in the updated node representation to perform initial path search; a disturbance operation is performed on the intermediate nodes in the initial path to construct a plurality of disturbance paths, the maximum cleaning objective function value and the satisfaction degree of the unmanned aerial vehicle parameter constraint of each path are calculated, and the path with the maximum objective function value and satisfying the constraint is selected as the final cleaning path, and the environment and the unmanned aerial vehicle parameter change are monitored in real time to trigger path re-planning to update the cleaning path.

[0074] Further, the dynamic path planning model has self-learning ability, which continuously optimizes the path planning algorithm parameters by analyzing historical cleaning data, for improving the efficiency and accuracy of path planning.

[0075] Further, the optimal cleaning path satisfies the following conditions: the overlap rate of adjacent cleaning trajectories is not more than 10%, and at least 2 repeated cleanings are performed on the photovoltaic panel region with a pollution degree higher than a preset value.

[0076] The technical solutions of the application will be explained below with a specific example.

[0077] Firstly, the space of the photovoltaic power station operation area is discretized. The area covered by the entire photovoltaic panel array is divided into uniform grid units according to a preset precision (such as 1 meter x 1 meter), and each grid unit is taken as a basic unit of environmental information collection. Through the multi-sensor (visual sensor, multi-spectral camera, etc.) carried by the unmanned aerial vehicle, the environmental data of each grid unit is collected in real time, including: 1) temporary obstacle information (such as maintenance equipment, bird flocks, etc., marked as "obstacle grid"); 2) photovoltaic panel surface pollution data (through multi-spectral reflectivity analysis, the pollution degree is quantified as a level value of 0-10, stored as a "pollution grid" attribute).

[0078] Subsequently, an undirected topological graph is constructed based on the discretized environmental information. Two types of key environmental feature points are set as nodes of the graph: one is the center coordinates of all "obstacle grids", which are taken as obstacle nodes, and their position coordinates and obstacle types (static / dynamic) are marked; the other is the center of the area with a pollution level exceeding a preset threshold (such as level >= 5) in the "pollution grid", which is taken as a high-pollution target node, and its position coordinates and pollution level are marked. For the remaining non-obstacle and low-pollution grids, they are not separately set as nodes, but as passable areas for path planning.

[0079] The edges between the nodes are generated by the following rules: for any two non-obstacle nodes (including high-pollution target nodes and virtual reference nodes of non-obstacle areas), if the straight line between the two nodes falls completely in the passable area (i.e., does not pass through any obstacle grid), an undirected edge is established. The weight of the edge comprehensively reflects the path cost, and the calculation formula is: edge weight = a x physical distance + b x average pollution level + g x environmental interference coefficient, wherein a, b and g are weight coefficients, the physical distance is the Euclidean distance of the two nodes, the average pollution level is the average of the pollution levels of the grids passed by the path, and the environmental interference coefficient is positively correlated with the real-time wind speed and sand concentration of the path area.

[0080] Finally, path search is performed based on the undirected topological graph. The improved A* algorithm takes the nodes in the graph as the key control points of path planning, preferentially selects high-pollution target nodes as the must-pass points in the search process, and dynamically adjusts the path direction through the edge weight-when it is detected that the movement of temporary obstacles causes the attribute of the node to change (such as the position update of the dynamic obstacle node), the invalid edge is deleted and the new edge is reconstructed in real time, the dynamic update of the topological graph is realized, and it is ensured that the unmanned aerial vehicle always completes the cleaning operation along the passable path with the optimal cost.

[0081] Corresponding to the method shown in Figure 1 The application also discloses a dynamic path planning control system of the unmanned aerial vehicle for Figure 1 The implementation of the method, and the specific structure is as shown in Figure 2 The application also discloses a dynamic path planning control system of the unmanned aerial vehicle for

[0082] The environmental information and the unmanned aerial vehicle information acquisition module is used for collecting the environmental information of the photovoltaic power station and the self parameter information of the unmanned aerial vehicle in real time, and the environmental information of the photovoltaic power station includes the position of the temporary obstacle and the pollution distribution on the surface of the photovoltaic panel.

[0083] The unmanned aerial vehicle cleaning path generation module is used for constructing an undirected topological graph based on the discretization of the environmental information of the photovoltaic power station, the nodes of the undirected topological graph are the position of the temporary obstacle and the pollution distribution on the surface of the photovoltaic panel, and the edges between the nodes represent the unmanned aerial vehicle cleaning path.

[0084] The dynamic path planning model construction module is used for taking the self parameter information of the unmanned aerial vehicle as the constraint condition of the unmanned aerial vehicle cleaning path, and constructing a dynamic path planning model with the maximum cleaning as the target.

[0085] The optimal cleaning path generation module is used for generating an optimal cleaning path by using the improved A* algorithm to adjust the preset initial path in real time according to the real-time updated environmental information and the equipment state parameter.

[0086] The embodiment further discloses a computer storage medium, and the computer storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the dynamic path planning control method of the unmanned aerial vehicle for cleaning photovoltaic panels.

[0087] The various embodiments are described in a progressive manner in the specification, and each embodiment focuses on the difference from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. For the device disclosed by the embodiments, since it corresponds to the method disclosed by the embodiments, the description is relatively simple, and the related parts can be referred to the method part.

[0088] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications of the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A dynamic path planning and control method for a photovoltaic panel cleaning drone, characterized in that, Includes the following steps: Real-time collection of environmental information from photovoltaic power plants and the drone's own parameter information. The environmental information from photovoltaic power plants includes the location of temporary obstacles and the distribution of pollution on the surface of photovoltaic panels. An undirected topological graph is constructed based on the discretization of environmental information of photovoltaic power plants. The nodes of the undirected topological graph represent the locations of temporary obstacles and the distribution of pollution on the surface of photovoltaic panels, and the edges between nodes represent the cleaning paths of drones. With the goal of maximizing cleaning, the drone's own parameter information is used as a constraint on the drone's cleaning path to construct a dynamic path planning model. The dynamic path planning model uses an improved A* algorithm to adjust the preset initial path in real time based on the real-time updated environmental information and equipment status parameters, thereby generating the optimal cleaning path.

2. The dynamic path planning and control method for a photovoltaic panel cleaning drone according to claim 1, characterized in that, The improved A* algorithm achieves path planning using the following formula: The path evaluation function is: f(n) = g(n) + h(n); The cost function g(n) is calculated using the following formula: g(n)=g0+λ1·L(n,n0)+λ1·E_fly(n)+λ3·E_clean(n); In the formula, g0 is the cumulative cost from the starting point to the parent node n0, L(n,n0) is the path length between node n and parent node n0, E_fly(n) is the flight energy consumption of this path segment, E_clean(n) is the cleaning energy consumption of this path segment, and λ1, λ2, and λ3 are dynamic adjustment coefficients. The formula for calculating the heuristic function h(n) is: h(n)=μ1·D(n,T)+μ2·W(n)+μ3·P(n); In the formula, D(n,T) is the straight-line distance from node n to target point T, W(n) is the environmental disturbance weight of the area where node n is located, P(n) is the pollution priority weight of the photovoltaic panel area corresponding to node n, and μ1, μ2, and μ3 are weight factors that are dynamically updated based on real-time monitoring data. The above function is used to calculate and optimize the path under multiple constraints.

3. The dynamic path planning and control method for a photovoltaic panel cleaning drone according to claim 1, characterized in that, The steps for constructing an undirected topological graph based on the discretization of environmental information of photovoltaic power plants are as follows: the working area of ​​the photovoltaic power plant is discretized into grid cells with a preset accuracy, and information on temporary obstacles and the pollution level of photovoltaic panels in each grid cell is collected. Using the center of the obstacle grid as the obstacle node and the center of the grid with a pollution level exceeding a preset threshold as the high pollution target node, a node set of an undirected topology graph is constructed. When the connection between two nodes does not pass through the obstacle grid, an undirected edge is established. The edge weight is obtained by weighted calculation based on physical distance, average pollution level of the path area, and environmental interference coefficient. The dynamic path planning model performs path search based on this undirected topology graph and updates the connection relationship and weight value of the edge in real time when the node attributes change.

4. The dynamic path planning and control method for a photovoltaic panel cleaning drone according to claim 1, characterized in that, Constructing a dynamic path planning model includes the following steps: Based on the adjacency matrix of the graph structure constructed by the dynamic path planning model, a multi-dimensional position encoding vector reflecting the characteristics of the graph structure is generated through matrix operations. The position encoding vectors of each dimension are concatenated in a preset way to form a position encoding matrix representing the structural position of the node in the graph. Based on the graph structure, different views are constructed. Two sets of graph encoders with the same parameter structure are used to process each view to generate node representation vectors. Positive and negative samples are distinguished by comparing the node representation vectors and a temperature coefficient is introduced to construct a loss function. After a preset number of training rounds, the initial node representations for subsequent modeling are output. Based on the initial node representation, a graph neural network is used to obtain the local context representation of the node. The local context representation is fused with the position encoding matrix to form the input information. The input information is then subjected to global information interaction and weight adjustment through an attention mechanism to generate an updated node representation that integrates local and global features. An improved path search algorithm is adopted. The path evaluation criteria are constructed by combining the target scene-related features and execution device parameter features in the updated node representation to perform initial path search. The initial path is perturbed to generate alternative paths. The optimal path is selected according to the preset objective function and device parameter constraints. The changes in scene and device parameters are monitored to trigger path replanning.

5. The dynamic path planning and control method for a photovoltaic panel cleaning drone according to claim 1, characterized in that, The dynamic path planning model has self-learning capabilities. By analyzing historical cleaned data, it continuously optimizes the path planning algorithm parameters to improve the efficiency and accuracy of path planning.

6. The dynamic path planning and control method for a photovoltaic panel cleaning drone according to claim 1, characterized in that, The optimal cleaning path meets the following conditions: the overlap rate of adjacent cleaning trajectories does not exceed 10%, and the photovoltaic panel area with a pollution level higher than the preset value is cleaned at least twice.

7. A dynamic path planning and control system for a photovoltaic panel cleaning drone, characterized in that, include: Environmental information and UAV information acquisition module: used to collect real-time environmental information of photovoltaic power station and UAV's own parameter information. The environmental information of photovoltaic power station includes the location of temporary obstacles and the distribution of pollution on the surface of photovoltaic panels; Drone cleaning path generation module: used to construct an undirected topology graph based on the discretized environmental information of photovoltaic power plants. The nodes of the undirected topology graph represent the locations of temporary obstacles and the distribution of contamination on the surface of photovoltaic panels, and the edges between nodes represent the drone cleaning path. Dynamic path planning model construction module: This module is used to construct a dynamic path planning model with the goal of maximizing cleaning efficiency, using the drone's own parameter information as constraints on the drone's cleaning path. Optimal Cleaning Path Generation Module: This module is used by the dynamic path planning model to adjust the preset initial path in real time based on the real-time updated environmental information and equipment status parameters, and to generate the optimal cleaning path using an improved A* algorithm.

8. A computer storage medium, characterized in that, The computer storage medium stores a computer program, which, when executed by a processor, implements the steps of a dynamic path planning and control method for a photovoltaic panel cleaning drone as described in any one of claims 1-6.

Images