Photovoltaic panel site intelligent inspection method and system and electronic equipment

Abstract

The invention discloses a photovoltaic panel site intelligent inspection method and system and electronic equipment, and the method comprises the steps: collecting site information of a photovoltaic power station to be inspected, the site information comprising surrounding environment information and terrain geographic information; based on the site information, constructing a three-dimensional digital simulation scene corresponding to the photovoltaic power station to be inspected; planning an optimal inspection path based on the three-dimensional digital simulation scene; and through the optimal inspection path, constructing a path prediction model of the unmanned aerial vehicle by adopting a long short-term neural network LSTM, and introducing a reinforcement learning mechanism to perform iterative training on the path prediction model of the unmanned aerial vehicle, thereby optimizing the flight strategy and path planning capability of the unmanned aerial vehicle. According to the invention, automatic inspection of a photovoltaic panel site is realized, maintenance and supervision are carried out by only a small number of personnel, the labor cost is greatly reduced, and unnecessary loss of the unmanned aerial vehicle in the flight process is reduced through an optimized flight path and intelligent flight control.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent inspection of photovoltaic panel sites, and particularly to an intelligent inspection method, system and electronic device for photovoltaic panel sites. Background Art

[0002] With the rapid development of photovoltaic power generation, the construction and management of large-scale photovoltaic power stations have put forward higher requirements for the inspection of photovoltaic panels. Facing the vast power station area, numerous devices and complex geographical environment, the traditional manual inspection method is gradually becoming inadequate. When manually inspecting photovoltaic devices in desert, mountainous and water areas, it is very difficult to conduct a comprehensive and detailed inspection. Coupled with the inconvenience caused by the installation height, the obstacles caused by the rugged terrain and the interference caused by the changeable weather, there are many loopholes in manual inspection, and it is difficult to meet the requirements for the quality and frequency of inspection. The operation and maintenance of power stations face certain challenges.

[0003] The introduction of unmanned aerial vehicle (UAV) technology has brought a revolutionary change to the operation and maintenance management of photovoltaic power stations. Through high-precision flight control and advanced monitoring equipment, it is possible to achieve a comprehensive and detailed inspection of the components of photovoltaic power stations.

[0004] However, there are still many problems in other related technologies. For example, the UAV has poor adaptability for inspection in complex environments, and it is difficult to quickly analyze and process a large amount of recorded data. The lack of intelligence always plans according to a specific path, which will lead to incomplete inspection coverage, a sharp increase in the power consumption of the UAV due to the inspection time-consuming, and accidents and failures when encountering emergencies such as biological obstacles. Summary of the Invention

[0005] In order to solve the above technical problems existing in the prior art, the present invention proposes an intelligent inspection method, system and electronic device for photovoltaic panel sites, realizing the automatic inspection of photovoltaic panel sites, greatly reducing the labor cost and lowering the equipment replacement and maintenance costs.

[0006] On the one hand, to achieve the above object, the present invention provides an intelligent inspection method for photovoltaic panel sites, including:

[0007] Collecting the site information of the photovoltaic power station to be inspected, wherein the site information includes the surrounding environment information and the topographic and geographical information;

[0008] Based on the site information, constructing a three-dimensional digital simulation scene corresponding to the photovoltaic power station to be inspected;

[0009] Planning an optimal inspection path based on the three-dimensional digital simulation scene;

[0010] Through the optimal inspection path, a path prediction model of the UAV is constructed using the long short-term neural network (LSTM), and a reinforcement learning mechanism is introduced to iteratively train the path prediction model of the UAV, optimizing the flight strategy and path planning ability of the UAV.

[0011] Preferably, the surrounding environment information includes the distribution of vegetation, the coverage of shadows, and the positions of obstacles; the topographic and geographical information includes the terrain height, the ground flatness, and the terrain undulation.

[0012] Preferably, constructing a three-dimensional digital simulation scene corresponding to the photovoltaic power station to be inspected includes:

[0013] Based on the building information model technology (BIM) of digital modeling, the site information is imported into BIM software for three-dimensional modeling to obtain the three-dimensional digital simulation scene.

[0014] Preferably, planning the optimal inspection path includes:

[0015] Using the improved ant colony algorithm to search for the optimal solution in the inspection path space by simulating the optimization process in nature, and using the shortest path length and the largest coverage range as evaluation indicators to obtain the optimal inspection path;

[0016] Among them, the improved ant colony algorithm optimizes path search by dynamically adjusting the heuristic factor and the pheromone update strategy.

[0017] Preferably, the improved ant colony algorithm includes:

[0018] Initializing the inspection point set and the pheromone matrix;

[0019] Generating an initial path through the fitness function by integrating the path length and the coverage range;

[0020] Dynamically adjusting the heuristic factor and the pheromone evaporation coefficient to enhance the global search ability;

[0021] Introducing a local search optimization strategy to correct the path and outputting the inspection path with the highest fitness value, that is, the optimal inspection path.

[0022] Preferably, the fitness function is:

[0023]

[0024] In the formula, D is the total path length; λ is the coverage range; are all weight parameters; h is the fitness function.

[0025] Preferably, the iterative training of the path prediction model of the UAV includes:

[0026] Normalize the optimal inspection path data;

[0027] Use an LSTM network to extract temporal features and predict the flight coordinates of the drone;

[0028] Optimize the flight strategy by combining the reward function of reinforcement learning to obtain the trained path prediction model of the drone. Among them, the reward function is dynamically adjusted based on path deviation, coverage completion, and obstacle avoidance.

[0029] Preferably, the reinforcement learning adopts the Deep Deterministic Policy Gradient (DDPG) algorithm, and gradually optimizes the flight control parameters of the drone through alternating training of the simulation environment and the actual site.

[0030] On the other hand, to achieve the above object, the present invention also provides a photovoltaic panel site intelligent inspection system, including:

[0031] Data acquisition module: used to collect the site information of the photovoltaic power station to be inspected;

[0032] Scene construction module: used to construct a three-dimensional digital simulation scene of the photovoltaic power station to be inspected through the site information;

[0033] Path planning module: used to plan the optimal inspection path through the three-dimensional digital simulation scene;

[0034] Drone training module: used to construct a path prediction and control model of the drone based on the optimal inspection path, and optimize the flight strategy and path planning of the drone.

[0035] The present invention also provides an electronic device, in which computer instructions are stored, and when the computer instructions are executed by a processor, the photovoltaic panel site intelligent inspection method described above is implemented.

[0036] Compared with the prior art, the present invention has the following advantages and technical effects:

[0037] (1) This application plans the optimal inspection path for the photovoltaic panel site by improving the ant colony algorithm. Compared with traditional manual planning or simple algorithm planning, it can effectively reduce the flight distance and time of the unmanned aerial vehicle (UAV), enabling the UAV to complete the inspection task with the shortest path, thereby significantly improving the inspection efficiency. The path prediction and control model constructed using the long short-term neural network (LSTM) can perform deep learning training based on the existing inspection path data, enabling the UAV to more accurately predict and control the flight path during flight, reducing flight deviations caused by human operation or environmental interference, and further enhancing the inspection speed and accuracy. By introducing a reinforcement learning mechanism to optimize the LSTM model, it can continuously adjust and improve the flight strategy according to the real-time feedback during the inspection process, enabling the UAV to make more reasonable decisions when facing different environmental conditions and emergencies, further improving the inspection quality, and ensuring the timely detection of faults and problems with the photovoltaic panels.

[0038] (2) This application realizes the automated inspection of the photovoltaic panel site, only requiring a small number of personnel for maintenance and supervision, greatly reducing the labor cost. At the same time, the optimized flight path and intelligent flight control reduce the unnecessary losses of the UAV during flight, reducing the damage to the UAV equipment caused by operations such as frequent takeoffs and landings and sharp turns, thereby extending the service life of the UAV and reducing the equipment replacement and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0040] Figure 1 is a schematic structural diagram of an intelligent inspection system for a photovoltaic panel site according to an embodiment of the present invention;

[0041] Figure 2 is an implementation flowchart of an intelligent inspection system for a photovoltaic panel site according to an embodiment of the present invention;

[0042] Figure 3 is a schematic structural diagram of an electronic device entity according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the drawings and combine the embodiments to detail this application.

[0044] It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0045] The present invention proposes an intelligent inspection method for a photovoltaic panel site, as Figure 2 follows:

[0046] Collect the site information of the photovoltaic power station to be inspected, where the site information includes surrounding environment information and topographic and geographical information;

[0047] Based on the site information, construct a three-dimensional digital simulation scene corresponding to the photovoltaic power station to be inspected;

[0048] Based on the three-dimensional digital simulation scene, plan the optimal inspection path;

[0049] Through the optimal inspection path, use the long short-term neural network LSTM to construct a path prediction model for the unmanned aerial vehicle (UAV), and introduce a reinforcement learning mechanism to iteratively train the path prediction model of the UAV, so as to optimize the flight strategy and path planning ability of the UAV.

[0050] Furthermore, the surrounding environment information includes the distribution of vegetation, the coverage of shadows, and the positions of obstacles; the topographic and geographical information includes the terrain height, the ground flatness, and the terrain undulation.

[0051] Specifically, the surrounding environment information of the photovoltaic panel site includes: the distribution of vegetation (such as trees, grasslands, etc.), the coverage of shadows (specifically referring to the areas where the light is dim or missing due to the occlusion of objects or terrain), and obstacles (such as buildings, fences, poles, etc.).

[0052] The vegetation situation can help understand the natural occlusion degree of the site, while the shadow situation is related to the lighting conditions of the photovoltaic panels at different times of the day. The existence of obstacles may affect the layout and overall efficiency of the photovoltaic panels.

[0053] The topographic and geographical information of the photovoltaic panel site includes: terrain height information, ground flatness, and terrain undulation information. Considering the flight height and flight speed of the UAV during the inspection process, understanding the slope and undulation of the site can avoid unnecessary shaking or collision risks of the UAV during takeoff, landing, or flight; at the same time, the terrain height information can help evaluate the drainage situation and possible soil erosion problems of the site. The ground flatness is directly related to the convenience and stability of the installation of photovoltaic panels, and the terrain undulation condition has an important impact on the layout and design of photovoltaic panels.

[0054] Furthermore, constructing a three-dimensional digital simulation scene corresponding to the photovoltaic power station to be inspected includes:

[0055] Import the site information into BIM software based on the building information model technology BIM for digital modeling to obtain the three-dimensional digital simulation scene.

[0056] Specifically, a three-dimensional digital simulation scenario of the photovoltaic power station to be inspected is constructed by using the Building Information Modeling (BIM) technology based on a digital model. The specific method includes: performing three-dimensional modeling of the photovoltaic power station by using BIM software. BIM technology is a building information model technology based on digital modeling, which can efficiently perform design, simulation, and collaboration. By importing the parameters collected by the acquisition module into the BIM software, three-dimensional modeling is carried out to obtain the three-dimensional simulation scenario of the photovoltaic power station to be inspected.

[0057] Furthermore, plan the optimal inspection path, including:

[0058] Using the improved ant colony algorithm, by simulating the optimization process in nature, search for the optimal solution in the inspection path space, and use the shortest path length and the largest coverage range as evaluation indicators to obtain the optimal inspection path;

[0059] Among them, the improved ant colony algorithm optimizes path search by dynamically adjusting the heuristic factor and the pheromone update strategy.

[0060] Specifically, adopt the improved ant colony algorithm to plan the optimal inspection path by continuously updating and iterating until the optimal solution in the inspection path space is found.

[0061] In this embodiment, taking the generated simulation scenario as the input, use the improved ant colony algorithm to search for the best solution in the inspection path space by simulating the optimization process in nature.

[0062] Furthermore, the improved ant colony algorithm includes:

[0063] Initialize the inspection point set and the pheromone matrix;

[0064] Generate an initial path through the fitness function by comprehensively considering the path length and the coverage range;

[0065] Dynamically adjust the heuristic factor and the pheromone evaporation coefficient to enhance the global search ability;

[0066] Introduce a local search optimization strategy to correct the path, and output the inspection path with the highest fitness value, that is, the optimal inspection path.

[0067] Specifically, by constructing a fitness function (i.e., the evaluation indicator), the quality of the inspection path is quantified. In this embodiment, the evaluation indicators are: the shortest path length and the largest coverage range. After the fitness function is established, the planning module immediately generates an initial inspection path. Subsequently, it is evaluated according to the fitness function, and the fitness value of each path is calculated. The higher the fitness value, the better the optimization degree of the path.

[0068] Adopting the improved ant colony algorithm to plan the optimal path includes:

[0069] 1. Initialize parameters:

[0070] First, set the basic parameters of the ant colony: the number of ants n, the pheromone evaporation coefficient α, the initial pheromone value ω0, the pheromone importance factor k, the heuristic information importance factor s, the maximum number of iterations pop_max, etc. Initialize the inspection point set Q = {q1, q2,... q m} of the photovoltaic panel site. The position coordinates of each inspection point are known, and the heuristic information matrix is defined where d ij is the distance between inspection points q i and q j .

[0071] 2. Initial pheromone matrix:

[0072] Initialize the initial pheromone ω ij of each path to ω0,

[0073] 3. Define the fitness function:

[0074] The fitness function h comprehensively considers the path length and the coverage range:

[0075]

[0076] where D is the total path length; λ is the coverage range (calculated according to the effective coverage area of the inspection points); are all weight parameters, satisfying

[0077] 4. Randomly select the starting point:

[0078] Each ant starts from a randomly selected inspection point q i .

[0079] 5. Construct the path:

[0080] The probability that ant p selects the next inspection point q i from the current inspection point q j is:

[0081]

[0082] where is the set of neighbor nodes accessible to ant p; ω ij is the pheromone concentration on the path; δ ij is the heuristic information.

[0083] 6. Update the taboo list:

[0084] Each ant records the inspected points that have been visited and adds the visited points to the taboo list U p .

[0085] 7. Completion path:

[0086] When all inspection points have been visited, return to the starting point to form a complete inspection path.

[0087] 8. Pheromone evaporation:

[0088] The pheromones of all paths decay proportionally:

[0089] ω ij = (1 - α)·ω ij .

[0090] 9. Pheromone enhancement:

[0091] Update the pheromone according to the fitness function h for the path completed by each ant:

[0092]

[0093] At this time, the path contains the edge (i, j), where A is a constant; D p is the total path length.

[0094] Update the pheromone by integrating the contributions of all ants:

[0095]

[0096] 10. Introduce improvement strategies to adjust dynamic parameters:

[0097] Dynamically adjust k, s, and α according to the number of iterations to enhance the global search ability of the algorithm:

[0098]

[0099] In the formula, k t is the value of the dynamic parameter at the t-th iteration, k0 represents the initial value of this dynamic parameter, pop_max is the maximum number of iterations, and t is the current number of iterations.

[0100] 11. Heuristic coverage correction:

[0101] Incorporate the coverage area into the heuristic information matrix:

[0102]

[0103] where λ ij is the total area covered from point q i to point q j .

[0104] 12. Local search optimization:

[0105] Apply a local optimization algorithm to the paths generated by each ant to further reduce the path length.

[0106] 13. Iterative update:

[0107] Repeat path construction, pheromone update, and improvement strategies until the maximum number of iterations pop_max is reached or the fitness value converges.

[0108] 14. Output the optimal path:

[0109] Return the path with the highest fitness function value among all iterations as the optimal inspection path.

[0110]

[0111] Where Q p is the maximum value of all paths traveled by ant p, D p is the total length of the path traveled by ant p, used to calculate the fitness function value, λ p is the coverage range traveled by ant p, Q * is the optimal inspection path.

[0112] Furthermore, the iterative training of the path prediction model for the UAV includes:

[0113] Normalize the optimal inspection path data;

[0114] Use an LSTM network to extract temporal features and predict the flight coordinates of the UAV;

[0115] Combine the reward function of reinforcement learning to optimize the flight strategy and obtain the trained path prediction model of the UAV, where the reward function is dynamically adjusted based on path deviation, coverage completion, and obstacle avoidance.

[0116] Specifically, use a long short-term neural network (LSTM) and further optimize the model performance using deep learning algorithms and reinforcement learning algorithms to guide the UAV to learn the optimal flight strategy after multiple iterative trainings in the simulation environment and the actual photovoltaic panel site.

[0117] After normalizing the recorded inspection path data, use a long short-term neural network (LSTM) to construct a path prediction and control model for the UAV, and further optimize the model performance using deep learning algorithms and reinforcement learning algorithms to guide the UAV to learn the optimal flight strategy after multiple iterative trainings in the simulation environment and the actual photovoltaic panel site, and finally achieve an efficient and accurate photovoltaic panel site inspection task.

[0118] Specifically as follows:

[0119] 1. Record data:

[0120] Use the improved ant colony algorithm to plan the optimal inspection path for the photovoltaic panel site.

[0121] Output result: Optimal path Q * : The optimal access order including all inspection points.

[0122] Path attribute records: Total path length D, coverage range λ of each inspection point, geographical coordinates (latitude and longitude or plane coordinates) of the inspection points.

[0123] Record the optimal path and related attributes in a standardized format CSV to a file.

[0124] 2. Normalization processing:

[0125] Perform normalization processing on the input path attribute data to ensure that the data is suitable for LSTM model training.

[0126] Use the normalization formula:

[0127]

[0128] In the formula, x is the original data value, x min 、x max are the minimum and maximum values of the data respectively.

[0129] Divide the dataset into a training set, a validation set, and a test set.

[0130] 3. Use LSTM to process and predict the flight path:

[0131] Model architecture:

[0132] Input layer: Normalized inspection path data (in time series form);

[0133] Hidden layer: Multiple LSTM layers to extract the temporal features of the path sequence;

[0134] Output layer: Predict the coordinates and distance of the next flight point.

[0135] Build the model:

[0136] r t = σ(V x x t + V r x t-1 + e r );

[0137] Among them, r t is the hidden state at the current time step t; V x x t is the input x t and the weight matrix V xLinear combination of; V r x t-1 Is the hidden state at the previous time step r t-1 And the linear combination of the hidden weight matrix Vr; e r Is the hidden bias term, and σ is the activation function.

[0138] Test the model on the validation set and calculate errors such as RMSE and MAE.

[0139] 4. Introduce reinforcement learning to optimize the model:

[0140] Define the reinforcement learning framework:

[0141] State S: The current position of the UAV and environmental information (waypoint coordinates, obstacle distribution, etc.);

[0142] Action A: The next movement direction and step size of the UAV;

[0143] Reward function R: Encourage flying a distance close to the optimal path and covering a larger area, and punish deviating from the optimal path and flying into the obstacle area.

[0144] Introduce deep reinforcement learning algorithms:

[0145] Use reinforcement learning algorithms such as DDPG or PPO to optimize path control.

[0146] Objective:

[0147]

[0148] Where U* is the optimal policy; θ is the discount factor, usually in the range of [0,1], Is the reward value received by the UAV from the environment after each action t, u is the action policy of the UAV, and T is the total number of actions of the UAV.

[0149] 5. Simulation training:

[0150] Integrate the LSTM and reinforcement learning models into the simulation environment, and the UAV optimizes its flight strategy through continuous iterative training.

[0151] Monitoring metrics include: average path error, coverage ratio.

[0152] 6. Actual site training and validation:

[0153] Deploy the model trained in the simulation environment to the actual UAV control system, and use the UAV to obtain real-time position information (such as GPS and IMU data).

[0154] Adjust the model in combination with the actual site data, including the differences between the actual path and the predicted path, and flight performance indicators (such as time consumption, coverage rate).

[0155] Test the efficiency and accuracy of the drone in completing the inspection task and compare it with the manually planned path.

[0156] Finally, based on the optimal inspection path planned by the planning module, the training module guides the drone to learn the optimal flight strategy after multiple iterative trainings in the simulated environment and the actual photovoltaic panel site.

[0157] The method of this embodiment realizes the automatic inspection of the photovoltaic panel site, only requiring a small number of personnel for maintenance and supervision, greatly reducing the labor cost. At the same time, the optimized flight path and intelligent flight control reduce the unnecessary losses of the drone during flight, reduce the damage to the drone equipment caused by operations such as frequent takeoff and landing, sharp turns, etc., thereby extending the service life of the drone and reducing the equipment replacement and maintenance costs.

[0158] In this embodiment, the improved ant colony algorithm is used to plan the optimal inspection path for the photovoltaic panel site. Compared with the traditional manual planning or simple algorithm planning, it can effectively reduce the flight distance and time of the drone, enabling the drone to complete the inspection task with the shortest path, thereby greatly improving the inspection efficiency; the path prediction and control model constructed by using the long short-term neural network (LSTM) can perform in-depth learning training based on the existing inspection path data, enabling the drone to more accurately predict and control the flight path during flight, reducing the flight deviation caused by human operation or environmental interference, and further improving the inspection speed and accuracy; introducing a reinforcement learning mechanism to optimize the LSTM model can continuously adjust and improve the flight strategy according to the real-time feedback during the inspection process, enabling the drone to make more reasonable decisions when facing different environmental conditions and emergencies, further improving the inspection quality, and ensuring the timely discovery of faults and problems of the photovoltaic panels.

[0159] This embodiment also provides an intelligent inspection system for a photovoltaic panel site, as Figure 1 , including:

[0160] Data acquisition module: used to acquire the site information of the photovoltaic power station to be inspected;

[0161] Scene construction module: used to construct a three-dimensional digital simulation scene of the photovoltaic power station to be inspected through the site information;

[0162] Path planning module: used to plan the optimal inspection path through the three-dimensional digital simulation scene;

[0163] Drone training module: used to construct a path prediction and control model of the drone based on the optimal inspection path, and optimize the flight strategy and path planning of the drone.

[0164] Specifically, first, the data acquisition module is used to collect the site information of the photovoltaic power station to be inspected, including: the surrounding environment information of the photovoltaic panel site, as well as the terrain and geographical information. Among them, the surrounding environment information of the photovoltaic panel site includes: the distribution of vegetation (such as trees, grasslands, etc.), the coverage of shadows (specifically referring to the areas where the light is dim or missing due to the occlusion of objects or terrain), and obstacles (such as buildings, fences, poles, etc.). The vegetation situation can help understand the natural occlusion degree of the site, while the shadow situation is related to the lighting conditions of the photovoltaic panels at different times of the day. The existence of obstacles may affect the layout and overall efficiency of the photovoltaic panels;

[0165] The terrain and geographical information of the photovoltaic panel site includes: the height information of the terrain, the ground flatness, and the undulation information of the terrain. Considering the flight height and flight speed of the drone during the inspection process, understanding the slope and undulation of the site can avoid unnecessary shaking or collision risks during the takeoff, landing, or flight of the drone; at the same time, the height information of the terrain can help evaluate the drainage situation of the site and possible soil erosion problems. The ground flatness is directly related to the convenience and stability of the installation of the photovoltaic panels, and the undulation condition of the terrain has an important impact on the layout and design of the photovoltaic panels.

[0166] After that, the scene construction module constructs a simulated scene of the photovoltaic power station to be inspected based on the site information. The specific process includes: using the building information model technology (BIM) based on digital modeling to perform three-dimensional modeling of the photovoltaic power station. By importing the above parameters into the BIM software, three-dimensional modeling is carried out to obtain a three-dimensional simulated scene.

[0167] After the above process is completed, the path planning module plans the inspection path based on the three-dimensional simulated scene. Taking the generated simulated scene as the input, the improved ant colony algorithm is used to search for the optimal solution in the inspection path space by simulating the optimization process in nature, so as to obtain an optimal inspection path.

[0168] In this embodiment, the evaluation indicators are: the shortest path length and the maximum coverage range. After the fitness function is established, the planning module immediately generates an initial inspection path. Subsequently, it is evaluated according to the fitness function, and the fitness values of each path are calculated. The higher the fitness value, the better the optimization degree of the path.

[0169] In this embodiment, the improved ant colony algorithm is used to plan the optimal path, and the path with the highest fitness function value in all iterations is returned as the optimal inspection path.

[0170] Finally, the drone training module uses the optimal inspection path obtained by the above algorithm to further train the drone to learn better flight strategies, specifically as follows:

[0171] Record and log the optimal inspection path planned by the planning module, and perform normalization preprocessing on the recorded data; use Long Short-Term Memory (LSTM) neural networks to process and predict the flight path of the drone, and construct a path prediction and control model for the drone; use the preprocessed data to perform deep learning training on the LSTM model, and further introduce a reinforcement learning mechanism to optimize the performance of the model; by defining a reward function, give corresponding rewards or punishments according to the performance of the drone during flight, and guide the drone to learn the optimal flight strategy; and perform multiple iterative trainings on the drone through the constructed three-dimensional simulation environment to improve its adaptability to complex environments; after certain results are obtained from the three-dimensional simulation environment training, transfer the drone to the actual site for further training and verification, and continuously optimize the flight strategy and path planning of the drone through multiple iterative trainings in the actual site, and finally achieve an efficient and accurate photovoltaic panel site inspection task.

[0172] This embodiment also provides an electronic device, in which computer instructions are stored, and when the computer instructions are executed by a processor, the intelligent inspection method for a photovoltaic panel site is implemented.

[0173] Specifically, as Figure 3 shown, it is a schematic diagram of the physical structure of the electronic device according to this embodiment. The electronic device includes a processor, a memory, and a communication bus. Among them, the processor and the memory complete mutual communication through the communication bus. The processor can call the logical instructions in the memory to execute the intelligent inspection method for a photovoltaic panel site.

[0174] An executable code is stored on the memory. When the executable code is processed by the processor, it can enable the processor to execute some or all of the methods described above.

[0175] In addition, the method according to this embodiment can also be implemented as a computer program or a computer program product, which includes computer program code instructions for executing some or all of the above steps of this application.

[0176] The above is only a preferred specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by this application should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. An intelligent inspection method for a photovoltaic panel site, characterized in that, Including: Collect the site information of the photovoltaic power station to be inspected, where the site information includes the surrounding environment information and the topographic and geographical information; Based on the site information, construct a three-dimensional digital simulation scene corresponding to the photovoltaic power station to be inspected; Based on the three-dimensional digital simulation scene, plan the optimal inspection path; Through the optimal inspection path, use the long short-term neural network LSTM to construct a path prediction model for the drone, and introduce a reinforcement learning mechanism to iteratively train the path prediction model of the drone to optimize the flight strategy and path planning ability of the drone.

2. The intelligent inspection method for a photovoltaic panel site according to claim 1, wherein The surrounding environment information includes the distribution of vegetation, the coverage of shadows, and the positions of obstacles; the topographic and geographical information includes the terrain height, the ground flatness, and the terrain undulation.

3. The intelligent inspection method for photovoltaic panel sites according to claim 1, characterized in that Constructing a three-dimensional digital simulation scene corresponding to the photovoltaic power station to be inspected includes: Based on the building information model technology BIM for digital modeling, import the site information into the BIM software for three-dimensional modeling to obtain the three-dimensional digital simulation scene.

4. The intelligent inspection method for a photovoltaic panel site according to claim 1, wherein Planning the optimal inspection path includes: Use the improved ant colony algorithm to search for the optimal solution in the inspection path space by simulating the optimization process in nature, and use the shortest path length and the largest coverage range as evaluation indicators to obtain the optimal inspection path; Among them, the improved ant colony algorithm optimizes path search by dynamically adjusting the heuristic factor and the pheromone update strategy.

5. The intelligent inspection method for a photovoltaic panel site according to claim 4, wherein, The improved ant colony algorithm includes: Initialize the inspection point set and the pheromone matrix; Generate an initial path through the fitness function that comprehensively considers the path length and the coverage range; Dynamically adjust the heuristic factor and the pheromone evaporation coefficient to enhance the global search ability; Introduce a local search optimization strategy to correct the path, and output the inspection path with the highest fitness value, that is, the optimal inspection path.

6. The intelligent inspection method for a photovoltaic panel site according to claim 5, characterized in that The fitness function is: where D is the total path length; λ is the coverage range; are all weight parameters; h is the fitness function.

7. The intelligent inspection method for a photovoltaic panel site according to claim 1, wherein Iteratively training the path prediction model of the drone includes: Normalize the optimal inspection path data; Use the LSTM network to extract temporal features and predict the flight coordinates of the drone; Combine the reward function of reinforcement learning to optimize the flight strategy to obtain the trained path prediction model of the drone, where the reward function is dynamically adjusted based on path deviation, coverage completion, and obstacle avoidance.

8. The intelligent inspection method for a photovoltaic panel site according to claim 7, characterized in that, The reinforcement learning uses the deep deterministic policy gradient DDPG algorithm, and gradually optimizes the flight control parameters of the drone through alternating training of the simulation environment and the actual site.

9. An intelligent inspection system for a photovoltaic panel site, characterized in that, Including: Data acquisition module: used to collect the site information of the photovoltaic power station to be inspected; Scene construction module: used to construct a three-dimensional digital simulation scene of the photovoltaic power station to be inspected through the site information; Path planning module: used to plan the optimal inspection path through the three-dimensional digital simulation scene; Drone training module: used to construct a path prediction and control model for the drone based on the optimal inspection path, and optimize the flight strategy and path planning of the drone.

10. An electronic device, characterized in that, The electronic device stores computer instructions, and when the computer instructions are executed by the processor, the intelligent inspection method for the photovoltaic panel site as described in any one of claims 1 to 8 is implemented.

Images