Photovoltaic cleaner road planning method and device

Abstract

The invention relates to a photovoltaic cleaner road planning method and device, and belongs to the technical field of photovoltaic cleaners, and the method comprises the steps: dividing a photovoltaic power station into a plurality of to-be-cleaned photovoltaic regions, and determining the cleaning priority of each to-be-cleaned photovoltaic region according to the pollution degree and power generation efficiency of a photovoltaic panel of the photovoltaic power station; and taking the plurality of photovoltaic areas to be cleaned as vertexes, taking the Manhattan distance between the photovoltaic areas to be cleaned as an edge, taking the cleaning priority of each photovoltaic area to be cleaned as a limiting condition, and adopting a preset multi-target evaluation function to dynamically correct an initial cleaning path obtained based on a genetic algorithm to obtain a shortest cleaning path. The technical problem that in the prior art, an intelligent cleaner is difficult to accurately position and navigate, so that repeated cleaning or cleaning omission easily occurs in a photovoltaic panel cleaning route, and the cleaning efficiency is low is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic washers, and in particular to a photovoltaic washer road planning method, device, photovoltaic washer control equipment and medium. Background Art

[0002] Photovoltaic power generation is increasingly becoming a key technology for reducing carbon emissions. However, during the operation of photovoltaic power stations, dust gradually accumulates on the surface of solar panels. This accumulation of dust can significantly affect the electrical, optical, and thermal performance of photovoltaic panels and cause energy loss. Data indicates that the power output degradation of modules caused by dust deposition is more severe in different regions, ranging from 7% to 13%. Furthermore, the accumulation of dust can produce a "hot spot effect," whereby the uncleaned corners of the panels transform from power generation units into power consumption units. Shaded photovoltaic cells become non-generating load resistors, consuming the power generated by the connected cells and causing heat generation. This process accelerates panel aging, reduces the photoresistance conversion rate, and in severe cases, can cause fires.

[0003] Currently, the main cleaning methods for solar panels still rely on manual cleaning or semi-automatic means. However, manual cleaning is inefficient, especially in large power plants, with high cleaning costs and serious water waste. Semi-automatic cleaning usually involves manual operation of heavy equipment to clean solar panels. Although this method is highly efficient, it is complex to operate and has very strict requirements on the terrain, making it difficult to apply to rooftop photovoltaic arrays or high-density photovoltaic arrays. Therefore, it is particularly important to develop a deep neural network learning method that can automatically identify heavy pollutants on the surface of solar panels and guide the cleaning intelligent cleaner to perform precise decontamination. Summary of the Invention

[0004] In view of this, it is necessary to provide a photovoltaic cleaner road planning method, device, photovoltaic cleaner control equipment and medium to solve the technical problem in the existing technology that the intelligent cleaner is difficult to accurately locate and navigate, which leads to repeated cleaning or cleaning omissions in the photovoltaic panel cleaning route, resulting in low cleaning efficiency.

[0005] In order to solve the above problems, in a first aspect, the present invention provides a photovoltaic cleaning device road planning method, comprising: Divide the photovoltaic power station into multiple photovoltaic areas to be cleaned, and determine the cleaning priority of each photovoltaic area to be cleaned based on the pollution level and power generation efficiency of the photovoltaic panels of the photovoltaic power station; With the multiple photovoltaic areas to be cleaned as vertices, the Manhattan distances between the photovoltaic areas to be cleaned as edges, and the cleaning priorities of the photovoltaic areas to be cleaned as constraints, a preset multi-objective evaluation function is used to dynamically correct the initial cleaning path obtained based on the genetic algorithm to obtain the shortest cleaning path.

[0006] In one possible implementation, the initial cleaning path obtained based on the genetic algorithm is dynamically corrected using a preset multi-objective evaluation function with the multiple photovoltaic areas to be cleaned as vertices, the Manhattan distances between the photovoltaic areas to be cleaned as edges, and the cleaning priority of the photovoltaic areas to be cleaned as constraints to obtain the shortest cleaning path, including: The vertex coordinates of each photovoltaic area to be cleaned are converted into binary numbers with a preset fixed number of bits, and are spliced to form chromosomes in the order of access; Starting from the starting point of the access sequence, the adjacent accessible photovoltaic panel areas to be cleaned are screened based on the octree space partitioning, and the selection probability is calculated according to the binary sequence using a preset boot function to determine the initial population; The initial cleaning order is determined by taking the derivative of the total path length as the fitness function and the cleaning priority of each photovoltaic area to be cleaned as the penalty function of the fitness function; The optimal detour plan is selected based on the path length, number of turns, and cleaning priority of the photovoltaic area to be cleaned to determine the shortest cleaning path.

[0007] In a possible implementation, after determining the shortest cleaning path of the photovoltaic cleaner, the method further includes: Based on the shortest cleaning path, a real-time photovoltaic cleaning device road image is acquired; The real-time photovoltaic washer road image is input into a well-trained deep neural network model to obtain the real-time moving path of the photovoltaic washer.

[0008] In a possible implementation, after obtaining the real-time moving path of the photovoltaic cleaning device, the method further includes: Acquire real-time road environment images; A well-trained road environment recognition model including a MobileNetv2 network is used to extract features from real-time road environment images to determine photovoltaic panel boundary information and dirt locations. The real-time moving path is optimized using the photovoltaic panel boundary information and the dirt location as optimization conditions to obtain an optimized moving path.

[0009] In one possible implementation, the road environment recognition model includes a MobileNetv2 network, an ASPP module, a first convolutional layer, and a feature fusion layer; The MobileNetv2 network is used to extract edge and texture features of real-time road environment images; The ASPP module includes a second convolutional layer, a first dilated convolutional layer, a second dilated convolutional layer, a third dilated convolutional layer, a pooling layer, and a third convolutional layer; the ASPP module is used to expand the receptive field of the edge and texture features through the multi-scale dilated convolutional layer and output the initial recognition result; The feature fusion layer is used to fuse the initial recognition result with the edge and texture features to determine the photovoltaic panel boundary information and the dirt location.

[0010] In a possible implementation, the optimizing the real-time moving path based on the photovoltaic panel boundary information and the dirt location as optimization conditions to obtain the optimized moving path includes: According to the photovoltaic panel boundary information, the distance between the photovoltaic washer and each boundary of the photovoltaic panel is obtained, and the minimum boundary distance closest to the photovoltaic washer is determined; Controlling the photovoltaic washer to move to the nearest photovoltaic panel boundary and travel at a preset distance from the photovoltaic panel boundary, wherein the preset distance is less than a minimum boundary distance; determining whether the photovoltaic cleaner is located at a corner of the photovoltaic panel based on the relationship between the photovoltaic panel boundary information and the position of the photovoltaic cleaner; If so, the photovoltaic cleaning device is controlled to move toward the dirt location.

[0011] In a possible implementation, the method further includes: Get the type of pollutants in photovoltaic panels; The cleaning mode of the photovoltaic cleaner is adjusted according to the type of contaminants, wherein the cleaning mode includes wet cleaning and dry cleaning.

[0012] In a possible implementation, the method further includes: Get the real-time power of photovoltaic cleaning equipment; Comparing the real-time power level with the power level threshold; If the real-time power level is less than the power threshold, a low power alarm is issued.

[0013] In a second aspect, the present invention further provides a photovoltaic cleaning device road planning device, comprising: A partitioning module is used to divide the photovoltaic power station into multiple photovoltaic areas to be cleaned, and determine the cleaning priority of each photovoltaic area to be cleaned according to the pollution degree and power generation efficiency of the photovoltaic panels of the photovoltaic power station; The cleaning sequence determination module is used to determine the shortest cleaning path of the photovoltaic cleaner using an improved genetic algorithm, with the multiple photovoltaic areas to be cleaned as vertices, the Manhattan distances between the photovoltaic areas to be cleaned as edges, and the cleaning priority of each photovoltaic area to be cleaned as a constraint condition.

[0014] In a third aspect, the present invention further provides a photovoltaic washer control device, comprising: a processor and a memory; The memory stores a computer-readable program executable by the processor; When the processor executes the computer-readable program, the steps in the photovoltaic washer path planning method described above are implemented.

[0015] In a fourth aspect, the present invention further provides a computer-readable storage medium storing one or more programs, which can be executed by one or more processors to implement the steps in the photovoltaic washer road planning method as described above.

[0016] The beneficial effects of the present invention are as follows: first, a photovoltaic power station is divided into multiple photovoltaic areas to be cleaned, and the cleaning priority of each photovoltaic area to be cleaned is determined according to the pollution degree and power generation efficiency of the photovoltaic panels of the photovoltaic power station, thereby enhancing the intelligent cleaner's ability to understand the environment and the response speed; then, with the multiple photovoltaic areas to be cleaned as vertices and the Manhattan distances between the photovoltaic areas to be cleaned as edges, the cleaning order problem of the photovoltaic power station is converted into a traveling salesman problem, and with the cleaning priority of each photovoltaic area to be cleaned as a constraint condition, with the multiple photovoltaic areas to be cleaned as vertices, the Manhattan distances between the photovoltaic areas to be cleaned as edges, and the cleaning priority of each photovoltaic area to be cleaned as a constraint condition, a preset multi-objective evaluation function is used to dynamically correct the initial cleaning path obtained based on the genetic algorithm to obtain the shortest cleaning path; through the multi-objective optimization solution of the genetic algorithm, not only the probability of generating the optimal cleaning order is improved, but also the convergence process of the algorithm is accelerated, and the robustness of the algorithm is enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A method flow chart of an embodiment of a photovoltaic cleaning device road planning method provided by the present invention; Figure 2 for Figure 1 In step S101, the Hamilton diagram of the photovoltaic power station after the area to be cleaned is divided; Figure 3 Schematic diagram of the changing trend of o with T solved by the adaptive genetic algorithm in step S102; Figure 4 This is a method flow chart of an embodiment of step S102 in an embodiment of the present invention; Figure 5 Schematic diagram of solar photovoltaic panel boundary line image processing in an embodiment of the present invention; Figure 6 Schematic diagram of an embodiment of a photovoltaic cleaning device road planning device provided by the present invention; Figure 7 This is a schematic diagram of the operating environment of an embodiment of the photovoltaic washer control device provided by the present invention. DETAILED DESCRIPTION

[0018] The preferred embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, and are not used to limit the scope of the present invention.

[0019] A specific embodiment of the present invention discloses a photovoltaic cleaning device road planning method, please refer to Figure 1 ,include: S101, dividing the photovoltaic power station into a plurality of photovoltaic areas to be cleaned, and determining a cleaning priority of each photovoltaic area to be cleaned according to the degree of contamination and power generation efficiency of the photovoltaic panels of the photovoltaic power station; It is understandable that the severity of pollution affects the power generation efficiency, that is, the higher the pollution level, the lower the power generation efficiency. Therefore, the areas with the most serious pollution and the lowest power generation efficiency have the highest cleaning priority, and the areas with the lightest pollution and higher power generation efficiency have the relatively lowest priority. Figure 2 As shown, by zoning the photovoltaic areas to be cleaned according to priority, the intelligent cleaner's ability to understand the environment and its response speed can be enhanced.

[0020] In a specific embodiment, x and y are set as the horizontal and vertical coordinates of the geometric center of each region polygon, respectively. Each point represents a region and is identified by a number. The priority of the region is divided into 6 levels, with a larger number indicating a higher priority.

[0021] S102, using the multiple photovoltaic areas to be cleaned as vertices, the Manhattan distances between the photovoltaic areas to be cleaned as edges, and the cleaning priorities of the photovoltaic areas to be cleaned as constraints, dynamically correcting the initial cleaning path obtained based on the genetic algorithm using a preset multi-objective evaluation function to obtain the shortest cleaning path.

[0022] It's important to note that after the PV power plant is divided into regions, the cleaning problem essentially transforms into the traveling salesman problem (TSP). This problem requires a cleaner to start from a starting point, visit each region in sequence, and traverse it only once, before returning to the starting point. The TSP is a classic problem in graph theory and can be solved by constructing a Hamiltonian graph. Each region is considered a vertex, and the distances between regions are considered edges. This constructs a Hamiltonian graph G = (V, E), where V represents the vertex set, numbered from 1 to n, and E represents the edge set, numbered from 1 to m. Since two regions are often not directly reachable in reality, the Manhattan distance is used to calculate the distance between vertices, thereby forming the edge set. Hierarchical planning for PV power plant cleaning tasks essentially involves finding the shortest Hamiltonian circuit.

[0023] Furthermore, the initial cleaning path obtained by the genetic algorithm is dynamically modified by presetting a multi-objective evaluation function to obtain the shortest cleaning path, and the fitness function is dynamically adjusted linearly to ensure the effective selection ability of the genetic algorithm, thereby increasing the probability of generating the optimal solution, accelerating the convergence process of the algorithm, and enhancing the robustness of the algorithm. Figure 3 As shown in Figure 2, IGA can quickly and accurately plan the operating sequence of the washers in the photovoltaic power plant. This hierarchical task planning strategy lays a solid foundation for the subsequent road recognition strategy of the washers in actual operation.

[0024] In this embodiment, a photovoltaic power station is first divided into multiple photovoltaic areas to be cleaned, and the cleaning priority of each photovoltaic area to be cleaned is determined based on the pollution level and power generation efficiency of the photovoltaic panels of the photovoltaic power station, thereby enhancing the intelligent cleaner's ability to understand the environment and the response speed. Subsequently, the cleaning order problem of the photovoltaic power station is converted into a traveling salesman problem with the multiple photovoltaic areas to be cleaned as vertices and the Manhattan distances between the photovoltaic areas to be cleaned as edges. The cleaning priority of each photovoltaic area to be cleaned is used as a constraint condition, and an improved genetic algorithm is used to determine the shortest cleaning path of the photovoltaic cleaner. The multi-objective optimization solution of the genetic algorithm not only improves the probability of generating the optimal cleaning order, but also accelerates the convergence process of the algorithm and enhances the robustness of the algorithm.

[0025] In a preferred embodiment, see Figure 4 , step S102 includes: S401, converting the vertex coordinates of each photovoltaic area to be cleaned into a binary number with a preset fixed number of bits, and splicing them into a chromosome in the order of access; In this embodiment, when using binary encoding, the coordinates of the photovoltaic areas to be cleaned are first connected in order of priority to form a path, such as ab, cd, ..., mn. Each photovoltaic panel coordinate in the path is then represented as a binary number, forming a chromosome, thus completing the chromosome encoding. The number of binary digits used to represent the location coordinates is based on the maximum number of binary digits required for the endpoint coordinates in the current environmental model; all other coordinates are represented using this binary digit.

[0026] S402, starting from the starting point of the access sequence, screening adjacent accessible photovoltaic panel areas to be cleaned based on the octree space partitioning, and using a preset boot function to calculate the selection probability according to the binary sequence to determine the initial population; It should be noted that the population is initialized using a combination of an octree and a guidance function. Starting from the starting PV panel, the octree principle is used to select the free PV panels adjacent to the starting PV panel as the next path PV panel. When multiple free PV panels are available, the value of the guidance function must be calculated. The guidance function is shown in Equation (X). After calculating the guidance factors for all free PV panels, a gambling strategy is used to select the PV panels. In this strategy, once a PV panel is selected on a path, it is marked and will not be selected again in subsequent path planning to avoid loops. The selection operation uses a combination of a roulette wheel and an elite retention strategy. The crossover operation uses a single-point crossover operator. The mutation operation uses a basic bit mutation operator. The correction operation is used to judge and repair individuals.

[0027]

[0028] The value of H(xay·) represents the guidance factor of (xayb) of the free photovoltaic panel, and its value reflects the probability of the free photovoltaic panel being selected; (xMyN) is the coordinate of the target photovoltaic panel.

[0029] S403, using the derivative of the total path length as a fitness function and the cleaning priority of each photovoltaic area to be cleaned as a penalty function of the fitness function to determine an initial cleaning order; It should be noted that the fitness function uses the inverse of the total path length as the fitness function, denoted as F, as shown in Equation (1). For path planning, the shorter the total path length, the better, that is, the larger F, the better.

[0030]

[0031] Where i represents the path number, and St is the Euclidean distance between time t and time t-1.

[0032] The objective function aims to minimize the sum of the distances between regions, meaning that smaller distances represent better performance. However, the fitness function requires larger values, so an appropriate conversion is necessary. Directly using the inverse of the distance as the fitness function may result in very small relative differences between fitness values, making the probability of individuals being selected during the selection process uniform, which would weaken the genetic algorithm's selection mechanism. Therefore, in this embodiment, a selection operator is designed by combining tournament selection with roulette wheel selection. The tournament selection method involves randomly selecting a preset number of individuals from the parent population and retaining the individuals with the highest fitness to pass on to the offspring. This process is repeated until the number of offspring meets a preset termination condition, ensuring that approximately S / N (rounded up) high-quality individuals are passed on. In each round of the tournament, a roulette wheel selection rule is used to select a parent from the tournament size for the subsequent crossover operation, while another parent is randomly selected in the same round. This method ensures both the transmission of elite individuals and the quality of the parents participating in the crossover operation.

[0033] Based on the operator selection process described above, a crossover operator based on a segmentation strategy is employed. Specifically, when the length of the route to be solved exceeds a preset threshold Q, sequential crossover is employed; otherwise, self-crossover is employed. The threshold Q is set based on analysis and optimization of historical algorithm performance. The introduction of the self-crossover operator and the adjustment of the threshold Q for specific scenarios aim to increase the probability of generating high-quality individuals during the crossover operation, accelerate algorithm convergence, and enhance the overall stability of the algorithm.

[0034] If a path violates the priority (e.g., a low-priority area is visited before a high-priority area), a penalty is imposed on the fitness value (e.g., multiplied by a decay coefficient), forcing the algorithm to eliminate the violating path.

[0035] S404: Select an optimal detour plan based on the path length, the number of turns, and the cleaning priority of the photovoltaic area to be cleaned to determine the shortest cleaning path.

[0036] In this embodiment, in the genetic algorithm design for cleaning robot path planning, a fourth-stage dynamic path correction operator is added on the basis of the traditional three-stage genetic operations of selection, crossover, and mutation, forming a "3+1" evolutionary mechanism, which effectively solves the technical bottleneck of traditional algorithms that easily generate invalid paths in complex scenarios.

[0037] Among them, the correction operators include: obstacle perception layer: real-time detection of conflicting nodes in the chromosome encoding path, and establishment of a topological map of obstacle photovoltaic panels; neighborhood reconstruction layer: using an eight-direction expansion search method to construct a feasible path fragment library within the 3×3 photovoltaic panels around the conflict point; optimal replacement layer: based on multi-objective evaluation functions such as path length, number of turns, and safety margin, intelligently select the optimal detour plan.

[0038] In some embodiments of the present invention, after determining the shortest cleaning path of the photovoltaic cleaner, the method further includes: Based on the shortest cleaning path, a real-time photovoltaic cleaning device road image is acquired; The real-time photovoltaic washer road image is input into a well-trained deep neural network model to obtain the real-time moving path of the photovoltaic washer.

[0039] In this embodiment, the road conditions are predicted by a deep neural network model, which can quickly identify and classify the roads for cleaning, automatically plan the cleaning routes, and accurately predict the future driving direction, thereby improving the efficiency of the cleaning work.

[0040] It should be noted that the structure of the deep neural network model is not limited here.

[0041] Furthermore, in the process of training the deep neural network model using the photovoltaic washer road images, the data augmentation method is used to expand the original data set. This strategy not only increases the volume of image data, but also greatly improves the generalization ability of the model, ensuring that it still performs well when faced with new data. First, a large number of road scene images are collected from the photovoltaic power station site to construct the initial data set. In order to further enhance the diversity of the data set and improve the robustness of model training, a series of image enhancement methods such as rotation, brightness and contrast adjustment are used to expand the size of the data set, forming the following: Figure 5 The enhanced dataset shown in Figure 1 enables more accurate and efficient recognition of the road environment surrounding the photovoltaic power station. Furthermore, given the specific input image size requirements of deep neural network learning models, after constructing the enhanced dataset, all images were normalized, scaling each image to 512x512 pixels. This step not only meets the model's input requirements but also accelerates training, allowing the gradient descent algorithm to converge to the optimal state more quickly, thereby comprehensively improving the efficiency and quality of model training.

[0042] Since solar panels have a certain tilt angle and a height difference from the ground, the solar panel cleaner faces the risk of falling during operation. In some embodiments of the present invention, after obtaining the real-time moving path of the photovoltaic cleaner, the following steps are further included: Acquire real-time road environment images; A well-trained road environment recognition model including a MobileNetv2 network is used to extract features from real-time road environment images to determine photovoltaic panel boundary information and dirt locations. The real-time moving path is optimized using the photovoltaic panel boundary information and the dirt location as optimization conditions to obtain an optimized moving path.

[0043] In this example, the ASPP architecture in the DeepLabv3+ base model cleverly utilizes multiple dilated convolutional layers to extract features, effectively expanding the receptive field of the output feature map. Dilated convolution (also known as dilated convolution) has the advantage of increasing the receptive field of features, allowing the feature layer to maintain a high pixel resolution.

[0044] It should be noted that, despite the complex road environments surrounding photovoltaic power stations, which include challenges such as overgrown weeds and gravel, blurred boundaries, and uneven road surfaces, the DeepLabv3+ model can accurately segment road areas in most cases. However, to address issues such as missed or incorrect road segmentation, inaccurate boundary segmentation, and lengthy training times due to large model parameters, this embodiment replaces the original model's backbone network, Xception, with an optimized MobileNetv2 network. This effectively reduces model parameters, accelerates model execution, and improves image data processing efficiency. Secondly, the ASPP (Atrous Spatial Pyramid Pooling) architecture is improved, combining different receptive field fusion with a dilated depthwise separable convolution strategy to enhance the correlation between information in different receptive fields and improve information utilization in the dilated convolutional layers, thereby improving model training efficiency. Finally, a Convolutional Block Attention Module (CBAM) is incorporated into the encoder. The CBAM module retains more critical image boundary feature information, improving feature extraction accuracy and enabling the model to more accurately capture road boundary details during road recognition.

[0045] In a specific embodiment, the road environment recognition model includes a MobileNetv2 network, an ASPP module, a first convolutional layer, and a feature fusion layer; The MobileNetv2 network is used to extract edge and texture features of real-time road environment images; The ASPP module includes a second convolutional layer, a first dilated convolutional layer, a second dilated convolutional layer, a third dilated convolutional layer, a pooling layer, and a third convolutional layer; the ASPP module is used to expand the receptive field of the edge and texture features through the multi-scale dilated convolutional layer and output the initial recognition result; The feature fusion layer is used to fuse the initial recognition result with the edge and texture features to determine the photovoltaic panel boundary information and the dirt location.

[0046] In some embodiments of the present invention, the optimizing the real-time moving path based on the photovoltaic panel boundary information and the dirt location as optimization conditions to obtain the optimized moving path includes: According to the photovoltaic panel boundary information, the distance between the photovoltaic washer and each boundary of the photovoltaic panel is obtained, and the minimum boundary distance closest to the photovoltaic washer is determined; Controlling the photovoltaic washer to move to the nearest photovoltaic panel boundary and travel at a preset distance from the photovoltaic panel boundary, wherein the preset distance is less than a minimum boundary distance; determining whether the photovoltaic cleaner is located at a corner of the photovoltaic panel based on the relationship between the photovoltaic panel boundary information and the position of the photovoltaic cleaner; If so, the photovoltaic cleaning device is controlled to move toward the dirt location.

[0047] In this embodiment, after the solar panel assembly is laminated and encapsulated, it is often encapsulated with an aluminum alloy border to protect and support the assembly. When a high-definition camera captures an image of a solar panel containing an aluminum alloy border, a deep neural network learning object detection algorithm is used for recognition processing. The pixel distance from the intelligent cleaner to the border line is calculated. Through comparison and judgment, the closest border line to the intelligent cleaner is determined and the intelligent cleaner is controlled to move toward it. To eliminate noise in the captured image, Gaussian filtering is used; to facilitate border detection, grayscale processing is further performed. Sharpening is used to highlight the aluminum alloy border line. A threshold is set for image segmentation, and the Roberts operator is used for border detection. Subsequently, area filtering is performed, and areas with connected domains less than the threshold area are set as the background. Hough line detection is used to fit the border line to the boundary line, and column scanning is used to determine the center line of the two lines. The image is scanned again in columns to calculate the average pixel distance from the center line to the bottom of the image. The above process is used to process images from other directions of the HD camera to determine the direction of the current border line closest to the intelligent cleaner and control the intelligent cleaner to move in that direction.

[0048] When the washer's HD camera first detects the closest point to the solar panel's upper boundary, it directs the smart washer toward the nearest boundary, entering state ③. The smart washer then uses the HD camera to continue collecting image information from other directions to ensure it steadily advances along the solar panel's upper boundary toward the nearest boundary. If the left boundary is detected as the closest point, the smart washer rotates 90° to the left and moves toward the left boundary until the left, right, and right rear photoelectric sensors all detect the solar panel's boundary. At this point, the HD camera rotates to collect image information from more directions. When the HD camera rotates 180° clockwise and the average pixel distance from the boundary centerline to the bottom of the image is the shortest, the smart washer is determined to be at the upper left corner of the solar panel, which is state ④.

[0049] Furthermore, if the right boundary line is detected as the closest distance, the smart washer will move toward it until the left front, right front, and left rear photoelectric sensors simultaneously detect the boundary line. When the high-definition camera rotates 180° clockwise and detects the shortest distance to the center line of the boundary line, it can be determined that the smart washer is in the upper right corner, that is, state 5. When the smart washer reaches the intersection of the two boundary lines of the solar panel, that is, any corner of the upper half of the area, its left front and right front photoelectric sensors will simultaneously detect the aluminum alloy boundary line of the solar panel. At this time, combined with the signals from the left and right rear of the smart washer and the rotation angle of the high-definition camera, the corner point position of the smart washer can be accurately determined, and then it can be guided to move to the starting position of the cleaning operation.

[0050] After both the left and right front photoelectric sensors detect the solar panel boundary and simultaneously receive a signal from the left rear photoelectric sensor, if the HD camera rotates 90° or 270° clockwise to capture the nearest boundary line, the smart washer is determined to be in the upper left corner. If it rotates 180° to capture information, it is determined to be in the upper right corner. Similarly, if a signal is received from the right rear photoelectric sensor, the same determination is made. Based on the corner point determination, the smart washer then selects a pre-set route and moves to its starting position, preparing to begin cleaning operations.

[0051] In some embodiments of the present invention, further comprising: Get the type of pollutants in photovoltaic panels; The cleaning mode of the photovoltaic cleaner is adjusted according to the type of contaminants, wherein the cleaning mode includes wet cleaning and dry cleaning.

[0052] In this embodiment, solar panels are exposed to various types of contamination due to prolonged exposure to outdoor environments. Different types of dirt require different cleaning strengths and methods, and scratching the solar panel surface must be avoided, as this will affect power generation efficiency. To address this issue, cleaning control instructions are generated for the intelligent cleaner based on the type of contaminant detected by the high-definition camera. This controls the intelligent cleaner's switching between wet and dry cleaning modes. During operation, the intelligent cleaner continuously acquires environmental perception data and updates the model in real time. In some embodiments of the present invention, further comprising: Get the real-time power of photovoltaic cleaning equipment; Comparing the real-time power level with the power level threshold; If the real-time power level is less than the power threshold, a low power alarm is issued.

[0053] In this embodiment, a power monitoring and early warning system is established by acquiring the real-time power consumption of the photovoltaic scrubber. This system tracks the power status of the intelligent scrubber in real time, ensuring that the intelligent scrubber promptly detects low power and prepares to return. This system also provides feedback on the remaining operating time, enabling timely route optimization and adjustments. Simultaneously, the intelligent scrubber's position is dynamically adjusted to ensure a smooth return. By adjusting the heading angle, the system maintains the accuracy of the intelligent scrubber's direction, thereby ensuring precise cleaning operations. By integrating route position and operating direction adjustment information, a multi-channel coordinated control signal is generated that synchronously regulates the position and direction of the intelligent scrubber, ensuring that the intelligent scrubber accurately executes the planned route and maintains the correct cleaning status through real-time adjustments.

[0054] Based on the above photovoltaic cleaning device road planning method, the embodiment of the present invention also provides a photovoltaic cleaning device road planning device, please refer to Figure 6 ,include: A partitioning module 610 is used to divide the photovoltaic power station into a plurality of photovoltaic areas to be cleaned, and determine the cleaning priority of each photovoltaic area to be cleaned according to the pollution degree and power generation efficiency of the photovoltaic panels of the photovoltaic power station; The cleaning sequence determination module 620 is used to dynamically correct the initial cleaning path obtained based on the genetic algorithm using a preset multi-objective evaluation function, with the multiple photovoltaic areas to be cleaned as vertices, the Manhattan distances between the photovoltaic areas to be cleaned as edges, and the cleaning priorities of the photovoltaic areas to be cleaned as constraints, to obtain the shortest cleaning path.

[0055] like Figure 7 As shown, based on the above-mentioned photovoltaic washer road planning method, the present invention also provides a photovoltaic washer control device. The photovoltaic washer control device can be a computing photovoltaic washer control device such as a mobile terminal, desktop computer, notebook, PDA, or server. The photovoltaic washer control device includes a processor 710, a memory 720, and a display 730. Figure 7 Only some of the components of the photovoltaic washer control device are shown, but it should be understood that implementation of all of the shown components is not required, and more or fewer components may be implemented instead.

[0056] In some embodiments, the memory 720 may be an internal storage unit of the photovoltaic washer control device, such as a hard drive or memory. In other embodiments, the memory 720 may also be an external storage unit of the photovoltaic washer control device, such as a plug-in hard drive, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Furthermore, the memory 720 may include both the internal storage unit and the external storage unit of the photovoltaic washer control device. The memory 720 is used to store application software installed in the photovoltaic washer control device and various data, such as program code for installing the photovoltaic washer control device. The memory 720 may also be used to temporarily store data that has been output or is about to be output. In one embodiment, the memory 720 stores a photovoltaic washer route planning program 740, which can be executed by the processor 710 to implement the photovoltaic washer route planning method according to various embodiments of the present application.

[0057] In some embodiments, the processor 710 may be a central processing unit (CPU), a microprocessor, or other data processing chip, configured to execute program codes stored in the memory 720 or process data, such as executing a photovoltaic washer-based road planning method.

[0058] In some embodiments, display 730 can be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 730 is used to display information about the photovoltaic washer control device based on the photovoltaic washer road planning and to display a visual user interface. Components 710-730 of the photovoltaic washer control device communicate with each other via a system bus.

[0059] Those skilled in the art will appreciate that all or part of the process steps of the above-described embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, such as a magnetic disk, an optical disk, a read-only memory, or a random access memory.

[0060] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed by the present invention should be covered by the scope of protection of the present invention.

Claims

1. A photovoltaic cleaning device road planning method, characterized in that: include: Divide the photovoltaic power station into multiple photovoltaic areas to be cleaned, and determine the cleaning priority of each photovoltaic area to be cleaned based on the pollution level and power generation efficiency of the photovoltaic panels of the photovoltaic power station; With the multiple photovoltaic areas to be cleaned as vertices, the Manhattan distances between the photovoltaic areas to be cleaned as edges, and the cleaning priorities of the photovoltaic areas to be cleaned as constraints, a preset multi-objective evaluation function is used to dynamically correct the initial cleaning path obtained based on the genetic algorithm to obtain the shortest cleaning path.

2. The photovoltaic cleaning device road planning method according to claim 1, characterized in that: The method uses a preset multi-objective evaluation function to dynamically correct an initial cleaning path obtained based on a genetic algorithm using the multiple photovoltaic areas to be cleaned as vertices, the Manhattan distances between the photovoltaic areas to be cleaned as edges, and the cleaning priority of each photovoltaic area to be cleaned as a constraint condition to obtain a shortest cleaning path, including: The vertex coordinates of each photovoltaic area to be cleaned are converted into binary numbers with a preset fixed number of bits, and are spliced to form chromosomes in the order of access; Starting from the starting point of the access sequence, the adjacent accessible photovoltaic panel areas to be cleaned are screened based on the octree space partitioning, and the selection probability is calculated according to the binary sequence using a preset boot function to determine the initial population; The initial cleaning order is determined by taking the derivative of the total path length as the fitness function and the cleaning priority of each photovoltaic area to be cleaned as the penalty function of the fitness function; The optimal detour plan is selected based on the path length, number of turns, and cleaning priority of the photovoltaic area to be cleaned to determine the shortest cleaning path.

3. The photovoltaic cleaning device road planning method according to claim 1, characterized in that: After determining the shortest cleaning path for the photovoltaic cleaner, it also includes: Based on the shortest cleaning path, a real-time photovoltaic cleaning device road image is acquired; The real-time photovoltaic washer road image is input into a well-trained deep neural network model to obtain the real-time moving path of the photovoltaic washer.

4. The photovoltaic cleaning device road planning method according to claim 3, characterized in that: After obtaining the real-time moving path of the photovoltaic cleaner, it also includes: Acquire real-time road environment images; A well-trained road environment recognition model including a MobileNetv2 network is used to extract features from real-time road environment images to determine photovoltaic panel boundary information and dirt locations. The real-time moving path is optimized using the photovoltaic panel boundary information and the dirt location as optimization conditions to obtain an optimized moving path.

5. The photovoltaic cleaning device road planning method according to claim 4, characterized in that: The road environment recognition model includes a MobileNetv2 network, an ASPP module, a first convolutional layer, and a feature fusion layer; The MobileNetv2 network is used to extract edge and texture features of real-time road environment images; The ASPP module includes a second convolutional layer, a first dilated convolutional layer, a second dilated convolutional layer, a third dilated convolutional layer, a pooling layer, and a third convolutional layer; the ASPP module is used to expand the receptive field of the edge and texture features through the multi-scale dilated convolutional layer and output the initial recognition result; The feature fusion layer is used to fuse the initial recognition result with the edge and texture features to determine the photovoltaic panel boundary information and the dirt location.

6. The photovoltaic cleaning device road planning method according to claim 4, characterized in that: The method of optimizing the real-time moving path based on the photovoltaic panel boundary information and the dirt location as optimization conditions to obtain the optimized moving path includes: According to the photovoltaic panel boundary information, the distance between the photovoltaic washer and each boundary of the photovoltaic panel is obtained, and the minimum boundary distance closest to the photovoltaic washer is determined; Controlling the photovoltaic washer to move to the nearest photovoltaic panel boundary and travel at a preset distance from the photovoltaic panel boundary, wherein the preset distance is less than a minimum boundary distance; determining whether the photovoltaic cleaner is located at a corner of the photovoltaic panel based on the relationship between the photovoltaic panel boundary information and the position of the photovoltaic cleaner; If so, the photovoltaic cleaning device is controlled to move toward the dirt location.

7. The photovoltaic cleaning device road planning method according to claim 1, characterized in that: Also includes: Get the type of pollutants in photovoltaic panels; The cleaning mode of the photovoltaic cleaner is adjusted according to the type of contaminants, wherein the cleaning mode includes wet cleaning and dry cleaning.

8. The photovoltaic cleaning device road planning method according to claim 1, characterized in that: Also includes: Get the real-time power of photovoltaic cleaning equipment; Comparing the real-time power level with the power level threshold; If the real-time power level is less than the power threshold, a low power alarm is issued.

9. A photovoltaic cleaning device road planning device, characterized in that: include: A partitioning module is used to divide the photovoltaic power station into multiple photovoltaic areas to be cleaned, and determine the cleaning priority of each photovoltaic area to be cleaned according to the pollution degree and power generation efficiency of the photovoltaic panels of the photovoltaic power station; The cleaning sequence determination module is used to dynamically correct the initial cleaning path obtained based on the genetic algorithm using a preset multi-objective evaluation function, with the multiple photovoltaic areas to be cleaned as vertices, the Manhattan distances between the photovoltaic areas to be cleaned as edges, and the cleaning priority of each photovoltaic area to be cleaned as a constraint condition, to obtain the shortest cleaning path.

10. A photovoltaic cleaning device control device, characterized in that: include: processor and memory; The memory stores a computer-readable program executable by the processor; When the processor executes the computer-readable program, the steps of the photovoltaic washer path planning method according to any one of claims 1 to 8 are implemented.

Images