Intelligent cleaning control method and system for water surface photovoltaic module

Abstract

The application provides a smart cleaning control method and system for a water surface photovoltaic module, relates to the technical field of cleaning control, and comprises the following steps: reading power generation data of the water surface photovoltaic module and backtracking evaluation of pollution influence; activating a UAV to conduct patrol and build a photovoltaic panel image; after feature extraction is performed on the photovoltaic panel image, identifying a pollution type, a distribution area and a distribution intensity; determining a pollution grade, and using the pollution grade, the backtracking evaluation result and an environmental state to make a scheduling decision of a cleaning scheme; using the pollution type, the distribution area and the distribution intensity to make scheme prediction optimization of the cleaning scheme, and combining a scheduling decision result to generate a cleaning control scheme. Through the application, the technical problem that the maintenance efficiency of the photovoltaic module is affected due to the lack of accurate evaluation of the pollution degree and the lack of pertinence of the cleaning strategy is solved, the cleaning strategy is adjusted in real time through accurate pollution evaluation, and the maintenance efficiency and power generation performance of the photovoltaic module are improved.

Description

Technical Field

[0001] This application relates to the field of cleaning control technology, and in particular to a smart cleaning control method and system for water-surface photovoltaic modules. Background Technology

[0002] Floating photovoltaic (PV) modules are susceptible to the adhesion of contaminants such as dust, water stains, and algae, which can block sunlight, reduce the transmittance of the glass surface, and decrease light absorption efficiency. Therefore, the cleanliness of PV modules directly affects their power generation efficiency. Traditional PV cleaning methods often rely on regular or fixed-cycle manual inspections and cleaning, lacking specificity and flexibility, and prone to omissions. Especially in large-scale floating PV power plants, manual inspections cannot comprehensively and promptly identify contamination points and are easily affected by factors such as human fatigue and weather. In addition, many PV power plants have not accurately assessed the actual impact of contamination on power generation efficiency, and cleaning plans are often fixed, leading to over- or under-cleaning.

[0003] In summary, existing technologies suffer from a lack of accurate assessment of contamination levels, resulting in insufficiently targeted cleaning strategies that negatively impact the maintenance efficiency of photovoltaic modules. Summary of the Invention

[0004] The purpose of this application is to provide a smart cleaning control method and system for water-surface photovoltaic modules, in order to solve the technical problem in the prior art that the lack of accurate assessment of the degree of pollution leads to insufficient targeting of cleaning strategies, thereby affecting the maintenance efficiency of photovoltaic modules.

[0005] In view of the above problems, this application provides a smart cleaning control method and system for water-surface photovoltaic modules.

[0006] Firstly, this application provides a smart cleaning control method for surface photovoltaic (PV) modules. This method is implemented through a smart cleaning control system for surface PV modules. The method includes: reading power generation data from the surface PV modules; constructing a power generation efficiency perception based on the data reading results; conducting a pollution impact retrospective evaluation using the power generation efficiency perception; establishing a retrospective evaluation result; activating a drone to inspect the surface PV modules along a preset flight path and constructing a PV panel image; extracting features from the PV panel image and identifying the pollution type, distribution area, and distribution intensity; determining the pollution level based on the pollution type, distribution area, and distribution intensity; and making scheduling decisions for cleaning schemes using the pollution level, retrospective evaluation results, and environmental conditions, generating scheduling decision results; predicting and optimizing the cleaning scheme using the pollution type, distribution area, and distribution intensity; and generating a cleaning control scheme based on the prediction optimization results and scheduling decision results.

[0007] Optionally, the equipment parameters of the water surface photovoltaic module are obtained, and the factory power generation efficiency is configured using the equipment parameters; the usage time of the water surface photovoltaic module is read, and the attenuation fitting of the factory power generation efficiency is performed based on the usage time to establish the attenuation fitting result; the efficiency impact compensation of the power generation efficiency perception is performed using the attenuation fitting result, and a pollution impact retrospective evaluation is established based on the efficiency impact compensation result.

[0008] Optionally, a pollution sample is established, and array tests of photovoltaic panels are performed based on the pollution sample to establish a test set, which includes the efficiency impact test results. After mapping and identifying the pollution sample and the test set, a pollution impact retrospective evaluation network is constructed using the mapped and identified pollution sample and the test set. The pollution impact retrospective evaluation network includes a depositional light transmission impact retrospective sub-network and an accumulation temperature rise impact retrospective sub-network. The efficiency impact compensation results are input into the pollution impact retrospective evaluation network, and retrospective analysis of light transmission impact and temperature rise impact is performed through the depositional light transmission impact retrospective sub-network and the accumulation temperature rise impact retrospective sub-network, respectively, to generate depositional data. The pollution impact retrospective evaluation is completed based on the depositional data.

[0009] Optionally, after image distortion correction of the photovoltaic panel image, illumination equalization adjustment is performed, and color normalization is applied. A convolutional neural network is used to traverse and extract features from the adjusted photovoltaic panel image to establish a pollution feature map. The traversal extraction includes color feature extraction, texture feature extraction, boundary contour feature extraction, and spatial location extraction. After multidimensional feature extraction, semantic fusion of multidimensional features with the same spatial location is performed to identify the pollution type, distribution area, and distribution intensity.

[0010] Optionally, a joint scoring function is constructed, as follows:

[0011] J(Zone i )=ω1·PL i +ω2·RE i +ω3·f(ES i ); where J(Zone) i ) is a joint scoring function that characterizes the urgency of cleaning. PL i According to the pollution level, RE i To retrospectively assess the impact score, f(ES) i The environmental state adjustment function is constructed from the current air temperature, wind speed and direction disturbances, and rainfall forecast. ω1, ω2, and ω3 are the weight coefficients of the pollution level item, the retrospective impact item, and the environmental impact item, respectively. The urgency of the cleaning plan is evaluated using the joint scoring function, and the scheduling decision is made based on the evaluation results.

[0012] Optionally, a cleaning time interval is determined based on the scheduling decision result; it is determined whether there is a rainfall impact within the cleaning time interval; if there is no rainfall impact, the time node of the last cleaning is obtained, and a backtracking interval is established based on the time node; pollution prediction fitting is performed using the backtracking interval, the cleaning time interval, the pollution type, distribution area and distribution intensity, and the cleaning scheme is optimized based on the pollution prediction fitting result.

[0013] Optionally, if there is a rainfall impact, the pollution prediction fitting result is reconstructed after generating impact compensation based on the rainfall impact; the reconstructed pollution prediction fitting result is used to complete the optimization of the cleaning scheme.

[0014] Optionally, the pollution prediction fitting results are used to perform strategy matching for cleaning strategies, and strategy matching results are established; a multi-objective evaluation function is constructed, the evaluation indicators of which include post-cleaning efficiency recovery rate, cleaning energy consumption, and environmental disturbance risk indicators during execution; the multi-objective evaluation function is used to perform optimization matching of the strategy matching results, and optimization is completed.

[0015] Optionally, a power grid dispatch plan is obtained, and a plan avoidance factor is established based on the power grid dispatch plan; the plan avoidance factor is used to establish constraints for the dispatch decision results.

[0016] Secondly, this application also provides a smart cleaning control system for surface photovoltaic modules, used to execute the smart cleaning control method for surface photovoltaic modules as described in the first aspect. The smart cleaning control system for surface photovoltaic modules includes: an impact retrospective evaluation module, used to read power generation data from the surface photovoltaic modules, construct power generation efficiency perception using the power generation data reading results, perform pollution impact retrospective evaluation using the power generation efficiency perception, and establish retrospective evaluation results; a drone inspection module, used to activate a drone to inspect the surface photovoltaic modules according to a preset flight path and construct photovoltaic panel images; a feature extraction module, used to extract features from the photovoltaic panel images and identify pollution type, distribution area, and distribution intensity; a scheduling decision module, used to determine the pollution level based on the pollution type, distribution area, and distribution intensity, and make scheduling decisions for the cleaning scheme using the pollution level, retrospective evaluation results, and environmental conditions, generating scheduling decision results; and a scheme optimization module, used to predict and optimize the cleaning scheme using the pollution type, distribution area, and distribution intensity, and generate a cleaning control scheme based on the prediction optimization results and scheduling decision results.

[0017] One or more technical solutions provided in this application have at least the following beneficial effects:

[0018] By reading power generation data from the surface photovoltaic modules, a power generation efficiency perception is constructed using the data. This perception is then used to conduct a retrospective evaluation of pollution impact, establishing the evaluation results. A drone is activated to patrol the surface photovoltaic modules along a preset flight path, constructing images of the photovoltaic panels. Feature extraction is performed on the photovoltaic panel images to identify the pollution type, distribution area, and intensity. Based on the pollution type, distribution area, and intensity, the pollution level is determined. The pollution level, retrospective evaluation results, and environmental conditions are used to make scheduling decisions for cleaning schemes, generating scheduling decision results. The cleaning scheme is then predicted and optimized using the pollution type, distribution area, and intensity. A cleaning control scheme is generated based on the prediction and optimization results and the scheduling decision results. In other words, by analyzing changes in power generation efficiency, a retrospective assessment of pollution impact is conducted. Simultaneously, drones are used for inspections to determine pollution levels. Based on the retrospective assessment results and environmental conditions, scheduling decisions are made. Pollution predictions are performed based on pollution type, distribution area, and intensity. Cleaning plans are then optimized to generate cleaning control schemes. This enables accurate identification and assessment of pollution, and real-time optimization of cleaning plans, making them more flexible and improving the maintenance efficiency and power generation performance of the water-surface photovoltaic modules.

[0019] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this application 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 merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating the intelligent cleaning control method for water-surface photovoltaic modules according to this application.

[0022] Figure 2 This is a schematic diagram of the intelligent cleaning control system for water-surface photovoltaic modules according to this application.

[0023] Figure labeling: Impact backtracking evaluation module 11, UAV inspection module 12, feature extraction module 13, scheduling decision module 14, and scheme optimization module 15. Detailed Implementation

[0024] This application provides a smart cleaning control method and system for floating photovoltaic modules, solving the technical problem in existing technologies where the lack of accurate assessment of pollution levels leads to insufficiently targeted cleaning strategies, thus affecting the maintenance efficiency of photovoltaic modules. By analyzing changes in power generation efficiency, the method performs retrospective evaluation of pollution impact. Simultaneously, drone inspections are conducted to determine pollution levels. Scheduling decisions are made based on the retrospective evaluation results and environmental conditions. Pollution prediction is performed based on pollution type, distribution area, and intensity, optimizing the cleaning plan and generating a cleaning control scheme. This achieves accurate identification and assessment of pollution, and real-time optimization of the cleaning plan, making the cleaning scheme more flexible and improving the maintenance efficiency and power generation performance of floating photovoltaic modules.

[0025] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0026] Example 1, please refer to the appendix. Figure 1 This application provides a smart cleaning control method for surface photovoltaic modules, wherein the smart cleaning control method for surface photovoltaic modules is executed by a smart cleaning control system for surface photovoltaic modules, and the smart cleaning control method for surface photovoltaic modules specifically includes the following steps:

[0027] S100: Read the power generation data of the photovoltaic modules on the water surface, construct a power generation efficiency perception based on the power generation data reading results, conduct a retrospective evaluation of pollution impact based on the power generation efficiency perception, and establish a retrospective evaluation result.

[0028] Furthermore, this application S100 includes:

[0029] Obtain the equipment parameters of the water surface photovoltaic module, and configure the factory power generation efficiency using the equipment parameters; read the usage time of the water surface photovoltaic module, and perform attenuation fitting of the factory power generation efficiency based on the usage time to establish attenuation fitting results; use the attenuation fitting results to compensate for the efficiency impact of the power generation efficiency perception, and establish a pollution impact retrospective evaluation based on the efficiency impact compensation results.

[0030] A pollution sample is established, and array tests of photovoltaic panels are conducted based on the pollution sample to establish a test set, which includes the efficiency impact test results. After mapping and identifying the pollution sample and the test set, a pollution impact retrospective evaluation network is constructed using the mapped and identified pollution sample and test set. The pollution impact retrospective evaluation network includes a depositional light transmission impact retrospective network and an accumulation temperature rise impact retrospective network. The efficiency impact compensation results are input into the pollution impact retrospective evaluation network, and retrospective analysis of the light transmission impact and temperature rise impact is performed through the depositional light transmission impact retrospective network and the accumulation temperature rise impact retrospective network, respectively, to generate depositional data. The pollution impact retrospective evaluation is completed based on the depositional data.

[0031] Specifically, sensors and intelligent monitoring components read power generation data from the surface photovoltaic modules. The intelligent monitoring components are responsible for automatically collecting, processing, and transmitting data in the full-cycle intelligent cleaning terminal for the surface photovoltaic modules. All sensor data is centrally processed by a PLC and then transmitted to the control center. Since sensor data may be affected by noise or interference, the PLC uses algorithms to clean and filter the data to ensure the accuracy and reliability of the final data. A PLC is a digital computer used for industrial automation control, centrally processing various data from sensors and intelligent monitoring components, and performing operations such as analysis, storage, scheduling, and transmission. The full-cycle intelligent cleaning terminal can execute cleaning operations when needed based on monitoring and analysis results. The control center is responsible for centralized management, control, and coordination, receiving sensor data, performing comprehensive analysis, and finally outputting control commands to the full-cycle intelligent cleaning terminal to ensure the efficient operation of the photovoltaic modules.

[0032] By reading the power generation data, a power generation efficiency perception system is constructed. Through analysis of this data, the current power generation efficiency of the floating photovoltaic (PV) modules is inferred. The current power generation efficiency is calculated by comparing the actual power generation with the theoretical maximum power generation. The theoretical maximum power generation is the maximum output power calculated based on the design parameters of the floating PV modules and actual environmental conditions (such as light intensity and temperature).

[0033] Obtain the equipment parameters of the floating photovoltaic (PV) modules. These parameters are usually provided by the manufacturer and specified on the equipment label or technical manual. They include the manufacturer's specifications, technical parameters, and performance under standard test conditions, such as maximum power output, operating voltage, current, and power generation efficiency. Based on these parameters, configure the factory power generation efficiency, which is the power generation efficiency of the floating PV modules under standard test conditions, i.e., the ideal power generation efficiency.

[0034] The usage duration of surface-mount photovoltaic (PV) modules is obtained, which refers to the cumulative operating time since installation, typically measured in hours or years. The performance of PV modules changes with usage time, especially after prolonged exposure to the environment, where efficiency may decline. The degradation of PV module efficiency is fitted based on its usage duration. Generally, the efficiency degradation of PV modules follows certain patterns, such as exponential or linear degradation. Based on the historical power generation efficiency of the surface-mount PV modules, i.e., the power generation efficiency at each point in time, a degradation trend is fitted to predict the decline in power generation efficiency over time. The degradation trend is usually fitted by analyzing long-term operating data and describes the pattern of PV module efficiency change over time.

[0035] Historical power generation efficiency of floating photovoltaic (PV) modules is typically recorded daily, monthly, quarterly, or even annually, reflecting the actual power generation efficiency over different time periods. Time series analysis methods (such as time series models) are used to observe the trend of power generation efficiency over time and determine whether there are significant seasonal variations. For power generation efficiency data with large fluctuations, data smoothing (such as moving averages) may be necessary to reduce noise interference with the fitting results. The power generation efficiency of PV modules typically declines exponentially with increasing usage time. A decay trend graph is plotted based on historical power generation efficiency and corresponding usage duration, and a decay formula is fitted accordingly. The usage duration of the floating PV module is input into the decay formula to obtain the current decay fitting result, thus determining the power generation efficiency that should be achieved under the current usage period—the theoretically expected power generation efficiency of the floating PV module at this time.

[0036] Based on the attenuation fitting results, efficiency impact compensation is applied to the actually measured power generation efficiency. This distinguishes the degradation effect from the actual power generation efficiency under current environmental conditions, eliminating the impact of natural degradation and accurately reflecting whether the power generation efficiency of the water-based photovoltaic modules has decreased due to pollution. The efficiency impact compensation result eliminates the influence of degradation factors on power generation efficiency assessment, making pollution assessment more accurate.

[0037] Data was collected from samples under different pollution conditions. The pollution samples were constructed by collecting photovoltaic panel performance data (such as the degree of efficiency reduction and the degree of surface pollution of the photovoltaic panels) under different pollution conditions. Test points were selected in different areas of the photovoltaic modules on the water surface, covering a variety of pollution conditions, such as normal cleanliness, light pollution, moderate pollution, and heavy pollution. The pollution type of different pollution samples was recorded, such as whether it was dust, sludge, algae, etc., and the degree of pollution.

[0038] Array tests of photovoltaic panels were conducted based on contaminated samples to establish a test set, which included test results on the impact of different levels and types of contamination on the efficiency of the photovoltaic panels. After obtaining the contaminated sample and test set data, the data was classified through mapping labels, that is, the contaminated samples and test sets were mapped to determine the impact of different contamination conditions on power generation efficiency. The goal of the mapping labels is to label each contaminated sample and its corresponding test set data, indicating the type and degree of contamination, as well as the associated power generation efficiency loss.

[0039] Using mapped and identified contamination samples and test sets as input, a contamination impact retrospective evaluation network is constructed. This network comprises two subnetworks: a deposition-induced light transmittance impact retrospective network and an accumulation-induced temperature rise impact retrospective network. The deposition-induced light transmittance impact retrospective network is used to analyze how contaminant deposition affects the light transmittance of the photovoltaic module surface. By inputting contamination samples, it obtains the light transmittance reduction under different contaminant deposition conditions, calculates the impact of contaminants on the light transmittance of the photovoltaic module, and thus obtains the power generation efficiency loss caused by each contaminant.

[0040] The design of the retrospective network for assessing the impact of sedimentation on light transmittance consists of an input layer, hidden layers, and an output layer, with the output being the percentage change in transmittance. The network is trained using an existing test dataset, and its parameters are optimized to accurately predict the impact of different pollutant types and levels on the transmittance of photovoltaic modules. A subset of the dataset is used for validation to check whether the transmittance change output by the network matches the actual test results. For example, with a sample inputting dust contamination, the retrospective network output should show a 10% transmittance loss.

[0041] The Accumulation Temperature Rise Impact Retrospective Network, by inputting contamination samples, derives the effect of contaminant accumulation on the surface temperature rise of photovoltaic (PV) modules, and thus the loss of PV module power generation efficiency due to contaminants. The design of the Accumulation Temperature Rise Impact Retrospective Network includes an input layer, a hidden layer, and an output layer, outputting the temperature changes of the PV module under different contamination conditions. The network is trained using a test dataset to optimize its parameters, enabling it to accurately predict the impact of contaminant accumulation on temperature rise. A training method similar to that of the Deposition Transmittance Impact Retrospective Network is adopted, but it primarily focuses on the relationship between temperature change and contaminant accumulation. A subset of the dataset is used for validation to ensure that the temperature rise changes output by the network match actual test results. Integrating the retrospective analysis results of the Deposition Transmittance Impact Retrospective Network and the Accumulation Temperature Rise Impact Retrospective Network reveals the power generation efficiency loss of PV modules under different contamination conditions.

[0042] The efficiency impact compensation results are input into the constructed pollution impact retrospective evaluation network. The core task of the network is to further analyze these compensation results and perform precise retrospective analysis. Based on the efficiency impact compensation results, the decrease in light transmittance and the impact of temperature rise are identified, and these are input into two sub-networks for retrospective analysis to determine the deposition data. This data specifically includes the amount, type, and affected area of ​​pollutants, as well as the impact of these pollutants on the light transmittance and temperature rise of photovoltaic modules. Based on the deposition data generated from the retrospective analysis, the final retrospective evaluation of the impact of pollutants on the power generation efficiency of photovoltaic modules is completed, determining the specific impact of current pollution on the power generation efficiency of photovoltaic modules.

[0043] By combining power generation data, attenuation fitting, pollution samples, and retrospective analysis, the impact of pollution on photovoltaic modules can be accurately assessed, avoiding misjudgment and over-cleaning. By inputting efficiency impact compensation results and combining them with the analysis of the pollution impact retrospective evaluation network, specific sedimentation data can be obtained, accurately quantifying the current impact of pollution on surface photovoltaic modules.

[0044] S200: Activate the drone to inspect the photovoltaic modules on the water surface according to the preset flight path and construct images of the photovoltaic panels.

[0045] S300: After extracting features from the photovoltaic panel image, identify the pollution type, distribution area, and distribution intensity.

[0046] Furthermore, this application S300 includes:

[0047] After image distortion correction of the photovoltaic panel image, illumination equalization adjustment is performed and color normalization is applied. A convolutional neural network is used to traverse and extract pollution features from the adjusted photovoltaic panel image to establish a pollution feature map. The traversal extraction includes color feature extraction, texture feature extraction, boundary contour feature extraction, and spatial location extraction. After multidimensional feature extraction, semantic fusion of multidimensional features with the same spatial location is performed to identify the pollution type, distribution area, and distribution intensity.

[0048] Specifically, the drone is activated and follows a preset flight path to inspect the photovoltaic modules on the water surface, capturing images of the photovoltaic panels using a high-definition camera. The preset flight path refers to the drone's planned flight route or trajectory, typically planned before flight via GPS and customized according to mission requirements, taking into account multiple factors such as the layout of the photovoltaic array, the size of the water surface, weather conditions, and inspection frequency. The flight path usually covers the entire area of ​​the photovoltaic array to ensure a comprehensive inspection.

[0049] The drone is activated and begins its pre-defined mission. Following a predetermined flight path, it captures images of the photovoltaic modules. Sensors on the drone (such as high-resolution cameras and infrared sensors) collect image data from the photovoltaic panels. After completing the inspection, the drone transmits the captured image data back to the control center via wireless communication.

[0050] At the control center, the images are analyzed using image processing software. Image distortion correction is performed on the photovoltaic panel images, specifically correcting geometric distortions in the images. Due to the optical characteristics of camera lenses, distortions such as barrel or pincushion distortion can occur at the edges of the image. Camera calibration techniques are used to correct the image by obtaining the camera's intrinsic parameters (such as focal length, optical center coordinates, and distortion coefficients). If the camera used by the drone has barrel distortion, the photovoltaic panel image will appear distorted at the four corners. By obtaining the camera's distortion coefficients and correcting the image, the edges of the image will be restored to straight lines.

[0051] Uneven lighting can occur due to factors such as shooting angle, time, and ambient light, resulting in some areas being too bright or too dark. To obtain a more balanced image, illumination equalization (or histogram equalization) is used to adjust the brightness distribution of the image, making the bright and dark areas more even and ensuring overall image visibility and detail. Areas with excessive brightness differences are adjusted to ensure more balanced image brightness. Finally, image normalization is performed, scaling the pixel values ​​to a standard range (e.g., [0,1]) to reduce differences under different lighting and shooting conditions, resulting in a more consistent image.

[0052] The adjusted photovoltaic panel image is traversed and extracted using a CNN, which automatically learns and extracts multi-level features from the image, including color feature extraction, texture feature extraction, boundary contour feature extraction, and spatial location extraction. Color feature extraction identifies the distribution of colors in the image, especially color changes caused by pollution. For example, dust or scale may cause the photovoltaic panel to appear darker or yellowish. Texture features capture the roughness and patterns of the photovoltaic panel surface. Pollutants (such as sewage or dust) often form different textures on the photovoltaic panel surface, making texture features an important basis for pollution identification. Boundary contour features help identify the shape and boundaries of pollutants. Pollutants usually form some obvious contours, and the CNN can extract these features through techniques such as edge detection. Spatial location features refer to the specific location of pollutants in the image. By performing convolution operations on the image, the CNN can extract the spatial distribution of pollutants in the image.

[0053] After multidimensional feature extraction, the next step is feature fusion. Since different features (such as color, texture, boundary, and location) may correspond to different regions in image space, these features need to be fused to obtain a more accurate determination of the type, distribution area, and intensity of pollution. Features extracted from various dimensions (color, texture, etc.) are merged into a unified spatial framework, ensuring that features at each location are accurately associated. For example, if the color and texture features of a region both show signs of pollution, then that region is determined to be a polluted region. Semantic fusion combines information from different features to form a complete semantic understanding. During the fusion process, the interrelationships and influences between features are considered to form a more complete representation of pollution features. For example, by fusing color, texture, and boundary features, a comprehensive understanding of the type of pollution (such as dust, water stains, oil stains, etc.), as well as the distribution area and intensity of pollution, can be obtained. Based on the fused features, a pollution feature map is built to visually display the type, distribution, and intensity of pollutants.

[0054] By analyzing the fused features, specific types of contaminants are identified, such as dust (dull color and rough texture), water stains (darkened color and smooth texture), and algae (irregular boundaries and color variations). Using boundary contours and spatial location information, the distribution area of ​​contaminants on the photovoltaic panel is precisely determined, separating contaminated areas from clean areas. The distribution area of ​​contamination is obtained by calculating the ratio of the pixel area of ​​the contaminated area to the total image area. For example, if dust covers 60% of the photovoltaic panel, the contamination distribution area is 0.6. The contamination intensity should be assessed based on the type of contamination, the physical properties of the contaminants (such as adhesion and concentration), the nature of the contamination (such as dust, oil stains, and water stains), and its impact on the photovoltaic module.

[0055] By utilizing drone patrols and image processing, the workload of manual inspections can be significantly reduced, and the types and distribution of pollution can be automatically identified, thus improving patrol efficiency. Image processing accurately identifies the type, distribution area, and intensity of pollution, enabling accurate cleaning decisions.

[0056] S400: Based on the pollution type, distribution area, and distribution intensity, determine the pollution level, and use the pollution level, retrospective evaluation results, and environmental status to make scheduling decisions for the cleaning plan, generating scheduling decision results.

[0057] Furthermore, this application S400 includes:

[0058] Construct a joint scoring function, which is as follows:

[0059] J(Zone i )=ω1·PL i +ω2·RE i +ω3·f(ES i); where J(Zone) i ) is a joint scoring function that characterizes the urgency of cleaning. PL i According to the pollution level, RE i To retrospectively assess the impact score, f(ES) i The environmental state adjustment function is constructed from the current air temperature, wind speed and direction disturbances, and rainfall forecast. ω1, ω2, and ω3 are the weight coefficients of the pollution level item, the retrospective impact item, and the environmental impact item, respectively. The urgency of the cleaning plan is evaluated using the joint scoring function, and the scheduling decision is made based on the evaluation results.

[0060] Specifically, different pollutants have different effects on photovoltaic (PV) modules. For example, dust mainly affects light transmittance, while water stains and algae may alter the reflectivity of the module surface, and oil stains directly affect the electrical performance of PV cells. The impact on PV power generation efficiency also varies depending on the type of pollution. The contaminated area refers to the distribution range of pollutants on the PV panel; the size of the contaminated area directly affects the pollution level assessment. If the contaminated area covers most of the PV panel, it indicates severe pollution, which may lead to a large-scale reduction in power generation efficiency. If the contaminated area occupies only a small area, the pollution is relatively mild, and the impact is smaller. Pollution intensity refers to the impact of pollutants on the performance of PV modules and can be quantified by the thickness and density of the pollution.

[0061] Based on a comprehensive assessment of the type, area, and intensity of pollution, pollution levels can be categorized into several distinct levels. For example, if the pollution type is dust, and the polluted area covers less than 20% of the total area, with relatively low pollution intensity (both thickness and density), it is classified as light pollution, having minimal impact on the photovoltaic panels and requiring only periodic cleaning. However, if the pollution type is dust + water stains + algae, and the polluted area covers approximately 30% of the total area, but the pollution intensity is high (both thickness and density), it will affect the heat dissipation of the photovoltaic modules on the water surface, generating excessive heat that could potentially damage the modules. Therefore, this is classified as heavy pollution, impacting power generation efficiency and requiring immediate cleaning.

[0062] The pollution level is converted into a corresponding numerical value. The higher the pollution level value, the more severe the pollution and the greater the urgency of cleaning. The retrospective impact score is a score obtained by quantifying the retrospective evaluation results, determined based on the long-term impact of pollutant deposition on photovoltaic panel performance. The environmental state adjustment function is a dynamic adjustment value calculated based on external environmental conditions (such as temperature, wind speed, wind direction, and rainfall forecast) to assess the accumulation of pollution and the effectiveness of cleaning. High temperatures may accelerate the deposition and fixation of pollutants (such as dust and stains) on the photovoltaic panel surface, while also affecting the performance of the photovoltaic panel and reducing its power generation efficiency. Wind speed affects the propagation and deposition of pollutants; strong winds cause more dust and pollutants to fall, while lower wind speeds may cause pollutants to remain on the photovoltaic panel. Wind direction affects the direction of pollutant propagation; when wind direction has a greater impact on pollutants, the urgency of cleaning is higher. When it rains, a small amount of calcium and magnesium ions dissolve in the rain and adhere to the glass surface of the photovoltaic module. If not cleaned in time, a thick and hard calcium-magnesium film will form on the surface of the photovoltaic module over time. Once the film forms and accumulates, it is extremely difficult to remove, which will seriously affect the power generation performance of photovoltaic modules and may even lead to problems such as hot spots on photovoltaic panels.

[0063] Based on the degree of influence of each factor on cleaning, weight coefficients are assigned to different environmental factors. For example, rainfall may have a greater impact than wind speed, so its weight coefficient may be higher. These weight coefficients are optimized using recent historical data (such as the cleaning effect of photovoltaic modules under different weather conditions) to ensure that the environmental state adjustment function reflects the urgency of cleaning. Finally, based on the weight coefficients of each environmental factor, the environmental state adjustment function f(ES) is constructed as follows. i The environmental state adjustment function is calculated as follows: ) = a·T + b·W + c·D, where T is the air temperature, W is the wind speed and direction, D is the rainfall forecast, and a, b, and c are the weighting coefficients of the three factors. For example, if the current temperature is 35℃, its air temperature urgency is 0.7; the wind speed is 5m / s, and the wind direction causes pollution accumulation, so its wind speed and wind force interference urgency is 0.6; the predicted rainfall probability is 0.8, so its rainfall urgency is 0.8. With preset weighting coefficients of 0.3 for air temperature, 0.2 for wind speed, and 0.5 for rainfall, the value of the environmental state adjustment function is 0.73.

[0064] Using pollution levels, retrospective assessment results, and environmental conditions, a joint scoring function is employed to evaluate the urgency of cleaning the current surface photovoltaic modules. The joint scoring function is as follows:

[0065] J(Zone i )=ω1·PL i +ω2·RE i +ω3·f(ES i ); where J(Zone)i ) is a joint scoring function that characterizes the urgency of cleaning. It combines the outputs of pollution level, retrospective impact score, and environmental state adjustment function to give a cleaning urgency score; PL i According to the pollution level, RE i To retrospectively assess the impact score, f(ES) i The environmental condition adjustment function is constructed from the current air temperature, wind speed and direction disturbances, and rainfall forecast. ω1, ω2, and ω3 are the weighting coefficients for the pollution level, retrospective impact, and environmental impact items, respectively. Generally, when the pollution level, pollution retrospective score, and environmental condition all indicate that the current pollution is relatively severe, the urgency of cleaning is high, indicating that immediate cleaning is required, as pollution prevention has a greater impact on power generation efficiency.

[0066] Based on the urgency results of the joint scoring, the cleaning demand level is determined, and scheduling decisions are made, including whether to clean, when to clean, and the selection of coverage areas. In other words, the priority of cleaning is determined based on the urgency assessment results obtained from the joint scoring function. Cleaning demand levels can be divided into multiple levels, such as high urgency, low urgency, and no cleaning required. Based on the urgency assessment and cleaning demand level, a decision is made on whether to execute the cleaning task, and a specific cleaning plan is formulated. Specifically, based on the scoring results, it is determined whether a cleaning task needs to be executed; the timing of cleaning is assessed to determine the optimal time for cleaning to achieve maximum efficiency; and the cleaning area is determined based on the distribution of contamination to reduce unnecessary cleaning work. If the contamination distribution is uneven, the most concentrated area is usually selected for cleaning first. Contamination areas are accurately identified and cleaned using drone images, generating a specific scheduling plan.

[0067] By constructing a joint scoring function that comprehensively considers multiple factors such as pollution level, retrospective evaluation results, and environmental status, the urgency of photovoltaic panel cleaning can be assessed. This helps to optimize the allocation of cleaning resources, improve cleaning efficiency, reduce maintenance costs, and ensure the power generation efficiency of photovoltaic modules.

[0068] Furthermore, this application also includes the following steps:

[0069] Obtain the power grid dispatch plan, establish a plan avoidance factor based on the power grid dispatch plan, and use the plan avoidance factor to establish constraints for the dispatch decision results.

[0070] Specifically, this involves obtaining the power grid dispatch plan, a power supply scheme formulated by the power grid dispatch center based on factors such as real-time load demand, generator operation status, and transmission network conditions. A planning avoidance factor is then established based on the power grid dispatch plan to prevent conflicts between cleaning activities and the dispatch plan, ensuring that the cleaning operation time does not overlap with critical periods in the plan. For example, if the power grid dispatch plan indicates high power demand during a certain period, large-scale cleaning activities may need to be avoided during that period to minimize the impact on power grid operation.

[0071] By establishing constraints on dispatch decisions based on planning avoidance factors, the sequence and duration of cleaning operations are optimized to adapt to the requirements of the power grid dispatch plan. In other words, a cleaning activity schedule is developed to ensure that the cleaning activities effectively maintain the photovoltaic modules without adversely affecting grid operation. By establishing planning avoidance factors and constraining dispatch decisions, the coordination between cleaning activities and the power grid dispatch plan is ensured, reducing the impact of cleaning activities on grid operation and contributing to improving the overall operational efficiency of photovoltaic power plants.

[0072] S500: Utilizes pollution type, distribution area, and distribution intensity to predict and optimize cleaning schemes, and generates cleaning control schemes based on the prediction and optimization results and scheduling decision results.

[0073] Furthermore, this application S500 includes:

[0074] The cleaning time interval is determined based on the scheduling decision result; it is determined whether there is a rainfall impact within the cleaning time interval; if there is no rainfall impact, the time node of the last cleaning is obtained, and a backtracking interval is established based on the time node; pollution prediction fitting is performed using the backtracking interval, the cleaning time interval, the pollution type, distribution area and distribution intensity, and the cleaning scheme is optimized based on the pollution prediction fitting result.

[0075] Specifically, based on the aforementioned scheduling decision results, a cleaning time interval is determined, which is the set time span between two cleaning operations. For example, if pollution accumulates rapidly in a certain area, the cleaning time interval may be shorter. It is then determined whether rainfall will affect the area within the cleaning time interval. If no rainfall occurs, the time point of the last cleaning is obtained, and the time difference from the last cleaning to the current time is calculated, i.e., the backtracking interval. Using factors such as the backtracking interval, cleaning time interval, pollution type, distribution area, and distribution intensity, a pollution prediction model is fitted. In other words, based on historical cleaning records and the current pollution distribution, a pollution prediction model is constructed to predict future pollution trends.

[0076] Collect cleaning records from the past few months or years, including cleaning dates, methods, areas, and types of contamination. Each record can include the contamination status of the photovoltaic panels at the time of cleaning, as well as an evaluation of the cleaning effect. Obtain the current distribution of contamination through drone inspections and power generation data reading, specifically including contamination type (such as dust, bird droppings, water stains, etc.), contamination intensity (such as contamination coverage area and concentration), and contamination distribution area (such as the central area and edge areas of the photovoltaic panels).

[0077] The pollution situation before and after cleaning, seasonal changes, and the correlation between climatic factors (such as temperature, rainfall, and wind speed) and pollution are extracted from historical cleaning records. A pollution prediction model is constructed based on the pollution type, distribution area, and intensity determined by the current pollution distribution. The pollution prediction model is trained using a training dataset (the dataset constructed from historical cleaning records), which is divided into training and test sets, typically 80% training and 20% test. A random forest model is trained using the training set. Taking pollution intensity prediction as an example, the model inputs are environmental features (temperature, wind speed, etc.), pollution features (pollution type, region), and historical pollution data; the target output is the pollution trend over a future period. During training, hyperparameters (such as the number of trees, maximum depth, etc.) are adjusted to optimize model performance. Optimal model parameters are selected through cross-validation. The model's predictive performance is evaluated using the test set, and the mean squared error is calculated. Convergence conditions are set for the model, such as a validation set loss change of less than 0.01 for five consecutive rounds or a training set accuracy reaching 95%. When the model reaches the convergence condition, training stops, and the model at this point is used as the pollution prediction model to obtain the future pollution trend.

[0078] The pollution type, distribution area, and intensity of the current pollution distribution are input into the pollution prediction model to obtain the future pollution trend, i.e., the pollution prediction fitting result. Based on the pollution prediction result and the cleaning time interval, multiple cleaning plans are matched, each of which may include different cleaning strategies and time nodes, such as cleaning strategies, cleaning area selection, cleaning methods, and cleaning timing. The best-performing plan among the multiple cleaning plans is selected as the prediction optimization result.

[0079] Furthermore, this application also includes the following steps:

[0080] If rainfall has an impact, the pollution prediction fitting results are reconstructed after generating impact compensation based on the rainfall impact; the reconstructed pollution prediction fitting results are then used to optimize the cleaning scheme.

[0081] Specifically, if rainfall is a factor, its impact is analyzed, and compensation is generated. Since dust contains a large amount of calcium and magnesium oxides, when rainfall occurs at the photovoltaic (PV) site, small amounts of calcium and magnesium ions dissolve in the rain and adhere to the glass surface of the PV modules. If not cleaned promptly, a thick and hard calcium-magnesium film will form on the PV module surface over time. Once formed and accumulated, this film is extremely difficult to remove, severely impacting the power generation efficiency of the PV modules and potentially causing problems such as hot spots on the PV panels.

[0082] To eliminate the interference of rainfall on pollution prediction, it is necessary to adjust the pollution prediction results through rainfall impact compensation. This involves analyzing the amount and distribution of pollutants caused by rainfall to correct the original pollution prediction results. Based on rainfall characteristics (such as rainfall amount and duration), a rainfall impact compensation factor is constructed to adjust the pollution prediction results. This compensation factor is then applied to the original pollution prediction results for corresponding adjustments. For example, rainfall can cause changes in the distribution area, intensity, and type of pollution in the prediction; the pollution intensity might increase from 30% to 36%, and the polluted area from 40% to 44%. Through rainfall compensation, the specific distribution of pollution is recalculated. By reconstructing the pollution prediction fitting results, a more accurate pollution distribution is ultimately obtained. For example, after rainfall, some areas may change from dust pollution to water stains or calcium-magnesium film pollution; the pollution intensity in some areas may increase by 20%; and the coverage area of ​​the polluted area may increase, potentially covering more areas.

[0083] Based on changes in pollution intensity, contaminated area, and environmental factors, several appropriate cleaning strategies are selected. Each cleaning strategy needs to be evaluated, taking into account factors such as post-cleaning efficiency recovery rate, cleaning energy consumption, and environmental disturbance risk indicators during execution. The cleaning solution with the highest overall score is selected as the final cleaning strategy.

[0084] Furthermore, this application also includes the following steps:

[0085] The pollution prediction fitting results are used to match cleaning strategies, and the strategy matching results are established. A multi-objective evaluation function is constructed, and the evaluation indicators of the multi-objective evaluation function include post-cleaning efficiency recovery rate, cleaning energy consumption, and environmental disturbance risk indicators during execution. The multi-objective evaluation function is used to optimize the strategy matching results, and the optimization is completed.

[0086] Specifically, based on the pollution prediction fitting results (i.e., pollution type, pollution distribution area, pollution intensity, etc.), multiple cleaning schemes matching the current pollution situation are selected to obtain strategy matching results. In the multi-objective optimization process, the effectiveness of the cleaning strategy needs to be evaluated based on multiple evaluation indicators, including post-cleaning efficiency recovery rate, cleaning energy consumption, and environmental disturbance risk indicators during execution. The post-cleaning efficiency recovery rate indicates the degree to which the power generation efficiency of the photovoltaic modules recovers after cleaning, assessing the efficiency at which the photovoltaic modules' power generation capacity recovers to its optimal level after the cleaning operation. Cleaning energy consumption is the energy consumed to complete the cleaning operation. High cleaning energy consumption means that the cleaning process consumes a large amount of equipment and energy resources, requiring reasonable optimization of energy consumption. Environmental disturbance risk indicators during execution refer to environmental disturbances that may occur during the cleaning process (such as wind speed, humidity changes, secondary pollution from pollutants, etc.). Environmental disturbances generated during the cleaning process need to be minimized to avoid adverse effects on the environment and photovoltaic modules.

[0087] The efficiency recovery rate, energy consumption, and environmental disturbance risk after cleaning are synthesized into a comprehensive objective function, which comprehensively considers the weights of different objectives. Using a particle swarm optimization algorithm based on a multi-objective evaluation function, the matching results of cleaning strategies are optimized to select an optimal cleaning scheme. The positions and velocities of particles are initialized. Particle positions represent various parameters of the cleaning scheme, such as cleaning frequency, cleaning method, and contamination type. Each particle's position is mapped to a specific cleaning scheme; particle velocity represents the particle's speed of movement in the solution space, controlling the convergence speed towards the optimal solution. Each particle's scheme is evaluated according to the multi-objective evaluation function, and then the particle's velocity and position are updated based on this score. The particle's position and velocity are updated based on its individual optimal position and the swarm's optimal position. Optimization stops when a termination condition is met, and the cleaning scheme at this point is taken as the predicted optimization result. The termination condition includes reaching a predetermined maximum number of iterations or a small change in the swarm's optimal solution.

[0088] The predictive optimization result is the optimal cleaning scheme based on pollution prediction fitting and multi-objective evaluation function optimization. Scheduling decision results include information such as cleaning urgency, cleaning time, cleaning area, and whether cleaning is necessary. Based on the predictive optimization result and scheduling decision results, a dynamic adaptive cleaning control scheme is generated, including scheduling drones for cleaning, adjusting the concentration and dosage of cleaning fluid, ensuring all cleaning equipment (such as water sources, cleaning agents, robotic arms, etc.) is available, and cleaning is carried out according to the predetermined path and method based on the cleaning scheme parameters. Sensors monitor data such as efficiency recovery, energy consumption, and environmental disturbance risk during the cleaning process. If anomalies occur (such as excessive energy consumption or increased environmental disturbance risk), the cleaning scheme is adjusted immediately.

[0089] The final cleaning control plan includes cleaning schedule, cleaning method selection, and cleaning area division, ensuring it is the most suitable cleaning control solution for the current surface photovoltaic modules. When a valve is opened, the corresponding camera feed is automatically invoked to monitor the valve's opening and closing status in real time. Simultaneously, a timed task function is added to the cleaning display page, allowing users to set scheduled cleaning tasks to achieve comprehensive and timely spray rinsing of the surface photovoltaic modules, preventing the accumulation of dust that forms a crystalline shell. By predictively optimizing the cleaning plan and generating a cleaning control scheme, cleaning activities are executed precisely, improving cleaning efficiency and effectiveness.

[0090] In summary, the intelligent cleaning control method for water-surface photovoltaic modules provided in this application has the following advantages:

[0091] Beneficial effects:

[0092] By reading power generation data from the surface photovoltaic modules, a power generation efficiency perception is constructed using the data. This perception is then used to conduct a retrospective evaluation of pollution impact, establishing the evaluation results. A drone is activated to patrol the surface photovoltaic modules along a preset flight path, constructing images of the photovoltaic panels. Feature extraction is performed on the photovoltaic panel images to identify the pollution type, distribution area, and intensity. Based on the pollution type, distribution area, and intensity, the pollution level is determined. The pollution level, retrospective evaluation results, and environmental conditions are used to make scheduling decisions for cleaning schemes, generating scheduling decision results. The cleaning scheme is then predicted and optimized using the pollution type, distribution area, and intensity. A cleaning control scheme is generated based on the prediction and optimization results and the scheduling decision results. In other words, by analyzing changes in power generation efficiency, a retrospective assessment of pollution impact is conducted. Simultaneously, drones are used for inspections to determine pollution levels. Based on the retrospective assessment results and environmental conditions, scheduling decisions are made. Pollution predictions are performed based on pollution type, distribution area, and intensity. Cleaning plans are then optimized to generate cleaning control schemes. This enables accurate identification and assessment of pollution, and real-time optimization of cleaning plans, making them more flexible and improving the maintenance efficiency and power generation performance of the water-surface photovoltaic modules.

[0093] Example 2: Based on the same inventive concept as the intelligent cleaning control method for water-surface photovoltaic modules in Example 1, this application also provides an intelligent cleaning control system for water-surface photovoltaic modules. Please refer to the appendix. Figure 2 The intelligent cleaning control system for water surface photovoltaic modules includes:

[0094] The impact retrospective evaluation module 11 is used to read power generation data of the water surface photovoltaic modules, construct power generation efficiency perception using the power generation data reading results, and conduct a pollution impact retrospective evaluation using the power generation efficiency perception to establish a retrospective evaluation result; the UAV inspection module 12 is used to activate UAVs to inspect the water surface photovoltaic modules according to a preset flight path and construct photovoltaic panel images; the feature extraction module 13 is used to extract features from the photovoltaic panel images and identify the pollution type, distribution area, and distribution intensity; the scheduling decision module 14 is used to determine the pollution level based on the pollution type, distribution area, and distribution intensity, and make scheduling decisions for the cleaning scheme using the pollution level, retrospective evaluation results, and environmental conditions to generate scheduling decision results; the scheme optimization module 15 is used to predict and optimize the cleaning scheme using the pollution type, distribution area, and distribution intensity, and generate a cleaning control scheme based on the prediction optimization results and scheduling decision results.

[0095] Furthermore, the impact retrospective evaluation module 11 in the intelligent cleaning control system for water-surface photovoltaic modules is also used for:

[0096] Obtain the equipment parameters of the water surface photovoltaic module, and configure the factory power generation efficiency using the equipment parameters; read the usage time of the water surface photovoltaic module, and perform attenuation fitting of the factory power generation efficiency based on the usage time to establish attenuation fitting results; use the attenuation fitting results to compensate for the efficiency impact of the power generation efficiency perception, and establish a pollution impact retrospective evaluation based on the efficiency impact compensation results.

[0097] Furthermore, the impact retrospective evaluation module 11 in the intelligent cleaning control system for water-surface photovoltaic modules is also used for:

[0098] A pollution sample is established, and array tests of photovoltaic panels are conducted based on the pollution sample to establish a test set, which includes the efficiency impact test results. After mapping and identifying the pollution sample and the test set, a pollution impact retrospective evaluation network is constructed using the mapped and identified pollution sample and test set. The pollution impact retrospective evaluation network includes a depositional light transmission impact retrospective network and an accumulation temperature rise impact retrospective network. The efficiency impact compensation results are input into the pollution impact retrospective evaluation network, and retrospective analysis of the light transmission impact and temperature rise impact is performed through the depositional light transmission impact retrospective network and the accumulation temperature rise impact retrospective network, respectively, to generate depositional data. The pollution impact retrospective evaluation is completed based on the depositional data.

[0099] Furthermore, the feature extraction module 13 in the intelligent cleaning control system for water-surface photovoltaic modules is also used for:

[0100] After image distortion correction of the photovoltaic panel image, illumination equalization adjustment is performed and color normalization is applied. A convolutional neural network is used to traverse and extract pollution features from the adjusted photovoltaic panel image to establish a pollution feature map. The traversal extraction includes color feature extraction, texture feature extraction, boundary contour feature extraction, and spatial location extraction. After multidimensional feature extraction, semantic fusion of multidimensional features with the same spatial location is performed to identify the pollution type, distribution area, and distribution intensity.

[0101] Furthermore, the scheduling decision module 14 in the intelligent cleaning control system for water-surface photovoltaic modules is also used for:

[0102] Construct a joint scoring function, which is as follows:

[0103] J(Zone i )=ω1·PL i +ω2·RE i +ω3·f(ES i ); where J(Zone) i ) is a joint scoring function that characterizes the urgency of cleaning. PL i According to the pollution level, RE i To retrospectively assess the impact score, f(ES) i The environmental state adjustment function is constructed from the current air temperature, wind speed and direction disturbances, and rainfall forecast. ω1, ω2, and ω3 are the weight coefficients of the pollution level item, the retrospective impact item, and the environmental impact item, respectively. The urgency of the cleaning plan is evaluated using the joint scoring function, and the scheduling decision is made based on the evaluation results.

[0104] Furthermore, the scheduling decision module 14 in the intelligent cleaning control system for water-surface photovoltaic modules is also used for:

[0105] Obtain the power grid dispatch plan, establish a plan avoidance factor based on the power grid dispatch plan, and use the plan avoidance factor to establish constraints for the dispatch decision results.

[0106] Furthermore, the solution optimization module 15 in the intelligent cleaning control system for water-surface photovoltaic modules is also used for:

[0107] The cleaning time interval is determined based on the scheduling decision result; it is determined whether there is a rainfall impact within the cleaning time interval; if there is no rainfall impact, the time node of the last cleaning is obtained, and a backtracking interval is established based on the time node; pollution prediction fitting is performed using the backtracking interval, the cleaning time interval, the pollution type, distribution area and distribution intensity, and the cleaning scheme is optimized based on the pollution prediction fitting result.

[0108] Furthermore, the solution optimization module 15 in the intelligent cleaning control system for water-surface photovoltaic modules is also used for:

[0109] If rainfall has an impact, the pollution prediction fitting results are reconstructed after generating impact compensation based on the rainfall impact; the reconstructed pollution prediction fitting results are then used to optimize the cleaning scheme.

[0110] Furthermore, the solution optimization module 15 in the intelligent cleaning control system for water-surface photovoltaic modules is also used for:

[0111] The pollution prediction fitting results are used to match cleaning strategies, and the strategy matching results are established. A multi-objective evaluation function is constructed, and the evaluation indicators of the multi-objective evaluation function include post-cleaning efficiency recovery rate, cleaning energy consumption, and environmental disturbance risk indicators during execution. The multi-objective evaluation function is used to optimize the strategy matching results, and the optimization is completed.

[0112] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Figure 1 The intelligent cleaning control method and specific examples for water-surface photovoltaic modules in Embodiment 1 are also applicable to the intelligent cleaning control system for water-surface photovoltaic modules in this embodiment. Through the foregoing detailed description of the intelligent cleaning control method for water-surface photovoltaic modules, those skilled in the art can clearly understand the intelligent cleaning control system for water-surface photovoltaic modules in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here. As for the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant details can be found in the method section.

[0113] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0114] Obviously, those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A smart cleaning control method for a water surface facing photovoltaic module, characterized in that, The method comprises the following steps: reading power generation data of the water surface photovoltaic assembly, constructing power generation efficiency perception using the reading result of the power generation data, performing pollution influence backtracking evaluation using the power generation efficiency perception, establishing backtracking evaluation result; activating the unmanned aerial vehicle to patrol the water surface photovoltaic assembly according to the preset flight path, and constructing a photovoltaic panel image; after feature extraction is performed on the photovoltaic panel image, identifying the pollution type, distribution area and distribution intensity; determining the pollution level based on the pollution type, distribution area and distribution intensity, and performing scheduling decision of the cleaning scheme using the pollution level, backtracking evaluation result and environmental state to generate scheduling decision result; performing scheme prediction optimization of the cleaning scheme using the pollution type, distribution area and distribution intensity, and generating the cleaning control scheme according to the prediction optimization result and the scheduling decision result; the backtracking evaluation of the pollution influence using the power generation efficiency perception comprises: obtaining equipment parameters of the water surface photovoltaic assembly, and configuring factory power generation efficiency using the equipment parameters; reading the use time length of the water surface photovoltaic assembly, performing attenuation fitting of the factory power generation efficiency according to the use time length, and establishing attenuation fitting result; performing efficiency influence compensation of the power generation efficiency perception using the attenuation fitting result, and establishing the pollution influence backtracking evaluation according to the efficiency influence compensation result, which comprises: establishing a pollution sample, performing array test of the photovoltaic panel according to the pollution sample, establishing a test set, and the test set comprises efficiency influence test result; after mapping and identifying the pollution sample and the test set, constructing a pollution influence backtracking evaluation network using the pollution sample and the test set, and the pollution influence backtracking evaluation network comprises a deposition light transmission influence backtracking sub-network and an accumulation temperature rise influence backtracking sub-network; inputting the efficiency influence compensation result into the pollution influence backtracking evaluation network, respectively performing backtracking analysis of light transmission influence and temperature rise influence through the deposition light transmission influence backtracking sub-network and the accumulation temperature rise influence backtracking sub-network, generating deposition data, and completing the pollution influence backtracking evaluation according to the deposition data; the method for generating the scheduling decision result comprises: obtaining a power grid scheduling plan, establishing a plan avoidance factor according to the power grid scheduling plan; performing establishment constraint of the scheduling decision result using the plan avoidance factor.

2. The intelligent cleaning control method for a water surface-oriented photovoltaic module according to claim 1, wherein, after the feature extraction is performed on the photovoltaic panel image, the pollution type, distribution area and distribution intensity are identified, which comprises: after image distortion is performed on the photovoltaic panel image, light balance adjustment is performed, and color normalization is performed; pollution feature map is established by traversing and extracting the adjusted photovoltaic panel image using a convolutional neural network, and the traversal and extraction comprises color feature extraction, texture feature extraction, boundary contour feature extraction and spatial position extraction; after multi-dimensional feature extraction is performed, semantic fusion of the multi-dimensional features in the same spatial position is performed to identify the pollution type, distribution area and distribution intensity.

3. The intelligent cleaning control method for a water surface-oriented photovoltaic module according to claim 1, wherein, the method for performing scheduling decision of the cleaning scheme using the pollution level, backtracking evaluation result and environmental state comprises: constructing a joint scoring function, and the joint scoring function is as follows: ； wherein, is a joint score function, representing the urgency of cleaning, is a pollution level, is a backtracking impact score, is an environmental state adjustment function, constructed from current air temperature, wind speed and direction disturbances, and rainfall predictions, are weight coefficients for the pollution level term, the backtracking impact term, and the environmental impact term, respectively; performing urgency evaluation of the cleaning scheme using the joint scoring function, and completing the scheduling decision according to the evaluation result.

4. The intelligent cleaning control method for a water surface-oriented photovoltaic module according to claim 1, wherein, The scheme prediction optimization optimization of the cleaning scheme using the pollution type, distribution area and distribution intensity comprises: determining a cleaning time interval according to the scheduling decision result; judging whether there is rainfall influence in the cleaning time interval; if there is no rainfall influence, obtaining a time node of the last cleaning, and establishing a backtracking interval according to the time node; performing pollution prediction fitting using the backtracking interval, the cleaning time interval, the pollution type, the distribution area and the distribution intensity, and completing cleaning scheme optimization optimization according to the pollution prediction fitting result.

5. The intelligent cleaning control method for a water surface-oriented photovoltaic module according to claim 4, wherein, The judgment of whether there is rainfall influence in the cleaning time interval further comprises: if there is rainfall influence, generating influence compensation according to the rainfall influence, reconstructing the pollution prediction fitting result, and completing the cleaning scheme optimization optimization using the reconstructed pollution prediction fitting result. The cleaning scheme optimization optimization using the reconstructed pollution prediction fitting result comprises:

6. The intelligent cleaning control method for a water surface-oriented photovoltaic module according to claim 5, wherein, performing strategy matching of the cleaning strategy using the pollution prediction fitting result, establishing a strategy matching result, constructing a multi-objective evaluation function, and the evaluation index of the multi-objective evaluation function comprises a cleaning efficiency recovery rate, a cleaning energy consumption and an environmental disturbance risk index during execution; performing optimization matching of the strategy matching result using the multi-objective evaluation function, and completing optimization optimization. The steps for implementing the intelligent cleaning control method for the water surface photovoltaic module in any one of claims 1 to 6, the intelligent cleaning control system for the water surface photovoltaic module comprises: an influence backtracking evaluation module for reading power generation data of the water surface photovoltaic module, constructing power generation efficiency perception using the power generation data reading result, performing pollution influence backtracking evaluation using the power generation efficiency perception, and establishing a backtracking evaluation result; 7. A smart cleaning control system for a water surface facing photovoltaic module, characterized by, an unmanned aerial vehicle patrol module for activating the unmanned aerial vehicle to patrol the water surface photovoltaic module according to a preset flight path, and constructing a photovoltaic panel image; a feature extraction module for identifying a pollution type, a distribution area and a distribution intensity after feature extraction of the photovoltaic panel image; a scheduling decision module for determining a pollution level based on the pollution type, the distribution area and the distribution intensity, and performing scheduling decision of a cleaning scheme using the pollution level, the backtracking evaluation result and an environmental state, and generating a scheduling decision result; a scheme optimization module for performing scheme prediction optimization optimization of the cleaning scheme using the pollution type, the distribution area and the distribution intensity, and generating a cleaning control scheme according to the prediction optimization optimization result and the scheduling decision result. ​ ​

Images