Method and system for controlling water consumption of photovoltaic panel cleaning robot

Abstract

The invention relates to the technical field of photovoltaic power generation operation and maintenance, in particular to a water consumption control method and system for a photovoltaic panel cleaning robot, and the method comprises the steps: synchronously collecting a pollution feature image, real-time environment parameters and robot body operation parameters of a current to-be-cleaned photovoltaic panel region; performing feature recognition on the pollution feature image to obtain a pollution type and a pollution level of the to-be-cleaned photovoltaic panel area; performing evaporation evaluation on the real-time environment parameters to obtain an evaporation intensity index of the photovoltaic panel area to be cleaned; parameters such as pollution types are used as index dimensions, indexing is performed in a preset water consumption reference mapping table, and basic unit water consumption is determined; the basic unit water consumption, the pollution level and the robot body operation parameters are input into a preset real-time cleaning optimization model, a segmented cleaning control parameter instruction is obtained, and the cleaning robot is controlled to execute cleaning operation according to the segmented cleaning control parameter instruction. According to the invention, water consumption dynamic calculation and multi-objective optimization control can be carried out based on real-time pollution characteristics and environmental parameters.

Description

Technical Field

[0001] This invention relates to the technical field of photovoltaic power generation operation and maintenance, and in particular to a method and system for controlling water consumption of a photovoltaic panel cleaning robot. Background Technology

[0002] Photovoltaic panel cleaning is a crucial maintenance step in ensuring power generation efficiency. Currently, automated cleaning robots are widely used in photovoltaic power generation maintenance, and their water usage control largely relies on preset fixed programs or control logic based on simple feedback, such as timed or fixed-stroke control. However, the contamination status of photovoltaic panel surfaces and on-site environmental conditions are complex and variable. Existing control methods struggle to calculate the optimal cleaning parameters that minimize water consumption while ensuring cleaning cleanliness, taking into account real-time, multi-dimensional factors. This often leads to redundant water usage strategies in practice to ensure cleaning effectiveness, failing to achieve an optimal balance between cleaning efficiency and water consumption under changing working conditions.

[0003] Therefore, there is an urgent need for a method that can dynamically calculate water consumption and perform multi-objective optimization control based on real-time pollution characteristics and environmental parameters, in order to solve the technical problem of accurately controlling and minimizing cleaning water consumption while ensuring preset cleaning standards. Summary of the Invention

[0004] This invention provides a method and system for controlling water consumption in a photovoltaic panel cleaning robot, which can effectively solve the problems in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for controlling water consumption in a photovoltaic panel cleaning robot includes: Simultaneously collect pollution characteristic images, real-time environmental parameters, and robot body operation parameters of the photovoltaic panel area to be cleaned; The pollution feature image is used to identify the pollution type and pollution level of the photovoltaic panel area to be cleaned; The evaporation intensity index of the photovoltaic panel area to be cleaned is obtained by evaluating the real-time environmental parameters. Using the pollution type, pollution level, and evaporation intensity index as index dimensions, the water consumption per unit area is determined by indexing in a pre-set water consumption baseline mapping table. The basic unit water consumption, the pollution level, and the robot's operating parameters are input into a preset real-time cleaning optimization model to obtain segmented cleaning control parameter instructions, and the cleaning robot is controlled to perform cleaning operations accordingly.

[0006] Furthermore, the robot's operating parameters include the current walking speed, the current brush pressure, and the current water pump pressure; The real-time environmental parameters include the surface temperature of the photovoltaic panel, light intensity, ambient temperature, and wind speed.

[0007] Furthermore, the pollution types include uniform dust coverage, sand accumulation, and bird droppings.

[0008] Furthermore, feature recognition is performed on the pollution feature image to obtain the pollution type and pollution level of the photovoltaic panel area to be cleaned, including: The pollution feature image is analyzed using a pre-trained convolutional neural network model, the classification result of the pollution type is output, and an image mask identifying the pollution area is generated. The pollution level is determined by comprehensively considering the proportion of pixels covered by the image mask, the average gray value within the area, and the texture feature parameters. The pollution level is quantified into multiple discrete levels. The higher the pollution coverage ratio, the greater the difference between the gray value and the background, or the coarser the texture, the higher the pollution level is determined.

[0009] Furthermore, the method for constructing the water reference mapping table includes: Under standard testing conditions, for each combination of pollution type and pollution level, and for different intervals corresponding to the evaporation intensity index, the minimum unit water consumption is determined by orthogonal experimental method to ensure that the preset basic cleanliness standard is met. Using the pollution type, pollution level, and evaporation intensity index range as input features, and the minimum unit water consumption as output label, a regression model is trained. The output of the regression model is then discretized to form the water consumption benchmark mapping table.

[0010] Furthermore, the segmented cleaning control parameter instructions include the segmented travel speed of the cleaning path, the segmented water flow rate, and the segmented brush head pressure.

[0011] Furthermore, the real-time cleaning optimization model is a constrained nonlinear programming model, whose objective function is to minimize total water consumption, and the constraints include: The cleanliness prediction value for each cleaning segment must be greater than or equal to a preset threshold. The cleanliness prediction value is calculated by the pollution level, water flow rate, brush head pressure and travel speed of the segment through a preset prediction sub-model.

[0012] Furthermore, the preset prediction sub-model is a machine learning model trained based on historical cleaning data. Its inputs are the contamination level of the cleaning segment, the planned water flow rate, the brush head pressure, and the travel speed, and its output is the cleanliness prediction value of the cleaning segment. The cleanliness prediction value is positively correlated with the water flow rate and the brush head pressure, and negatively correlated with the travel speed and the contamination level. Furthermore, the negative impact of the contamination level on the cleaning effect exhibits an exponential decay trend.

[0013] Furthermore, the real-time cleaning optimization model employs a particle swarm optimization algorithm; During the iterative solution process of the particle swarm optimization algorithm, the search step size of the particle swarm optimization algorithm is adjusted according to the difference between the cleanliness prediction value corresponding to the current candidate solution and the preset threshold. When the cleanliness prediction value is close to the preset threshold, the search step size is reduced to perform a fine search.

[0014] On the other hand, the present invention also provides a water consumption control system for a photovoltaic panel cleaning robot, comprising: The data acquisition module is used to simultaneously collect pollution characteristic images, real-time environmental parameters, and robot body operating parameters of the photovoltaic panel area to be cleaned. The pollution feature recognition module is used to perform feature recognition on the pollution feature image to obtain the pollution type and pollution level of the photovoltaic panel area to be cleaned; An environmental parameter processing module is used to evaluate the evaporation of the real-time environmental parameters and obtain the evaporation intensity index of the photovoltaic panel area to be cleaned. The baseline water consumption determination module is used to index the pollution type, pollution level and evaporation intensity index in a preset water consumption baseline mapping table to determine the basic unit water consumption for the current photovoltaic panel area to be cleaned. The optimization control module is used to input the basic unit water consumption, the pollution level and the robot body operating parameters into a preset real-time cleaning optimization model to obtain segmented cleaning control parameter instructions, and control the cleaning robot to perform cleaning operations accordingly.

[0015] The technical solution of this invention can achieve the following technical effects: This invention couples the characteristic identification results of pollution type and pollution level with the environmental evaporation intensity assessment results to form a composite index for retrieving the water use benchmark mapping table. This ensures that the determined basic unit water consumption reflects both the inherent difficulty of pollutant removal and the risk of water loss caused by the environment, making the water consumption benchmark value fit the actual physical process and solving the problem of insufficient adaptability caused by setting water consumption based solely on the degree of pollution or fixed experience values. By simultaneously inputting the basic unit water consumption and the real-time collected robot body operating parameters into the real-time cleaning optimization model, with minimizing water consumption as the core objective and cleanliness prediction value as the key constraint, the benchmark water consumption is used to provide the optimization starting point and boundary that conforms to the current working conditions. At the same time, the response state of the cleaning mechanism is compensated and adjusted according to the real-time operating parameters, and the segmented cleaning control parameter instructions that can adapt to the differentiated pollution and evaporation state of each area under the global water consumption constraint are solved. The feature recognition and evaporation assessment steps provide an initial decision basis for water consumption calculation; the water use benchmark mapping step efficiently transforms multi-source features into an operable water use benchmark based on prior knowledge; and the real-time cleaning optimization model step integrates the initial decision, real-time status, and multi-objective constraints to generate execution instructions, enabling the cleaning robot to autonomously maintain the optimal balance between preset cleaning standards and water consumption efficiency under complex and variable working conditions.

[0016] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the water consumption control method for a photovoltaic panel cleaning robot in an embodiment of the present invention. Figure 2 This is a logic block diagram of the water consumption control system for the photovoltaic panel cleaning robot in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] like Figure 1 As shown, the water consumption control method for a photovoltaic panel cleaning robot of the present invention specifically includes the following steps: Step S1: Synchronously collect pollution characteristic images, real-time environmental parameters, and robot body operation parameters of the photovoltaic panel area to be cleaned; Step S2: Perform feature recognition on the pollution feature image to obtain the pollution type and pollution level of the photovoltaic panel area to be cleaned; Step S3: Evaporation assessment of the real-time environmental parameters to obtain the evaporation intensity index of the photovoltaic panel area to be cleaned; Step S4: Using the pollution type, pollution level, and evaporation intensity index as index dimensions, index the preset water consumption baseline mapping table to determine the basic unit water consumption for the current photovoltaic panel area to be cleaned. Step S5: Input the basic unit water consumption, the pollution level and the robot body operating parameters into the preset real-time cleaning optimization model to obtain segmented cleaning control parameter instructions, and control the cleaning robot to perform cleaning operations accordingly.

[0022] In this embodiment, by coupling the feature identification results of pollution type and pollution level with the environmental evaporation intensity assessment results, a composite index is used to retrieve the water usage benchmark mapping table. This ensures that the determined basic unit water usage simultaneously reflects the inherent difficulty of pollutant removal and the risk of water loss caused by the environment, making the water usage benchmark value conform to the actual physical process and solving the problem of insufficient adaptability caused by setting water usage solely based on pollution level or fixed empirical values. By simultaneously inputting the basic unit water usage and the real-time collected robot body operating parameters into the real-time cleaning optimization model, with minimizing water usage as the core objective and cleanliness prediction value as the key constraint, the benchmark water usage provides a consistent... The optimization starting point and boundary of the current working condition are determined, and the response state of the cleaning mechanism is compensated and adjusted based on real-time operating parameters. Under global water consumption constraints, segmented cleaning control parameter instructions that can adapt to the differentiated pollution and evaporation states of each area are obtained. The feature recognition and evaporation assessment steps provide initial decision-making basis for water consumption calculation. The water consumption benchmark mapping step efficiently transforms multi-source features into operable water consumption benchmarks based on prior knowledge. The real-time cleaning optimization model step integrates the initial decision, real-time state and multi-objective constraints to generate execution instructions, enabling the cleaning robot to autonomously maintain the optimal balance between preset cleaning standards and water consumption efficiency under complex and variable working conditions.

[0023] In some embodiments of the present invention, the robot's operating parameters include current walking speed, current brush pressure, and current water pump pressure; the real-time environmental parameters include photovoltaic panel surface temperature, light intensity, ambient temperature, and wind speed; to achieve data collection on the current state of the photovoltaic panel area to be cleaned, the following steps are specifically included: Step S11: Fix a high-definition industrial camera at the center of the front beam of the cleaning robot, with the lens vertically facing the photovoltaic panel surface and the focal length adapted to the robot's working height to ensure that the shooting range covers a single segment of the current cleaning path; symmetrically arrange integrated sensing units on both sides of the camera, each unit containing a photovoltaic panel surface temperature sensor, an ambient temperature sensor, a light intensity sensor, and a miniature wind speed sensor. The sensing unit probes are exposed and maintain a fixed distance from the photovoltaic panel surface to avoid contact with the panel or contaminants during cleaning operations; the robot's walking motor, brush drive motor, and water pump are respectively equipped with incremental encoders, pressure sensors, and pressure transmitters to collect operating parameters in real time; Step S12: The photovoltaic panels to be cleaned are divided into continuous rectangular acquisition segments according to the width and length of the robot's single cleaning operation. The boundary coordinates of each segment are stored in the robot control system. When the robot moves to the starting position of a certain acquisition segment, the control system triggers a synchronous acquisition signal. This signal is simultaneously sent to the high-definition industrial camera, the integrated sensing unit, and the detection elements of each motor / pump, triggering the camera to capture pollution feature images, the sensing unit to detect environmental parameters, and the detection elements to collect operating parameters. Step S13: When the acquisition signal is triggered, the control system synchronously records the current positioning coordinates of the robot, which are provided by the robot's own positioning module. These coordinates are used as the area identifier of the current acquisition segment. The pollution feature image captured by the high-definition industrial camera is embedded with the positioning coordinates and acquisition timestamp. The environmental parameters detected by the integrated sensing unit and the operating parameters collected by each detection element are all associated with the same positioning coordinates and timestamp. All data are stored in the robot's local storage unit with the positioning coordinates and timestamp as the index, forming a multi-source parameter set for a single acquisition segment.

[0024] In this embodiment, the design of preset acquisition segments, synchronous trigger signals, and coordinate association indexes solves the problem of temporal and spatial mismatch in multi-source data acquisition. Synchronous triggering ensures that pollution images, environmental parameters, and operating parameters correspond to the same physical moment, eliminating timing errors caused by sensor response delays or differences in sampling periods. The acquired data is strictly associated with the robot's precise real-time positioning coordinates to avoid incorrect combination of perception information from different spatial locations. All data is stored with a unified spatiotemporal coordinate index, constructing a single acquisition segment multi-source parameter set with strictly consistent spatiotemporal attributes, improving the accuracy of pollution assessment, evaporation calculation, and control parameter optimization.

[0025] In some embodiments of the present invention, the pollution feature image is analyzed using a pre-trained convolutional neural network model to output a classification result of the pollution type and generate an image mask identifying the pollution area. The pollution types include uniform dust coverage, sand accumulation, and bird droppings. The pollution level is comprehensively determined based on the pixel ratio covered by the image mask, the average gray value within the area, and texture feature parameters. The pollution level is quantified into multiple discrete levels; a higher pollution coverage ratio, a greater difference between the gray value and the background, or a coarser texture indicates a higher pollution level. Specific implementation details are as follows: Step S21: Construct a pre-trained convolutional neural network model, specifically including: collecting image samples of photovoltaic panels with uniform dust coverage, sand accumulation, bird droppings spot pollution, and clean photovoltaic panels. The samples need to cover the surface state of photovoltaic panels under different light intensities and ambient temperatures; and labeling each sample image with two labels: one is the pollution type label, including uniform dust coverage, sand accumulation, bird droppings spot pollution, and clean; the other is the pollution area boundary coordinate label in the form of pixel coordinate groups. A convolutional neural network model is built based on the ResNet50 architecture. A pollution-specific feature extraction layer is added between the convolutional layers and fully connected layers of the model. Specialized convolutional kernels are set for features of various pollution types. For example, the texture of uniformly covered dust is continuously distributed, so a 1×5 horizontal convolutional kernel and a 5×1 vertical convolutional kernel are configured; the edge contour of sand accumulation is irregular, so a 3×3 edge detection convolutional kernel is configured; bird droppings are isolated clumps, so a 5×5 point feature convolutional kernel is configured. The model training uses cross-validation, dividing the labeled samples into training, validation, and test sets. The Adam optimizer is selected, with an initial learning rate of 0.001. When the validation accuracy does not improve for 5 consecutive epochs, the learning rate is reduced to 1 / 10 of the original. After training, the pollution feature image obtained in step S1 is input, and the model outputs two results: one is the pollution type classification result, and the other is a binarized image mask, where the pixel value of the polluted area is set to 1 and the pixel value of the clean background area is set to 0.

[0026] Step S22: Perform morphological closing operation optimization on the binarized image mask output by the model; specifically, use 3×3 rectangular structuring elements, expand the mask by filling the small holes inside the contaminated area, and then perform erosion operation to eliminate isolated noise points at the mask edge, ensuring that the contaminated area outline is complete and free of redundant noise; by traversing the optimized image mask pixel by pixel, count the total number of contaminated pixels with a pixel value of 1, calculate the ratio of the total number of pixels to the total number of pixels in the image, and obtain the contaminated coverage ratio.

[0027] Step S23: Based on the optimized image mask, segment the contaminated area sub-image from the original contaminated feature image, retaining only the original image area corresponding to the mask pixel value of 1, and convert the sub-image into an 8-bit grayscale image; acquire grayscale images of the same type of clean photovoltaic panel under standard illumination conditions, and calculate its grayscale mean as the standard grayscale value; calculate the grayscale mean of the contaminated area sub-image, and calculate the absolute difference with the standard grayscale value to obtain the grayscale difference value, which is used to reflect the information related to the contaminated thickness; use the gray-level co-occurrence matrix method to calculate the texture feature parameters of the contaminated area sub-image. Specifically, set the distance parameter of the gray-level co-occurrence matrix to 1 pixel, the angle parameters to 0°, 45°, 90°, and 135°, and the grayscale level to 256 levels. Calculate the entropy value used to reflect the texture complexity, the contrast value used to reflect the texture brightness difference, and the energy value used to reflect the texture uniformity at the four angles, and take the average value of each parameter at the four angles as the final texture feature parameters.

[0028] Step S24: Preset multiple discrete pollution levels, and determine the threshold range corresponding to each level through orthogonal experiments; Under a standard test environment, for different combinations of pollution coverage ratio, grayscale difference, and texture feature parameters, determine the minimum water consumption required to clean to the preset standard, classify the cleaning difficulty level according to water consumption, and then determine the parameter threshold range corresponding to each level in reverse; Calculate the comprehensive score using a weighted summation method, with the pollution coverage ratio having the highest weight, and the grayscale difference and texture feature comprehensive value having the same weight; The texture feature comprehensive value is calculated by normalizing the entropy value, contrast, and energy to the same interval, summing the three and taking the average; Determine the pollution level based on the threshold range that the comprehensive score falls into, with a higher score indicating a higher pollution level and greater cleaning difficulty.

[0029] In this embodiment, the convolutional neural network model adds a pollution-specific feature extraction layer and a dedicated convolutional kernel to specifically identify the morphological and texture differences of the three types of pollution, reduce classification confusion in general models, and improve the matching degree between classification results and actual pollution types. After the image mask is optimized by morphological closing operation, it achieves precise isolation between the polluted area and the background. The pollution coverage ratio, gray-level difference, and texture feature parameters are all calculated based on the polluted area, eliminating the interference of clean background pixels. The pollution level is calculated by weighting the pollution coverage ratio, gray-level difference, and texture feature parameters. A single parameter cannot fully characterize the impact of pollution on cleaning. The integration of multi-dimensional parameters can comprehensively match actual cleaning needs.

[0030] In a specific implementation, as one example, evaporation intensity indicators for the photovoltaic panel area to be cleaned are obtained by evaluating real-time environmental parameters. The evaporation rate of moisture on the photovoltaic panel surface is a key environmental interference factor determining the effectiveness of the cleaning water. The temperature rise of the photovoltaic panel surface under illumination, the influence of its specific surface roughness and tilt angle on the airflow near the wall, and the changes in surface thermodynamic properties caused by the contamination layer all make the evaporation process significantly different from that of open water bodies or general wet surfaces. If the evaluation model fails to incorporate the above key influencing factors, the obtained evaporation intensity indicators cannot accurately reflect the actual retention risk of the cleaning fluid on the actual working surface, leading to inaccurate water volume control. To solve the above problems, this embodiment constructs a structured computational framework that includes driving terms, transport terms, and surface state correction terms. This framework is based on classical evaporation theory and introduces special parameters and correction functions for photovoltaic application scenarios, transforming the collected multi-point environmental parameters into a single quantitative indicator characterizing the current instantaneous evaporation potential. The specific implementation is as follows: Step S31: Extract instantaneous measurements of photovoltaic panel surface temperature, ambient temperature, wind speed, and light intensity from the synchronously acquired data frames; simultaneously, retrieve inherent parameters related to the photovoltaic panel model and installation location from the pre-stored database, including the photovoltaic panel surface absorptivity to solar radiation, the photovoltaic panel installation tilt angle, and aerodynamic roughness parameters obtained through wind tunnel experiments or on-site calibration of the airflow characteristics of the photovoltaic panel array surface.

[0031] Step S32: Calculate the core thermodynamic driving potential energy of evaporation, which is characterized by the saturated vapor pressure difference. Specifically, based on the measured surface temperature of the photovoltaic panel and the measured ambient temperature, the corresponding saturated vapor pressure values ​​are calculated using the saturated vapor pressure calculation formula. The difference between the two saturated vapor pressure values ​​is the core thermodynamic driving potential energy that drives water to evaporate from the panel to the air.

[0032] Step S33: Calculate the net energy input for water vaporization. This calculation mainly considers solar radiation energy. First, calculate the solar incidence angle based on the photovoltaic panel installation tilt angle, current geographical location, and time. Multiply the measured light intensity by the panel surface radiative absorptivity, and then by the cosine of the incidence angle to obtain the short-wave radiation energy absorbed by the panel surface. Next, calculate the net long-wave radiation exchange value between the panel surface and the atmosphere based on the Stefan-Boltzmann law, according to the surface temperature and ambient temperature. Subtract the long-wave radiation loss from the absorbed short-wave radiation energy to obtain the net radiation energy input value acting on the panel surface and the water film.

[0033] Step S34: Calculate the water vapor mass transport efficiency under wind speed. This calculation introduces a wind speed correction function based on the logarithmic law. The wind speed correction function takes the measured wind speed as input and combines it with the specific aerodynamic roughness parameters of the photovoltaic panel to calculate the transport coefficient that characterizes the ability of water vapor to diffuse from the panel surface to the atmosphere. This ensures that the air retention effect is compensated at low wind speeds and that the increase in transport efficiency is smoothly limited in accordance with the laws of fluid boundary layer at high wind speeds.

[0034] Step S35: The net radiation energy input value and the water vapor mass transfer efficiency value are combined according to their respective physical weights in evaporation to obtain a preliminary evaporation rate estimate. This merging process follows the principle of energy and mass conservation, and the weighting coefficients are determined by the local slope of the temperature-saturated vapor pressure relationship curve and the wet / dry constant. Then, a dynamic correction factor is introduced. The dynamic correction factor is obtained from a preset lookup table based on the current regional pollution level determined in real time in step S2. This lookup table is established through experiments and reflects the comprehensive influence of different pollution types and thicknesses on surface temperature, hydrophilicity, and thermal insulation, thereby dynamically adjusting the evaporation rate estimate to obtain the corrected evaporation rate.

[0035] Step S36: Divide the corrected evaporation rate by the standard evaporation rate reference value calculated by this model under predefined standard environmental conditions. The standard environmental conditions include standard temperature, standard humidity, standard wind speed, standard light intensity, and clean surface condition. This division operation produces a dimensionless ratio, which is the final evaporation intensity index. An evaporation intensity index greater than 1 indicates that the current environmental evaporation capacity is higher than the standard conditions, while a value less than 1 indicates that it is lower than the standard conditions.

[0036] In this embodiment, a structured physical hybrid model is established to integrate thermodynamic, radiative, and aerodynamic factors affecting the evaporation process. By introducing specific parameters for the surface characteristics of photovoltaic panels and a logarithmic-law-based wind speed correction function, the model is made closer to the real microclimate conditions of the photovoltaic panel surface. By incorporating a dynamic correction factor based on real-time pollution levels, the evaporation assessment can respond to the dynamic changes in surface properties during the cleaning operation. Finally, a standardized evaporation intensity index is output, enabling the water volume control strategy to be adaptively adjusted based on the quantitative assessment of the environmental evaporation potential.

[0037] In some embodiments of the present invention, the basic unit water consumption for the current photovoltaic panel area to be cleaned is determined by offline construction of a water consumption benchmark knowledge base and water consumption determination logic for real-time operations. Specifically, the construction of the water-based benchmark knowledge base was completed through systematic laboratory calibration, as detailed below: In a standard laboratory, using photovoltaic panel samples of the same model as the target to be cleaned, three types of standard contamination samples were prepared: uniform dust coverage, sand accumulation, and bird droppings spot contamination. Each type of contamination was classified into multiple discrete contamination levels based on its coverage density, particle size, or attachment area, and the classification criteria were consistent with the level determination logic in step S2. The contaminated sample is placed in an experimental chamber where environmental conditions can be precisely controlled. The experimental chamber can independently adjust the temperature, humidity, wind speed, and light intensity. By combining and adjusting the above parameters, multiple discrete evaporation intensity levels, ranging from extremely low evaporation to extremely strong evaporation, are simulated. This level division corresponds to the evaporation intensity index range output in step S3. For each defined combination of parameters (pollution type i, pollution level j, evaporation intensity level k), an orthogonal cleaning experiment is performed; a high-precision flow meter and nozzle are used to clean the sample with a specific water volume per unit area; after cleaning, the surface cleanliness is measured using a standard instrument; by iteratively adjusting the water volume per unit area, the minimum water volume that enables the cleanliness to reach the preset qualified threshold for the first time is found; this minimum water volume is the baseline water volume under this set of operating conditions. Minimum water consumption data for all combinations of (pollution type i, pollution level j, evaporation intensity level k) are collected to form a three-dimensional benchmark database. This three-dimensional benchmark database is used as a training set to train a gradient boosting decision tree regression model with pollution type, pollution level, and evaporation intensity index as input features and benchmark water consumption as the output target. This model is used to learn the complex nonlinear mapping relationship and interaction effect between the three input dimensions and water consumption. Finally, the model is subjected to dense grid inference in the entire input space (3 types of pollution × multiple pollution levels × multiple evaporation intensity indices). The predicted water consumption corresponding to each grid point is archived to generate a high-resolution pre-set water consumption benchmark mapping table for quick querying.

[0038] Furthermore, in real-time operations, the basic unit water consumption is queried and determined. After obtaining the pollution type, pollution level, and evaporation intensity index of the current area, the system first attempts to perform an exact match query in the preset water consumption baseline mapping table. The query key value is (pollution type, discrete level of pollution level, discrete interval of evaporation intensity index). If an exact match is found, the corresponding water consumption will be read directly as the basic unit of water consumption. If the query key value lies between the discrete grids of the mapping table, indicating a transitional operating condition that has not been pre-stored, the system will then call the regression model for real-time inference. Continuous pollution type values ​​and pollution level values, along with evaporation intensity index values, are input into the regression model, and the model outputs continuous basic unit water consumption prediction values. These prediction values ​​are then used as the basic unit water consumption under the current operating condition.

[0039] This embodiment overcomes the problem of insufficient accuracy of traditional discrete lookup table methods when parameters fall between grids by constructing continuous mapping relationships using regression models and combining them with real-time interpolation inference. Through model calculation, it achieves continuous and smooth water consumption benchmark output across the entire parameter range, improving the precision of water consumption setting. By using regression models as the underlying logic, the system can perform nonlinear and interactive precise fitting and interpolation of continuously changing input features. Its output basic unit water consumption is more in line with the actual needs under complex working conditions than methods based on simple linear interpolation or nearest neighbor lookup.

[0040] In some embodiments of the present invention, in order to achieve the goal of minimizing total water consumption and meeting segmented cleanliness standards, this embodiment combines the robot body operating parameters in step S1, the segmented pollution levels in step S2, the basic unit water consumption in step S4, and the data collection segment division logic in step S1 to achieve segmented control. The specific implementation is as follows: Step S51: Construct a cleanliness prediction sub-model. Specifically, collect a training sample set composed of historical data, including the measured cleanliness data under different combinations of pollution levels, water flow, brush head pressure, and travel speed in the orthogonal experiment in step S4, as well as the segmented cleaning parameters and corresponding cleanliness detection data recorded in actual operations. The samples cover three types of pollution and fully discrete pollution levels. The sample input features are defined as segmented pollution levels, planned water flow, brush head pressure, and travel speed, and the output features are the measured cleanliness values. The pollution level characteristics were quantitatively preprocessed. Based on the experimental data, an exponential decay relationship between pollution level and cleanliness was fitted, and the exponential coefficient was determined. This coefficient was used to transform the pollution level characteristic values, making the model adapt to the exponential decay law of pollution level on cleaning effect. A gradient boosting regression algorithm was selected to construct a prediction sub-model. Based on the algorithm's ability to identify the interaction effects between features, a nonlinear relationship was fitted between cleanliness and water flow rate and brush head pressure (positive correlation), and cleanliness and travel speed and pollution level (negative correlation). The samples were divided into training set, validation set, and test set according to the proportion. Cross-validation was used to optimize the model hyperparameters, and the prediction error of the test set was controlled within a reasonable range. The goodness of fit met the prediction requirements.

[0041] Step S52: Construct a constrained nonlinear programming optimization model. Specifically, the model takes minimizing the total water consumption as the objective function. The calculation logic of the total water consumption is closely related to the subsequent segmented control parameters and the segmented division standard mentioned above: the total water consumption is the sum of the products of the water flow rate of each segment and the corresponding cleaning time of the segment. The segment cleaning time is calculated by the segment area and the travel speed. The segment area follows the segment division standard of step S1, so that the calculation basis is consistent with the segmented logic mentioned above. The constraints of the nonlinear programming optimization model include: Cleanliness constraints: The predicted cleanliness values ​​for each segment, output by the aforementioned prediction sub-model, must not be lower than a preset qualified threshold. This threshold is consistent with the cleanliness compliance threshold in step S4 of the experiment. The equipment operation safety constraints are as follows: the brush head pressure does not exceed the rated pressure of the robot brush plate, the traveling speed is within the safe operating range designed for the robot, and the water flow does not exceed the rated output of the water pump. The constraint range is determined in combination with the robot operating parameters collected in step S1 to avoid overload damage to the actuator. The parameters are linked and constrained. The water flow rate and the brush head pressure must be matched. When the brush head pressure increases, the water flow rate is adjusted proportionally to prevent the plate from being worn or the cleaning from being incomplete due to excessive pressure and insufficient water flow.

[0042] Step S53: Solve the constrained optimization model above using the particle swarm optimization algorithm: the particle code corresponds to the segmented control parameter dimension, and each particle contains three parameters: the travel speed, water flow rate, and brush head pressure of each cleaning segment. The particle dimension is consistent with the number of subsequent cleaning segments to ensure that the solution can be directly adapted to the segmented control requirements. When initializing the particle swarm, the initial range of particles is set based on the robot's current operating parameters collected in step S1, and the initial range of water flow rate is limited by the basic unit water consumption determined in step S4 to reduce invalid iterations. During the iterative solution process, a dynamic step size strategy is used to link cleanliness constraints with prediction results. Specifically, a gap threshold range is first set. When the gap between the cleanliness prediction value corresponding to a candidate solution and the preset threshold exceeds the upper limit of the range, a larger search step size is used to accelerate the iteration convergence speed. When the gap is within the range, the search step size is reduced to refine the search and improve the accuracy of the optimal solution. During the iteration process, each generation of particles must satisfy the aforementioned three types of constraints. Particles that violate the constraints are corrected to ensure they fall within the feasible region. An iteration termination condition is set. After the deviation of the optimal solution for multiple consecutive generations stabilizes within the allowable range or the number of iterations reaches the preset upper limit, the iteration is terminated. The current global optimal solution is output as the segmented cleaning control parameter instruction, which includes the specific values ​​of the travel speed, water flow rate, and brush head pressure for each segment.

[0043] Step S54: Generate and execute segmented control parameters; specifically, based on the acquisition segment in step S1, refine the segmentation rules by binding contamination features with robot operation capabilities. Within the same acquisition segment, based on the image mask and contamination level determination results in step S2, split the area into independent cleaning segments with different contamination levels and types; adjust the segment area according to the contamination level, with higher contamination levels resulting in smaller segment areas, and align the segment boundaries with the robot's single cleaning width to avoid parameter adaptation confusion caused by cross-segment cleaning; after segmentation, assign a unique identifier to each segment, associate it with the corresponding contamination level, area, and location coordinates to form a segmentation information table for use during parameter generation; When generating parameters, the basic unit water consumption determined in step S4 is first weighted and allocated according to the pollution level of each segment to obtain the initial water flow of each segment. The higher the pollution level, the greater the weight. Then, the initial water flow, the pollution level of each segment, and the real-time robot operation parameters in step S1 are input into the aforementioned dynamic step size particle swarm optimization algorithm. The algorithm solves and outputs the precise control parameters of each segment, forming a control instruction set with segment identifiers. The instructions clearly specify the effective start coordinates and end coordinates of each segment parameter to ensure that the parameters correspond precisely to the segment positions. The execution process employs segmented synchronous control logic. The main controller on the robot calls the control parameters output by the aforementioned algorithm in the order of segment identifiers, driving the walking mechanism, brush drive mechanism, and water pump to execute the corresponding parameters synchronously. When switching between segments, the controller uses a linear transition strategy to adjust the parameters, avoiding sudden parameter changes that could cause equipment shock or fluctuations in cleaning effect. During operation, a dual-dimensional real-time detection mechanism is established. Cleanliness detection uses a laser reflectivity meter installed on the front crossbeam of the robot. After each segment is cleaned, multiple points are immediately detected for that segment, and the average value is taken as the actual cleanliness. The robot's operating parameters are collected in real time through an incremental encoder and pressure sensor, with the collection frequency synchronized with the cleanliness detection.

[0044] When the actual cleanliness is lower than the preset threshold and the deviation exceeds the allowable range, the aforementioned optimization model is called to solve the problem again, the corresponding segment water flow is adjusted first, the brush head pressure and travel speed are finely adjusted simultaneously, and the rewashing operation is performed after correction. If the brush head pressure or water flow exceeds the aforementioned safety constraints, the corresponding segment operation should be immediately suspended. The model should then regenerate parameters adapted to the equipment performance, and operation should resume after verification.

[0045] In this embodiment, the pollution level characteristics are preprocessed by indexation to make the prediction sub-model conform to the physical law of nonlinear growth in the difficulty of pollutant cleaning. Combined with the gradient boosting regression algorithm, the complex interaction effects between water flow, pressure, velocity and pollution level are captured, improving the accuracy of cleanliness prediction. In the optimization solution stage, the dynamic step size particle swarm optimization algorithm adaptively adjusts the search behavior based on the difference between the cleanliness prediction value and the threshold, taking into account both the convergence speed and the quality and accuracy of the solution under the premise of satisfying constraints. The generation and execution of segmented control parameters are based on the front-end acquisition segment as the spatial reference, and the equipment safety constraints and parameter linkage rules are embedded to ensure the adaptability of the instructions to the robot's real-time performance and the safety of the operation process. Through the dual-dimensional real-time detection and feedback re-optimization mechanism of cleanliness and operating parameters, the non-compliant local areas can be corrected online in a single operation, thereby maintaining a stable balance between the water consumption optimization target and the cleanliness compliance constraint in the dynamic operation environment.

[0046] Based on the same inventive concept as the water consumption control method for a photovoltaic panel cleaning robot in the foregoing embodiments, the present invention also provides a water consumption control system for a photovoltaic panel cleaning robot, such as... Figure 2 As shown, the system includes: The data acquisition module is used to simultaneously collect pollution characteristic images, real-time environmental parameters, and robot body operating parameters of the photovoltaic panel area to be cleaned. The pollution feature recognition module is used to perform feature recognition on the pollution feature image to obtain the pollution type and pollution level of the photovoltaic panel area to be cleaned; An environmental parameter processing module is used to evaluate the evaporation of the real-time environmental parameters and obtain the evaporation intensity index of the photovoltaic panel area to be cleaned. The baseline water consumption determination module is used to index the pollution type, pollution level and evaporation intensity index in a preset water consumption baseline mapping table to determine the basic unit water consumption for the current photovoltaic panel area to be cleaned. The optimization control module is used to input the basic unit water consumption, the pollution level and the robot body operating parameters into a preset real-time cleaning optimization model to obtain segmented cleaning control parameter instructions, and control the cleaning robot to perform cleaning operations accordingly.

[0047] The system described above in this invention can effectively realize a method for controlling water consumption of a photovoltaic panel cleaning robot, and the technical effects it can achieve are as described in the above embodiments, which will not be repeated here.

[0048] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of the application as defined herein, and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Thus, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for controlling water consumption in a photovoltaic panel cleaning robot, characterized in that, include: Simultaneously collect pollution characteristic images, real-time environmental parameters, and robot body operation parameters of the photovoltaic panel area to be cleaned; The pollution feature image is used to identify the pollution type and pollution level of the photovoltaic panel area to be cleaned; The evaporation intensity index of the photovoltaic panel area to be cleaned is obtained by evaluating the real-time environmental parameters. Using the pollution type, pollution level, and evaporation intensity index as index dimensions, the water consumption per unit area is determined by indexing in a pre-set water consumption baseline mapping table. The basic unit water consumption, the pollution level, and the robot's operating parameters are input into a preset real-time cleaning optimization model to obtain segmented cleaning control parameter instructions, and the cleaning robot is controlled to perform cleaning operations accordingly.

2. The water consumption control method for a photovoltaic panel cleaning robot according to claim 1, characterized in that, The robot's operating parameters include current walking speed, current brush pressure, and current water pump pressure; The real-time environmental parameters include the surface temperature of the photovoltaic panel, light intensity, ambient temperature, and wind speed.

3. The water consumption control method for a photovoltaic panel cleaning robot according to claim 2, characterized in that, The pollution types include uniform dust coverage, sand accumulation, and bird droppings.

4. The water consumption control method for a photovoltaic panel cleaning robot according to claim 3, characterized in that, The pollution feature image is subjected to feature recognition to obtain the pollution type and pollution level of the photovoltaic panel area to be cleaned, including: The pollution feature image is analyzed using a pre-trained convolutional neural network model, the classification result of the pollution type is output, and an image mask identifying the pollution area is generated. The pollution level is determined by comprehensively considering the proportion of pixels covered by the image mask, the average gray value within the area, and the texture feature parameters. The pollution level is quantified into multiple discrete levels. The higher the pollution coverage ratio, the greater the difference between the gray value and the background, or the coarser the texture, the higher the pollution level is determined.

5. The water consumption control method for a photovoltaic panel cleaning robot according to claim 4, characterized in that, The method for constructing the water use baseline mapping table includes: Under standard testing conditions, for each combination of pollution type and pollution level, and for different intervals corresponding to the evaporation intensity index, the minimum unit water consumption is determined by orthogonal experimental method to ensure that the preset basic cleanliness standard is met. Using the pollution type, pollution level, and evaporation intensity index range as input features, and the minimum unit water consumption as output label, a regression model is trained. The output of the regression model is then discretized to form the water consumption benchmark mapping table.

6. The water consumption control method for a photovoltaic panel cleaning robot according to claim 5, characterized in that, The segmented cleaning control parameter commands include the segmented travel speed of the cleaning path, the segmented water flow rate, and the segmented brush head pressure.

7. The water consumption control method for a photovoltaic panel cleaning robot according to claim 6, characterized in that, The real-time cleaning optimization model is a constrained nonlinear programming model, whose objective function is to minimize total water consumption. The constraints include: The cleanliness prediction value for each cleaning segment must be greater than or equal to a preset threshold. The cleanliness prediction value is calculated by the pollution level, water flow rate, brush head pressure and travel speed of the segment through a preset prediction sub-model.

8. The water consumption control method for a photovoltaic panel cleaning robot according to claim 7, characterized in that, The preset prediction sub-model is a machine learning model trained based on historical cleaning data. Its inputs are the contamination level of the cleaning segment, the planned water flow rate, the brush head pressure, and the travel speed. Its output is the predicted cleanliness value of the cleaning segment. The cleanliness prediction value is positively correlated with the water flow rate and the brush head pressure, and negatively correlated with the travel speed and the contamination level. Furthermore, the negative impact of the contamination level on the cleaning effect exhibits an exponential decay trend.

9. The water consumption control method for a photovoltaic panel cleaning robot according to claim 8, characterized in that, The real-time cleaning optimization model employs a particle swarm optimization algorithm. During the iterative solution process of the particle swarm optimization algorithm, the search step size of the particle swarm optimization algorithm is adjusted according to the difference between the cleanliness prediction value corresponding to the current candidate solution and the preset threshold. When the cleanliness prediction value is close to the preset threshold, the search step size is reduced to perform a fine search.

10. A water consumption control system for a photovoltaic panel cleaning robot, characterized in that, include: The data acquisition module is used to simultaneously collect pollution characteristic images, real-time environmental parameters, and robot body operating parameters of the photovoltaic panel area to be cleaned. The pollution feature recognition module is used to perform feature recognition on the pollution feature image to obtain the pollution type and pollution level of the photovoltaic panel area to be cleaned; An environmental parameter processing module is used to evaluate the evaporation of the real-time environmental parameters and obtain the evaporation intensity index of the photovoltaic panel area to be cleaned. The baseline water consumption determination module is used to index the pollution type, pollution level and evaporation intensity index in a preset water consumption baseline mapping table to determine the basic unit water consumption for the current photovoltaic panel area to be cleaned. The optimization control module is used to input the basic unit water consumption, the pollution level and the robot body operating parameters into a preset real-time cleaning optimization model to obtain segmented cleaning control parameter instructions, and control the cleaning robot to perform cleaning operations accordingly.

Images