Spraying cleaning control method of photovoltaic cleaning unmanned aerial vehicle

Abstract

The invention relates to the technical field of photovoltaic cleaning control, and discloses a spraying cleaning control method of a photovoltaic cleaning unmanned aerial vehicle. The method comprises the following steps: receiving photovoltaic panel surface state data and environmental parameters, comparing dirt grade standards to obtain dirt distribution characteristics and stubborn stain positions, and determining initial spraying parameters through a self-adaptive control algorithm in combination with the environmental parameters; after the spraying executing mechanism is driven to conduct preliminary cleaning, an image collecting device is used for obtaining surface image data, and a deviation value is calculated by comparing a cleanliness threshold value; and in combination with flight attitude data of the unmanned aerial vehicle, a fuzzy decision system is adopted to dynamically adjust initial spraying parameters to generate a real-time control instruction, an execution mechanism is controlled to perform fine cleaning until surface image data accords with a cleanliness threshold, and a final scheme is generated. According to the method, through real-time feedback and dynamic adjustment, spraying parameters are made to adapt to different dirt states and environment conditions, targeted cleaning is achieved, and the intelligence and accuracy of photovoltaic panel cleaning are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic cleaning control, and in particular to a spray cleaning control method for a photovoltaic cleaning drone. Background Art

[0002] As a key vehicle for clean energy utilization, photovoltaic power plants' power generation efficiency is closely linked to the surface cleanliness of their panels. Prolonged exposure to outdoor environments can easily lead to the accumulation of various types of contaminants, including dust, bird droppings, and oil, on the panels. This contaminant significantly reduces light transmittance and reduces the conversion efficiency of the panels. Statistics show that in dusty areas, improperly cleaned panels can reduce power generation by over 20%. Therefore, regular cleaning is essential to maintaining efficient operation of photovoltaic power plants.

[0003] Photovoltaic panel cleaning primarily relies on manual cleaning, robotic cleaning, and conventional drone cleaning. While manual cleaning allows for direct contact with the panel surface, the tens of thousands of panels in large-scale photovoltaic power plants make cleaning time-consuming and labor-intensive. Furthermore, working at height poses a risk of falling. Furthermore, manual cleaning quality is significantly affected by operator experience, making it prone to missed cleaning spots and uneven cleaning.

[0004] Robotic cleaning equipment is typically fixed on a specific track and is suitable for flat-surface photovoltaic arrays. However, its range of motion is limited, making it less adaptable to photovoltaic panels installed on complex terrain, such as mountains and rooftops. Furthermore, the rigid contact of the robotic arm can scratch the surface of the photovoltaic panels, especially thin-film photovoltaic modules. Such damage can directly affect their service life.

[0005] Conventional drone cleaning technology has been gradually adopted in the photovoltaic cleaning field in recent years, but its control method often uses a preset parameter mode, that is, it operates according to a fixed spray pressure, flow rate, and angle. In actual application, the dirt types on the surface of photovoltaic panels vary greatly, ranging from loose dust to hardened stains formed by rain. The cleaning force required for different dirt types varies significantly. At the same time, environmental parameters such as wind speed, temperature, and light can affect the spraying effect. When the wind speed is high, the spray water flow is easily blown away and cannot accurately affect the dirt location. When the temperature is too high, the water evaporates too quickly, which may cause detergent residue.

[0006] Existing drone cleaning methods lack the ability to perceive and respond to real-time surface conditions. Fixed parameters make it difficult to completely remove stubborn stains, while overspraying even lightly contaminated areas can waste water and energy. Furthermore, during flight, drones can drift due to airflow disturbances, causing the spray position to misalign with the target area, further impacting cleaning effectiveness.

[0007] Some traditional cleaning control methods use a timed cleaning mode, ignoring the variations in the actual contamination levels of photovoltaic panels and resulting in inappropriate cleaning timing. Others, while incorporating simple sensor feedback, only adjust a single parameter and fail to comprehensively consider the relationship between dirt distribution, environmental changes, and the drone's own state, making it difficult to achieve truly precise cleaning. These issues collectively restrict the efficiency and economic viability of photovoltaic cleaning, necessitating an intelligent control method that can dynamically adapt to complex operating conditions. Summary of the Invention

[0008] The purpose of the present invention is to provide a spray cleaning control method for a photovoltaic cleaning drone to solve the problems raised in the above background technology.

[0009] To achieve the above objectives, the present invention provides a method for controlling spray cleaning of a photovoltaic cleaning drone, the method comprising:

[0010] receiving surface condition data from a photovoltaic panel surface monitoring device and environmental parameters collected by an environmental sensor, comparing the surface condition data with a preset dirt grade standard, and obtaining dirt distribution characteristics and stubborn stain locations in the surface condition data;

[0011] Determining the initial spraying parameters of the UAV using an adaptive control algorithm according to the dirt distribution characteristics, the location of the stubborn stains, and the environmental parameters;

[0012] Using the initial spray parameters to drive the spray actuator to perform preliminary cleaning, while acquiring surface image data after cleaning through an image acquisition device;

[0013] Comparing the surface image data with a preset cleanliness threshold, and calculating a deviation between the actual cleanliness and the cleanliness threshold;

[0014] According to the deviation value, combined with the flight attitude data of the UAV, a fuzzy decision system is used to dynamically adjust the initial spray parameters to generate real-time spray control instructions;

[0015] According to the real-time spray control instruction, the spray actuator is controlled to perform a fine cleaning operation until the surface image data meets the cleanliness threshold, and a final spray control plan is generated.

[0016] Preferably, the method of determining the initial spraying parameters of the UAV using an adaptive control algorithm according to the dirt distribution characteristics, the location of the stubborn stains, and the environmental parameters includes:

[0017] Gridding the dirt distribution characteristics and the stubborn stain positions to obtain a dirt density value for each grid area;

[0018] Normalizing the wind speed and temperature data in the environmental parameters to generate environmental impact factors;

[0019] Inputting the dirt density value and the environmental impact factor into a parameter model of an adaptive control algorithm, wherein the parameter model is trained using historical cleaning data and outputs an initial spray pressure and water flow rate;

[0020] Combined with the current flight altitude and horizontal movement speed of the UAV, the spray angle and flow rate corresponding to the initial spray pressure and the water flow velocity are calculated as the initial spray parameters.

[0021] Preferably, the method of dynamically adjusting the initial spray parameters using a fuzzy decision system based on the deviation value and the flight attitude data of the UAV to generate a real-time spray control instruction includes:

[0022] Dividing the deviation value into a plurality of fuzzy subsets, each fuzzy subset corresponding to a different deviation level;

[0023] Collecting the flight attitude data of the UAV, including pitch angle, roll angle and yaw angle, and filtering the flight attitude data to obtain stable attitude parameters;

[0024] Inputting the deviation level and the stable posture parameter into a fuzzy decision system, wherein the fuzzy decision system includes a preset fuzzy rule library and outputs a spray parameter adjustment amount;

[0025] According to the spray parameter adjustment amount, the spray angle and flow rate in the initial spray parameters are corrected to generate the real-time spray control instruction.

[0026] Preferably, the step of acquiring surface image data after cleaning by an image acquisition device, comparing the surface image data with a preset cleanliness threshold, and calculating a deviation between the actual cleanliness and the cleanliness threshold comprises:

[0027] grayscale the surface image data to extract brightness features and texture features in the image;

[0028] Inputting the brightness feature and the texture feature into a preset cleanliness assessment model and outputting an actual cleanliness value;

[0029] Obtaining a preset cleanliness threshold, and calculating the difference between the actual cleanliness value and the cleanliness threshold as the deviation value;

[0030] When the deviation value is greater than a preset allowable range, it is marked as an area requiring secondary cleaning, and the position coordinates of the area are recorded.

[0031] Preferably, the method of determining the initial spray parameters of the UAV by using an adaptive control algorithm includes:

[0032] Collect surface condition data, environmental parameters, spray parameters, and corresponding cleaning effect data from historical cleaning operations to construct a training dataset;

[0033] Standardizing the training data set and dividing it into a training set and a validation set;

[0034] The training set is used to train a parameter model of the adaptive control algorithm, wherein the parameter model adopts a multi-layer perceptron structure, the input layer is the dirt distribution characteristics and environmental influencing factors, and the output layer is the spray pressure and water flow rate;

[0035] The performance of the trained parameter model is evaluated using the validation set. When the evaluation index reaches a preset threshold, the training is stopped and the optimal parameters of the parameter model are saved.

[0036] Preferably, the fuzzy decision system includes a preset fuzzy rule base, and the method for constructing the fuzzy rule base includes:

[0037] Collect expert experience and historical cleaning cases to extract the mapping relationship between deviation level, flight attitude parameters and spray parameter adjustment amount;

[0038] Converting the mapping relationship into fuzzy rules, each fuzzy rule includes a premise and a conclusion part, the premise is a fuzzy subset of the deviation level and the flight attitude parameter, and the conclusion is a fuzzy subset of the spray parameter adjustment amount;

[0039] Performing conflict detection and merging processing on the fuzzy rules, removing redundant rules, and generating an initial fuzzy rule base;

[0040] The initial fuzzy rule base is iteratively optimized through actual cleaning operation data, and the membership function parameters of the rules are adjusted to obtain an optimized fuzzy rule base.

[0041] Preferably, controlling the spray actuator to perform a fine cleaning operation according to the real-time spray control instruction includes:

[0042] Parsing the real-time spray control instruction to obtain the target spray angle, target flow rate and execution time;

[0043] Converting the target spray angle into a steering gear angle of a spray actuator, and converting the target flow rate into a speed control signal of a water pump;

[0044] driving the steering gear and the water pump to operate according to the steering gear rotation angle and the speed control signal to perform fine cleaning;

[0045] During the cleaning process, the working status of the steering gear and water pump is monitored in real time. When any abnormality occurs, the operation is stopped immediately and an alarm signal is issued.

[0046] Preferably, the receiving of surface state data from a photovoltaic panel surface monitoring device and environmental parameters collected by an environmental sensor includes:

[0047] Performing time series segmentation on the surface state data, dividing the continuously collected data into multiple data windows;

[0048] The Kalman filter algorithm is used to perform noise suppression on each data window to obtain the denoised surface state data;

[0049] Normalizing the humidity and light intensity data in the environmental parameters to generate an environmental correction coefficient;

[0050] The denoised surface state data is fused with the environmental correction coefficient to obtain fused surface state features.

[0051] Preferably, the using the training set to train the parameter model of the adaptive control algorithm includes:

[0052] Dividing the environmental parameters in the training set into a plurality of environmental scene categories, each category corresponding to a different climatic condition;

[0053] For each environmental scenario category, a sub-model of the parameter model is trained separately to obtain multiple scenario-based sub-models;

[0054] The output results of the multiple scenario-based sub-models are integrated through a voting mechanism to generate a comprehensive spray parameter prediction value;

[0055] The error between the predicted value of the comprehensive spray parameter and the actual cleaning effect data is calculated, and the weight parameters of each sub-model are adjusted by back propagation.

[0056] Preferably, the real-time monitoring of the working status of the steering gear and the water pump includes:

[0057] Collect the deviation between the actual rotation angle of the servo and the target rotation angle, as well as the deviation between the actual speed of the water pump and the target speed;

[0058] When the deviation value continuously exceeds the preset threshold value for a first preset time period, it is determined to be a minor fault and PID adjustment is started for compensation;

[0059] When the deviation value exceeds the emergency threshold or continuously exceeds the preset threshold for a second preset time period, it is determined to be a serious fault, and shutdown protection is immediately triggered and a fault code is recorded.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] This PV-cleaning drone spray cleaning control method improves upon the shortcomings of existing PV cleaning technologies through multi-dimensional data fusion and a dynamic adjustment mechanism. Its core advantage lies in its adaptability to complex working conditions. It can flexibly adjust cleaning strategies based on the actual surface condition of the PV panels, breaking the limitations of fixed parameters in traditional cleaning methods.

[0062] After receiving surface condition data and environmental parameters, the system determines dirt distribution characteristics and the location of stubborn stains by comparing them to dirt grade standards. This allows initial spray parameter settings to be based on a precise understanding of the pollution situation, rather than relying on empirical values. This parameter configuration based on actual pollution data avoids the under- or over-cleaning that can occur with traditional methods due to inaccurate judgment of dirt types. For sparsely distributed light dust, the system automatically reduces spray pressure and flow; for concentrated stubborn stains, the system specifically increases parameters to ensure effective removal.

[0063] The image feedback mechanism after initial cleaning provides an intuitive basis for evaluating cleaning effectiveness. By comparing the cleaned surface image data with a preset cleanliness threshold, the deviation between the actual cleanliness and the target value can be quantified. This quantitative analysis replaces traditional subjective judgment and makes the evaluation of cleaning effectiveness more objective. The introduction of deviation values ​​provides a clear direction for subsequent parameter adjustments, avoiding the waste of resources caused by blind adjustments.

[0064] The fuzzy decision-making system, combined with the drone's flight attitude data, further improves control accuracy. When operating at high altitudes, drones are subject to fluctuations in flight attitude due to factors such as airflow and wind speed. If the spraying parameters remain fixed, this can easily lead to offset spraying positions or uneven spraying force. This method collects flight attitude data in real time and incorporates it into a parameter adjustment model, allowing parameters such as spray angle and distance to be adjusted synchronously with changes in the drone's attitude, ensuring that the spraying point always accurately covers the target area. This dynamic adaptability can significantly reduce missed areas, especially when dealing with large-scale photovoltaic arrays.

[0065] This method demonstrates greater targeted effectiveness in treating stubborn stains. Traditional cleaning methods often require multiple, repeated cleanings, which is not only time-consuming but also increases water consumption. This method, however, locates stubborn stains in the initial stage. After the initial cleaning, combined with image feedback, it dynamically increases the spray intensity or adjusts the spray angle for any remaining stains. This allows for a single, detailed clean to remove them, reducing the frequency of repetitive cleaning.

[0066] The inclusion of environmental parameters makes the cleaning process more adaptable. Under different environmental conditions, the evaporation rate of water and the diffusion range of the spray vary. For example, in a high-temperature environment, if the spray is carried out according to normal temperature parameters, the rapid evaporation of water may cause detergent residue on the surface; in a windy environment, spraying at a fixed angle will cause the water flow to deviate from the target due to the influence of airflow. This method takes environmental factors into account in both initial parameter setting and dynamic adjustment. For example, in high temperatures, the flow rate is appropriately increased to delay evaporation, and in strong winds, the spray angle is adjusted to offset the influence of airflow, so that the cleaning effect remains stable in various environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a working principle diagram of the spray cleaning control method of the photovoltaic cleaning drone described in the present invention;

[0068] Figure 2 Flow chart of the method for determining initial spray parameters;

[0069] Figure 3 Flowchart for surface image data comparison and deviation value calculation;

[0070] Figure 4 This is a flow chart of the fine cleaning control of the spray actuator. DETAILED DESCRIPTION

[0071] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0072] See also Figure 1 The present invention provides a method for controlling the spray cleaning of photovoltaic panels by a drone. The method combines surface condition monitoring, environmental parameter analysis, adaptive control, and dynamic adjustment mechanisms to achieve efficient and accurate cleaning of photovoltaic panels. The specific implementation steps are as follows:

[0073] Receive surface status data from the photovoltaic panel surface monitoring device and environmental parameters collected by the environmental sensor, compare the surface status data with the preset dirt grade standard, and obtain the dirt distribution characteristics and stubborn stain locations in the surface status data. Among them, the surface status data covers the dirt coverage rate and stain type (such as dust, bird droppings, oil, etc.) on the photovoltaic panel surface, and the environmental parameters include wind speed, temperature, humidity, light intensity, etc. The dirt grade standard is divided according to the preset dirt coverage rate range and the stubbornness of the stain. By comparison, the dirt distribution pattern in different areas of the photovoltaic panel surface (such as the dirt density in the edge area is higher than that in the center area) and the specific coordinates of stubborn stains (such as long-term accumulated hardened stains) can be determined.

[0074] Based on the dirt distribution characteristics, the location of stubborn stains, and the environmental parameters, an adaptive control algorithm is used to determine the drone's initial spray parameters. These parameters include spray pressure, water velocity, spray angle, and flow rate. These parameters are determined by comprehensively considering the dirt distribution density (distribution characteristics), the difficulty of cleaning stubborn stains (location), and the impact of environmental factors (such as high wind speeds that may cause water flow deviation). The adaptive control algorithm dynamically outputs the appropriate initial parameters based on these factors.

[0075] The initial spray parameters are used to drive the spray actuator for preliminary cleaning, while the image acquisition device simultaneously captures image data of the surface after cleaning. The spray actuator operates according to the initial parameters to complete the initial cleaning of the photovoltaic panel surface. After cleaning, the image acquisition device (such as a high-definition camera) captures the surface and captures image data containing information about the cleaning effect. The image must cover the entire area of ​​the photovoltaic panel to ensure that no area is missed.

[0076] The surface image data is compared with a preset cleanliness threshold, and the deviation between the actual cleanliness and the threshold is calculated. The cleanliness threshold is a preset standard for the cleanliness of the photovoltaic panel surface (e.g., dirt coverage less than 1%). The actual cleanliness is obtained by analyzing the image data (e.g., calculating the residual dirt area percentage). The difference between the two is the deviation value. A positive deviation value indicates that the cleanliness standard has not been met and further cleaning is required.

[0077] Based on the deviation value and the drone's flight attitude data, a fuzzy decision-making system dynamically adjusts the initial spray parameters and generates real-time spray control instructions. Flight attitude data includes the drone's pitch, roll, and yaw angles, reflecting its flight stability. The fuzzy decision-making system adjusts the initial spray parameters (e.g., increasing spray pressure and adjusting spray angle to accommodate changes in the drone's attitude) based on the deviation value (e.g., a large deviation indicates poor cleaning effectiveness) and flight attitude stability (e.g., large attitude fluctuations may affect spray accuracy) to generate real-time control instructions.

[0078] Based on the real-time spray control instructions, the spray actuator is controlled to perform fine cleaning operations until the surface image data meets the cleanliness threshold, generating a final spray control plan. The spray actuator performs fine cleaning operations two or more times according to the real-time control instructions. After each cleaning, the image acquisition device acquires surface image data and compares it with the cleanliness threshold until the image data meets the threshold requirement. At this point, the spray parameters and control logic used throughout the cleaning process are summarized to form the final spray control plan.

[0079] Example 1: See Figure 2When receiving surface condition data from the photovoltaic panel surface monitoring device and environmental parameters collected by environmental sensors, the system first performs time series segmentation on the surface condition data. This surface condition data is collected continuously and covers the contamination status of the photovoltaic panel surface at different times, including information such as the coverage, distribution density, and type of contamination. Time series segmentation divides this continuous data into multiple independent data windows at regular intervals. Each data window contains complete surface condition information for a specific time period.

[0080] A Kalman filter algorithm is used to suppress noise in each partitioned data window. During data acquisition, due to sensor accuracy limitations or environmental interference, surface condition data may be contaminated with irrelevant noise, which can affect the assessment of the panel's actual surface condition. The Kalman filter algorithm, by establishing a dynamic model and continuously predicting and correcting the data, effectively filters out this noise, producing more accurate, denoised surface condition data that more accurately reflects the panel's surface contamination.

[0081] For the environmental parameters collected by environmental sensors, the humidity and light intensity data need to be normalized to generate an environmental correction coefficient. The numerical ranges of humidity and light intensity vary greatly, and the units are different. Directly using these raw data for analysis will have an adverse effect on the results. Normalization is to convert these data into a unified numerical range to eliminate the interference caused by different units and numerical ranges. The environmental correction coefficient obtained after this processing can reflect the degree of influence of humidity and light intensity on the surface state of photovoltaic panels. For example, high humidity may cause dirt to adhere more tightly to the surface of photovoltaic panels, while strong light may affect the monitoring effect of surface conditions.

[0082] The denoised surface state data is then fused with the generated environmental correction coefficients to produce a fused surface state signature. This fusion process comprehensively considers the actual fouling conditions on the photovoltaic panel surface as well as the influence of environmental factors, making the resulting surface state signature more comprehensive and accurate, providing a reliable basis for subsequent determination of initial spray parameters.

[0083] When using an adaptive control algorithm to determine the drone's initial spray parameters based on dirt distribution characteristics, stubborn stain locations, and environmental parameters, the system first performs a gridding process to determine the dirt distribution characteristics and stubborn stain locations. Gridding involves dividing the entire photovoltaic panel surface into several equally sized grid regions, each corresponding to a specific portion of the panel surface. The amount and distribution of dirt within each grid region are then counted, and the dirt density value for each grid region is calculated. This dirt density value provides a clear understanding of the severity of dirt in each grid region.

[0084] Normalizing the wind speed and temperature data within the environmental parameters generates an environmental impact factor. Wind speed and temperature are crucial environmental factors influencing spray cleaning effectiveness. Excessive wind speed can cause the spray flow to deviate from its intended direction, while excessively high temperatures can accelerate water evaporation, affecting cleaning effectiveness. Normalization also aims to eliminate differences in numerical values ​​and units between the two, converting them to a unified range. The resulting environmental impact factor quantifies the combined impact of wind speed and temperature on the cleaning process.

[0085] The dirt density and environmental factors for each grid area are input into the adaptive control algorithm's parameter model. This parameter model is trained using extensive historical cleaning data, which contains information on optimal spray parameters and cleaning results for different dirt densities and environmental factors. Based on the input dirt density and environmental factors, the parameter model automatically outputs the appropriate initial spray pressure and water flow rate for the situation.

[0086] The initial spray parameters are calculated based on the drone's current flight altitude and horizontal speed, corresponding to the spray angle and flow rate. The drone's flight altitude determines the coverage and impact of the spray water flow upon reaching the photovoltaic panel surface, while the horizontal speed influences the duration of the spray on each area. By comprehensively considering these factors, the initial spray pressure and flow rate are converted into specific spray angles and flow rates, enabling the spray actuator to perform initial cleaning according to these parameters, ensuring optimal initial cleaning results under various flight conditions.

[0087] The entire process, from data collection and processing to the calculation of initial spray parameters, is closely integrated with the actual conditions and environmental factors of the photovoltaic panels. This makes the determined initial spray parameters highly targeted and adaptable, laying a good foundation for subsequent cleaning operations.

[0088] Example 2: See Figure 3 , the surface image data after cleaning is obtained through an image acquisition device, and the surface image data is compared with a preset cleanliness threshold. When calculating the deviation between the actual cleanliness and the cleanliness threshold, the surface image data is first grayscaled. The surface image of the photovoltaic panel captured by the image acquisition device is usually a color image, containing a large amount of redundant information. Grayscale processing can convert the color image into a grayscale image, retaining the brightness information of the image, while reducing the amount of data, facilitating subsequent feature extraction and analysis. During the grayscale processing, a specific algorithm is used to convert the pixel values ​​of the red, green, and blue channels into a single grayscale value according to a certain ratio, so that each pixel in the image is represented by only one grayscale value, simplifying the data structure of the image.

[0089] After grayscaling, brightness and texture features are extracted from the image. The brightness feature reflects the reflectivity of the photovoltaic panel surface. Clean areas, due to their smooth surface, have strong reflectivity and high brightness values; areas with dirt have weak reflectivity and low brightness values. By calculating parameters such as the average brightness value and brightness variance for each region in the image, the brightness feature, which reflects the surface cleanliness, can be obtained. The texture feature reflects the roughness of the photovoltaic panel surface. Clean areas have a relatively uniform and smooth texture, while areas with dirt have irregular texture variations. Texture features can be extracted using methods such as gray-level co-occurrence matrices and entropy. These features can further distinguish clean from dirty areas.

[0090] The extracted brightness and texture features are input into a pre-set cleanliness assessment model, which outputs the actual cleanliness value. The cleanliness assessment model is trained based on a large amount of historical data. This data includes the brightness and texture features of images of photovoltaic panel surfaces with different levels of cleanliness, as well as the corresponding actual cleanliness (determined through manual inspection or other precise methods). By learning the patterns in this data, the model can accurately predict the actual cleanliness value of the photovoltaic panel surface based on the input features. This value is usually expressed as a specific number, with higher numbers indicating cleaner surfaces.

[0091] Obtain a preset cleanliness threshold. This threshold is set based on the normal operating requirements and cleaning standards of the PV panel and represents the required cleanliness level of the panel surface. Calculate the difference between the actual cleanliness value and the cleanliness threshold; this difference is the deviation value. If the deviation value exceeds the preset allowable range, the cleanliness level of the area does not meet the requirements and requires a second cleaning. The area requiring a second cleaning is marked and its coordinates are recorded. The coordinates are converted from the corresponding relationship between the pixel coordinates of the image and the actual position of the PV panel, allowing for precise positioning of subsequent cleaning operations.

[0092] Based on the deviation value and combined with the drone's flight attitude data, a fuzzy decision-making system dynamically adjusts the initial spray parameters. When generating real-time spray control instructions, the deviation value is first divided into multiple fuzzy subsets, each corresponding to a different deviation level. The magnitude of the deviation value reflects the gap between the actual cleanliness and the target cleanliness. By dividing the fuzzy subsets, the continuous deviation value can be discretized into different levels, such as "negative large," "negative medium," "negative small," "zero," "positive small," "positive medium," and "positive large." Each level corresponds to a certain range of deviation values. This division method facilitates the incorporation of deviation values ​​into the fuzzy decision-making system for processing.

[0093] At the same time, the drone's flight attitude data is collected, including pitch, roll, and yaw angles. The pitch angle indicates the drone's tilt in the front-to-back direction, the roll angle indicates the drone's tilt in the left-to-right direction, and the yaw angle indicates the drone's rotation angle around the vertical axis. These data can reflect the drone's flight stability. Changes in the drone's flight attitude can affect the accuracy of the spray. For example, a pitch angle that is too large may cause the spray direction to shift forward or backward, affecting the cleaning effect. Since the flight attitude data may be affected by factors such as drone vibration and airflow interference during the collection process, there is a certain amount of noise, so it needs to be filtered. The filtering process can use methods such as mean filtering and Kalman filtering to eliminate noise in the data and obtain stable attitude parameters, so that the attitude data can more truly reflect the actual flight attitude of the drone.

[0094] The deviation level and stable attitude parameters are input into a fuzzy decision-making system, which contains a pre-set fuzzy rule base and outputs spray parameter adjustments. This fuzzy rule base, built based on expert experience in the cleaning field and extensive historical cleaning cases, contains a series of "if-then" rules, such as "If the deviation level is negative and the pitch angle is positive, then the spray pressure adjustment is positive and large," and "If the deviation level is negative and the roll angle is negative and small, then the spray pressure adjustment is positive and small." These rules describe how to adjust spray parameters under different deviation levels and flight attitudes. The fuzzy decision-making system fuzzifies the input deviation level and stable attitude parameters to match them with the fuzzy conditions in the rule base. It then performs inference based on the matched rules and finally defuzzifies the inference results into specific spray parameter adjustments, such as spray pressure adjustments, water flow rate adjustments, and spray angle adjustments.

[0095] Based on the obtained spray parameter adjustment amount, the spray angle and flow rate in the initial spray parameters are corrected to generate real-time spray control instructions. The initial spray parameters are determined before the cleaning operation begins and may not be fully adapted to the actual cleaning situation. Through correction, the spray parameters can be made more in line with the current cleanliness requirements and the flight status of the drone. For example, if the spray angle adjustment amount is +3 degrees and the initial spray angle is 25 degrees, the corrected spray angle is 28 degrees; if the flow rate adjustment amount is -0.5L / min and the initial flow rate is 3L / min, the corrected flow rate is 2.5L / min. The corrected spray pressure, water flow rate, spray angle, flow rate and other parameters are integrated together to form a real-time spray control instruction, which will be transmitted to the spray actuator to control the actuator to perform fine cleaning operations.

[0096] Example 3: When using an adaptive control algorithm to determine the drone's initial spray parameters, it is necessary to first collect various data from historical cleaning operations to construct a training dataset. This data includes the surface condition of the photovoltaic panels, such as the type, coverage, and thickness of dirt in different areas; environmental parameters such as wind speed, temperature, humidity, and light intensity during cleaning; actual spray parameters used, such as spray pressure, water velocity, angle, and flow rate; and corresponding cleaning effect data, namely, the cleanliness of the photovoltaic panel surface after cleaning. During the data collection process, it is necessary to ensure that the data covers scenarios of different seasons, different weather conditions, and different levels of dirt to ensure the comprehensiveness and representativeness of the dataset and provide sufficient samples for subsequent model training.

[0097] The constructed training dataset is standardized and then divided into training and validation sets. The purpose of standardization is to eliminate the effects of differences in dimensions and numerical ranges between different data, bringing all data to the same order of magnitude. During standardization, specific methods are used to transform the data into a distribution with a mean of 0 and a standard deviation of 1, making different types of data comparable. After standardization, the dataset is divided into training and validation sets according to a specific ratio. The training set is used to learn the model's parameters, while the validation set is used to evaluate the model's training effectiveness and avoid overfitting or underfitting.

[0098] When using a training set to train the parameter model of the adaptive control algorithm, the environmental parameters in the training set are first divided into multiple environmental scenario categories, each corresponding to different climatic conditions. For example, wind speed can be divided into low wind speed (such as 0-3m / s), medium wind speed (such as 3-6m / s), and high wind speed (such as above 6m / s); based on the temperature range, it can be divided into low temperature (such as 0-15℃), normal temperature (such as 15-30℃), and high temperature (such as above 30℃). Then, combined with factors such as humidity and light, multiple environmental scenarios are formed, such as "low wind speed-normal temperature-medium humidity" and "high wind speed-high temperature-low humidity". Each scenario category contains multiple sets of relevant data under the climatic conditions.

[0099] For each environmental scenario category, a sub-model of the parameter model is trained separately, resulting in multiple scenario-specific sub-models. Each sub-model utilizes a multi-layer perceptron structure consisting of an input layer, a hidden layer, and an output layer. The input layer receives dirt distribution characteristics and environmental influencing factors, the hidden layer processes data and extracts features using multiple neurons, and the output layer outputs spray pressure and water flow rate. During training, each sub-model uses only data from the corresponding environmental scenario category. By continuously adjusting the connection weights between neurons in each layer, the sub-model learns the mapping between dirt conditions and spray parameters in that scenario.

[0100] The voting mechanism fuses the outputs of multiple scenario-based sub-models to generate a comprehensive sprinkler parameter prediction value. The voting mechanism assigns different weights to each sub-model based on how well it matches the current environmental scenario. The higher the match, the greater the weight. For example, if the current environment is "high wind speed, high temperature, and low humidity," the sub-model corresponding to this scenario will have a higher weight than other sub-models. The formula for calculating the comprehensive sprinkler parameter prediction value is:

[0101]

[0102] Where P represents the predicted value of comprehensive spray parameters, n represents the number of sub-models, and w i represents the weight of the i-th sub-model, p i Represents the output value of the i-th sub-model.

[0103] The error between the predicted comprehensive spray parameters and the actual cleaning performance data is calculated, and the weight parameters of each sub-model are adjusted using a back-propagation algorithm. Error calculation typically uses metrics such as mean squared error to reflect the degree of deviation between the predicted and actual values. The back-propagation algorithm propagates the error from the output layer to the input layer, adjusting the connection weights of neurons in each layer and the weights of the sub-models based on the error to bring the model's predicted values ​​closer to the actual cleaning performance data. This process is repeated until the model's prediction error is reduced to a preset range. At this point, the resulting model can more accurately output the appropriate spray pressure and water flow rate based on the input dirt distribution characteristics and environmental factors.

[0104] After training, the parameter model can quickly output the corresponding initial spray pressure and water flow rate in actual applications when new dirt distribution characteristics and environmental influencing factors are input. Combined with the flight altitude and horizontal movement speed of the drone, the spray angle and flow rate are calculated to form complete initial spray parameters, providing parameter support for the initial cleaning operation of the photovoltaic cleaning drone.

[0105] Example 4: The construction of the fuzzy rule base in the fuzzy decision-making system needs to start with the collection of expert experience knowledge and historical cleaning cases. The expert experience knowledge comes from technicians who have been engaged in photovoltaic panel cleaning for a long time. Based on practice, they have summarized methods for adjusting spray parameters in different situations. For example, when there are large areas of stubborn stains on the surface of the photovoltaic panel and the flight attitude of the drone is unstable, it is necessary to increase the spray pressure and adjust the angle to ensure the cleaning effect; for local minor dirt, only fine-tuning of the water flow rate is required. Historical cleaning cases cover complete cleaning process records under different environmental conditions and different dirt states, including the deviation level of each cleaning (the difference between the actual cleanliness and the threshold), the flight attitude parameters of the drone (such as the specific values ​​of the pitch angle and roll angle), the spray parameter adjustment measures adopted, and the final cleaning results. By sorting out these cases, it can be found which spray parameter adjustment methods are more effective under specific deviation level and flight attitude combinations.

[0106] The mapping relationships between deviation levels, flight attitude parameters, and spray parameter adjustments are extracted from collected expert experience and historical cases. These mapping relationships are the basis of the rule base. For example, "When the deviation level is significantly low (i.e., the actual cleanliness is far below the threshold) and the pitch angle of the drone is large, it is necessary to significantly increase the spray pressure and appropriately increase the spray angle" and "When the deviation level is slightly low and the roll angle is small, it is only necessary to slightly increase the water flow rate." These relationships require clear corresponding logic between input (deviation level, flight attitude) and output (spray parameter adjustment) to ensure that the corresponding adjustment strategy can be found for each specific scenario.

[0107] The extracted mapping relationships are converted into fuzzy rules, each of which contains premises and conclusions. The premises are composed of fuzzy subsets of deviation levels and flight attitude parameters, while the conclusions are fuzzy subsets of spray parameter adjustments. Fuzzy subsets are described by linguistic variables. For example, the fuzzy subsets of deviation levels can be divided into "significantly low," "slightly low," "basically up to standard," "slightly high," and "significantly high"; the fuzzy subsets of flight attitude parameters can be divided into "angle too large," "moderate angle," and "angle too small"; and the fuzzy subsets of spray parameter adjustments can be divided into "substantially increase," "slightly increase," "remain unchanged," "slightly decrease," and "substantially decrease." A typical fuzzy rule might be: "If the deviation level is significantly low and the pitch angle is too large, the spray pressure adjustment amount is a significant increase, and the spray angle adjustment amount is a slight decrease." This expression method converts actual data into a language description that is easy to understand and apply, which conforms to the characteristics of fuzzy decision-making.

[0108] The constructed fuzzy rules are subjected to conflict detection and merging to remove redundant rules and form an initial fuzzy rule base. Conflict detection mainly checks whether there are rules with the same premise but contradictory conclusions. For example, the premise of two rules is "the deviation level is slightly low and the roll angle is moderate", but one conclusion is "the water flow speed increases slightly" and the other is "the water flow speed decreases slightly". This situation requires verification based on expert experience and historical cases to retain more effective rules. Redundant rules refer to rules that can be covered by other rules. For example, there is a rule that says "if the deviation level is significantly low, the spray pressure increases significantly", while another rule says "if the deviation level is significantly low and the yaw angle is moderate, the spray pressure increases significantly". The latter has more specific premises, but the conclusion is consistent with the former. It is a redundant rule and can be deleted to simplify the rule base structure and improve decision-making efficiency.

[0109] The initial fuzzy rule base is iteratively optimized using actual cleaning operation data, and the membership function parameters of the rules are adjusted to obtain the optimized fuzzy rule base. The membership function quantifies the degree to which a variable belongs to a fuzzy subset. For example, a certain actual deviation value may be 70% "significantly low" and 30% "slightly low." In actual application, if the cleaning effect does not meet expectations after adjusting the parameters according to the initial rules (for example, the deviation value still exceeds the allowable range), the cause needs to be analyzed. It may be that the division of the membership function is unreasonable, resulting in inaccurate rule matching. For example, a deviation value originally judged as "slightly low" should actually be closer to "significantly low". In this case, the boundary value of the membership function needs to be adjusted to make the fuzzy classification of the variable more accurate. This optimization process needs to be repeated. After each cleaning operation, the actual data is compared with the output of the rule base, and the membership function parameters are continuously adjusted until the rule base can output effective spray parameter adjustments in most cases to adapt to different cleaning scenarios and equipment conditions.

[0110] Example 5: See Figure 4 , when controlling the spray actuator to perform fine cleaning operations according to the real-time spray control instructions, the real-time spray control instructions must be parsed first. The real-time spray control instructions are digital signals containing multiple parameters. The parsing process needs to convert these signals into specific executable parameters, including target spray angle, target flow rate and execution time. The target spray angle determines the direction of the sprinkler head, ensuring that the water flow can accurately cover the area that needs to be cleaned; the target flow rate specifies the amount of water sprayed per unit time, which directly affects the cleaning intensity; and the execution time clarifies the duration of the spray operation under this set of parameters. After the parsing is completed, these parameters will serve as the basis for subsequent control of the spray actuator action.

[0111] The target spray angle is converted into the servo angle of the sprinkler actuator. The servo of the sprinkler actuator is responsible for adjusting the direction of the sprinkler head. A preset correspondence exists between the target spray angle and the servo angle, which allows the specific angle the servo needs to rotate to be calculated. For example, if the target spray angle is 30 degrees, the correspondence determines that the servo needs to rotate to 65 degrees to ensure that the sprinkler head can accurately align with the target angle. Simultaneously, the target flow rate is converted into a speed control signal for the water pump. The speed of the water pump directly affects the water output. The target flow rate and the pump speed are linked via a pre-calibrated curve. This curve can be used to convert the target flow rate into a corresponding speed value, which is then used to generate a corresponding electrical signal to control the rotation speed of the water pump motor.

[0112] The servo and water pump are driven according to the calculated servo angle and speed control signals to perform precision cleaning. Upon receiving the angle signal, the servo rotates to the target angle via its internal motor and transmission structure, driving the sprinkler head to adjust to the set spray angle. Driven by the speed control signal, the water pump's motor runs at the specified speed, discharging water from the sprinkler head at the target flow rate. Working together, the two systems perform targeted cleaning of the photovoltaic panel surface according to real-time spray control commands, ensuring adequate water coverage and cleaning intensity for key areas requiring treatment.

[0113] During the cleaning process, the operating status of the servo and water pump are monitored in real time. Sensors collect the actual servo angle and water pump speed, compare these actual values ​​with the target values, and calculate the deviation. For example, if the servo's target angle is 65 degrees and the actual angle is 63 degrees, the deviation is -2 degrees; if the water pump's target speed is 1500 rpm and the actual speed is 1480 rpm, the deviation is -20 rpm. These deviations reflect the gap between the actuator's actual action and the commanded speed.

[0114] When the deviation value continuously exceeds the preset threshold for a first preset duration, it is determined to be a minor fault and PID adjustment is initiated for compensation. The preset threshold is set according to the normal operating accuracy of the equipment, and the first preset duration is determined according to the actual working conditions, usually a few seconds. For example, if the servo angle deviation continuously exceeds ±3 degrees for 3 seconds, or the water pump speed deviation continuously exceeds ±50 rpm for 3 seconds, it is determined to be a minor fault. PID adjustment generates a compensation signal through calculations of the three links of proportional, integral, and differential, and adjusts the servo drive current or the water pump motor voltage to bring the actual angle or speed closer to the target value, eliminating the deviation and restoring normal working conditions.

[0115] When the deviation value exceeds the emergency threshold or continues to exceed the preset threshold for a second preset time, it is determined to be a serious fault, and the shutdown protection is immediately triggered and the fault code is recorded. The emergency threshold is a safety limit that is much larger than the preset threshold, such as the servo angle deviation exceeds ±10 degrees, or the water pump speed deviation exceeds ±300 rpm; the second preset time is usually longer than the first preset time, such as 10 seconds. When a serious fault occurs, continuing the operation may cause damage to the equipment or serious deterioration of the cleaning effect. Therefore, it is necessary to stop the servo and water pump immediately, and keep the drone hovering or returning to avoid dangers such as collisions. The fault code contains information such as the type of fault (such as servo jam, water pump blockage), the time of occurrence, etc., and is stored in the device's memory to facilitate subsequent maintenance personnel to query and repair.

[0116] During the entire fine cleaning process, the above-mentioned monitoring and troubleshooting steps are carried out continuously to ensure that abnormal conditions of the servo and water pump are discovered and handled in a timely manner, and to ensure the safety and stability of the cleaning operation until the surface cleanliness of the photovoltaic panel reaches the preset threshold.

[0117] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0118] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A photovoltaic cleaning drone spray cleaning control method, characterized in that: include: receiving surface condition data from a photovoltaic panel surface monitoring device and environmental parameters collected by an environmental sensor, comparing the surface condition data with a preset dirt grade standard, and obtaining dirt distribution characteristics and stubborn stain locations in the surface condition data; Determining the initial spraying parameters of the UAV using an adaptive control algorithm according to the dirt distribution characteristics, the location of the stubborn stains, and the environmental parameters; Using the initial spray parameters to drive the spray actuator to perform preliminary cleaning, while acquiring surface image data after cleaning through an image acquisition device; Comparing the surface image data with a preset cleanliness threshold, and calculating a deviation between the actual cleanliness and the cleanliness threshold; According to the deviation value, combined with the flight attitude data of the UAV, a fuzzy decision system is used to dynamically adjust the initial spray parameters to generate real-time spray control instructions; According to the real-time spray control instruction, the spray actuator is controlled to perform a fine cleaning operation until the surface image data meets the cleanliness threshold, and a final spray control plan is generated.

2. A photovoltaic cleaning drone spray cleaning control method according to claim 1, characterized in that: The method of determining the initial spraying parameters of the UAV by using an adaptive control algorithm according to the dirt distribution characteristics, the location of the stubborn stains, and the environmental parameters includes: Gridding the dirt distribution characteristics and the stubborn stain positions to obtain a dirt density value for each grid area; Normalizing the wind speed and temperature data in the environmental parameters to generate environmental impact factors; Inputting the dirt density value and the environmental impact factor into a parameter model of an adaptive control algorithm, wherein the parameter model is trained using historical cleaning data and outputs an initial spray pressure and water flow rate; Combined with the current flight altitude and horizontal movement speed of the UAV, the spray angle and flow rate corresponding to the initial spray pressure and the water flow velocity are calculated as the initial spray parameters.

3. A photovoltaic cleaning drone spray cleaning control method according to claim 2, characterized in that: The method of dynamically adjusting the initial spray parameters using a fuzzy decision system based on the deviation value and the flight attitude data of the UAV to generate a real-time spray control instruction includes: Dividing the deviation value into a plurality of fuzzy subsets, each fuzzy subset corresponding to a different deviation level; Collecting the flight attitude data of the UAV, including pitch angle, roll angle and yaw angle, and filtering the flight attitude data to obtain stable attitude parameters; Inputting the deviation level and the stable posture parameter into a fuzzy decision system, wherein the fuzzy decision system includes a preset fuzzy rule library and outputs a spray parameter adjustment amount; According to the spray parameter adjustment amount, the spray angle and flow rate in the initial spray parameters are corrected to generate the real-time spray control instruction.

4. A photovoltaic cleaning drone spray cleaning control method according to claim 3, characterized in that: The method includes: obtaining surface image data after cleaning by an image acquisition device, comparing the surface image data with a preset cleanliness threshold, and calculating a deviation between the actual cleanliness and the cleanliness threshold. grayscale the surface image data to extract brightness features and texture features in the image; Inputting the brightness feature and the texture feature into a preset cleanliness assessment model and outputting an actual cleanliness value; Obtaining a preset cleanliness threshold, and calculating the difference between the actual cleanliness value and the cleanliness threshold as the deviation value; When the deviation value is greater than a preset allowable range, it is marked as an area requiring secondary cleaning, and the position coordinates of the area are recorded.

5. The method for controlling spray cleaning of a photovoltaic cleaning drone according to claim 1, characterized in that: The method of determining the initial spray parameters of the UAV by using an adaptive control algorithm includes: Collect surface condition data, environmental parameters, spray parameters, and corresponding cleaning effect data from historical cleaning operations to construct a training dataset; Standardizing the training data set and dividing it into a training set and a validation set; The training set is used to train a parameter model of the adaptive control algorithm, wherein the parameter model adopts a multi-layer perceptron structure, the input layer is the dirt distribution characteristics and environmental influencing factors, and the output layer is the spray pressure and water flow rate; The performance of the trained parameter model is evaluated using the validation set. When the evaluation index reaches a preset threshold, the training is stopped and the optimal parameters of the parameter model are saved.

6. The method for controlling spray cleaning of a photovoltaic cleaning drone according to claim 3, characterized in that: The fuzzy decision system includes a preset fuzzy rule base, and the method for constructing the fuzzy rule base includes: Collect expert experience and historical cleaning cases to extract the mapping relationship between deviation level, flight attitude parameters and spray parameter adjustment amount; Converting the mapping relationship into fuzzy rules, each fuzzy rule includes a premise and a conclusion part, the premise is a fuzzy subset of the deviation level and the flight attitude parameter, and the conclusion is a fuzzy subset of the spray parameter adjustment amount; Performing conflict detection and merging processing on the fuzzy rules, removing redundant rules, and generating an initial fuzzy rule base; The initial fuzzy rule base is iteratively optimized through actual cleaning operation data, and the membership function parameters of the rules are adjusted to obtain an optimized fuzzy rule base.

7. The method for controlling spray cleaning of a photovoltaic cleaning drone according to claim 1, characterized in that: The step of controlling the spray actuator to perform a fine cleaning operation according to the real-time spray control instruction includes: Parsing the real-time spray control instruction to obtain the target spray angle, target flow rate and execution time; Converting the target spray angle into a steering gear angle of a spray actuator, and converting the target flow rate into a speed control signal of a water pump; driving the steering gear and the water pump to operate according to the steering gear rotation angle and the speed control signal to perform fine cleaning; During the cleaning process, the working status of the steering gear and water pump is monitored in real time. When any abnormality occurs, the operation is stopped immediately and an alarm signal is issued.

8. The method for controlling spray cleaning of a photovoltaic cleaning drone according to claim 1, characterized in that: The receiving of surface state data from a photovoltaic panel surface monitoring device and environmental parameters collected by an environmental sensor includes: Performing time series segmentation on the surface state data, dividing the continuously collected data into multiple data windows; The Kalman filter algorithm is used to perform noise suppression on each data window to obtain the denoised surface state data; Normalizing the humidity and light intensity data in the environmental parameters to generate an environmental correction coefficient; The denoised surface state data is fused with the environmental correction coefficient to obtain fused surface state features.

9. The method for controlling spray cleaning of a photovoltaic cleaning drone according to claim 5, characterized in that: The method of using the training set to train the parameter model of the adaptive control algorithm includes: Dividing the environmental parameters in the training set into a plurality of environmental scene categories, each category corresponding to a different climatic condition; For each environmental scenario category, a sub-model of the parameter model is trained separately to obtain multiple scenario-based sub-models; The output results of the multiple scenario-based sub-models are integrated through a voting mechanism to generate a comprehensive spray parameter prediction value; The error between the predicted value of the comprehensive spray parameter and the actual cleaning effect data is calculated, and the weight parameters of each sub-model are adjusted by back propagation.

10. A photovoltaic cleaning drone spray cleaning control method according to claim 7, characterized in that: The real-time monitoring of the working status of the steering gear and the water pump includes: Collect the deviation between the actual rotation angle of the servo and the target rotation angle, as well as the deviation between the actual speed of the water pump and the target speed; When the deviation value continuously exceeds the preset threshold value for a first preset time period, it is determined to be a minor fault and PID adjustment is started for compensation; When the deviation value exceeds the emergency threshold or continuously exceeds the preset threshold for a second preset time period, it is determined to be a serious fault, and shutdown protection is immediately triggered and a fault code is recorded.

Images