An adaptive intelligent cleaning control system and method for photovoltaic panels

Abstract

This invention discloses an adaptive intelligent cleaning control system and method for photovoltaic panels. The invention relates to the field of photovoltaic power generation technology and includes an image acquisition module for dynamically acquiring visible light and near-infrared images of the photovoltaic panels; a dust contamination level classification module for adaptive classification of dust contamination levels through image analysis; a cleaning decision module for determining whether to perform cleaning based on power generation revenue and cleaning cost analysis; and a parameter control module for generating control parameters such as robotic arm speed and water pressure using deep reinforcement learning. This invention achieves intelligent and precise control of photovoltaic panel cleaning, improving cleaning efficiency and reducing operation and maintenance costs.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation technology, specifically to an adaptive intelligent cleaning control system and method for photovoltaic panels. Background Technology

[0002] Dust accumulation on the surface of photovoltaic panels can significantly reduce their power generation efficiency. Existing intelligent cleaning technologies have the following drawbacks in practical applications:

[0003] Current detection methods mostly rely on a single visible light camera, which can only acquire surface color and texture information. They are not capable of recognizing deeper features such as dust thickness. Especially in scenarios where the dust color is similar to that of photovoltaic panels, it is easy to miss or misjudge. Furthermore, the classification of dust pollution levels is often based on fixed thresholds and does not combine multi-dimensional features such as coverage and grayscale difference for dynamic adjustment. This makes it difficult to adapt to the differences in dust distribution under different environments, resulting in a mismatch between the classification of levels and the actual degree of pollution.

[0004] Existing methods often use fixed cycles for cleaning decisions, without comprehensively considering economic factors such as real-time electricity prices and cleaning costs. This may lead to over-cleaning or delayed cleaning, increasing operation and maintenance costs.

[0005] The adjustment of cleaning execution parameters lacks an adaptive mechanism and relies heavily on manual experience or preset parameters. It cannot be optimized in real time according to the level of contamination and equipment status, resulting in low cleaning efficiency.

[0006] Therefore, there is an urgent need for an adaptive intelligent cleaning control system and method to achieve precise and intelligent cleaning of photovoltaic panels. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides an adaptive intelligent cleaning control system and method for photovoltaic panels, which solves the problems of unintelligent detection, decision-making, and parameter control in photovoltaic panel cleaning.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a photovoltaic panel adaptive intelligent cleaning control system, comprising:

[0009] The image acquisition module uses dual cameras to acquire visible light and near-infrared images of the photovoltaic panel at a frequency f within a dynamic time window;

[0010] The classification module averages multiple visible light images to obtain an average visible light image and averages multiple near-infrared light images to obtain an average near-infrared light image. It uses the average visible light image to calculate the coverage R and the average near-infrared light image to calculate the average absolute gray level difference ΔG. It then constructs a dynamic feature space mapping method to achieve adaptive classification of dust pollution levels.

[0011] The cleaning decision module calculates the power generation revenue Sg and cleaning cost Sr of the photovoltaic panel group under different pollution levels, and determines whether to perform the cleaning operation.

[0012] The parameter control module constructs a track-type robotic arm on top of the photovoltaic panel array and is equipped with a high-pressure water gun. Based on the Deep Deterministic Strategy Gradient Algorithm (DDPG), it generates three continuous control parameters: robotic arm speed, water pressure, and water volume, which are used to control the robotic arm and the high-pressure water gun in real time.

[0013] As a further embodiment of the present invention, the dual cameras are selected as an industrial RGB camera and a near-infrared camera, wherein the industrial RGB camera is equipped with an infrared cut-off filter and the near-infrared camera is equipped with an 850nm narrowband filter.

[0014] As a further aspect of the present invention, the specific steps for setting the dynamic time window are as follows:

[0015] Obtain the latitude and longitude of the photovoltaic power station, the current date, and the orientation of the photovoltaic panels;

[0016] Use the astronomical algorithm library to calculate the time period when the solar altitude angle is less than 30°.

[0017] Ambient light intensity is monitored in real time using a light sensor to determine the time periods when the light intensity is greater than 3000 lux;

[0018] The time period at the intersection of solar altitude angle <30° and illumination >3000 lux is used to form two candidate windows, and the first candidate window is selected as the effective window;

[0019] If the shadow length is greater than 1 / 3 of the board length during three consecutive days of data collection, the window will be automatically advanced / delayed by 15 minutes. As a further aspect of this invention, after determining the dynamic time window, the formula is used... Set the acquisition frequency f of the dual cameras, where T is the length of the dynamic time window.

[0020] As a further aspect of the present invention, the specific steps for determining the coverage R based on the average visible light image are as follows:

[0021] The average visible light image was converted to a grayscale image, and salt-and-pepper noise was removed using a 3×3 median filter.

[0022] The grayscale image is converted to the HSV color space, the brightness V channel is extracted, and the V channel is binarized using the Otsu adaptive thresholding method to segment the image into dusty and clean areas.

[0023] The binary image is opened to remove small noise areas, and then closed to fill the holes in the dust area. The number of pixels in the dust area, Ndust, is counted and the total number of pixels in the image, Ntotal. The coverage is calculated using the formula R = Ndust / Ntotal.

[0024] As a further aspect of the present invention, the specific steps for calculating ΔG based on the average near-infrared light image are as follows:

[0025] A 5×5 window median filter is applied to the average near-infrared image;

[0026] The filtered image is divided into several sub-blocks, and a local mean threshold is calculated for each sub-block. Based on the local threshold, the image is binarized into dust area and background area.

[0027] The above two steps are performed on the reference infrared image of the same group of photovoltaic panels in a clean state and the current image to be detected to obtain Iref and Icurr, respectively.

[0028] The absolute grayscale difference of Icurr and Iref is calculated pixel by pixel according to the formula ΔI(x,y)=|Icurr(x,y)-Iref(x,y)|.

[0029] According to the formula Calculate the average absolute gray level difference for the dusty area.

[0030] As a further aspect of the present invention, the specific steps for constructing a dynamic feature space mapping method for hierarchical classification are as follows:

[0031] After normalizing R and ΔG, map them to the polar coordinate system: according to the formula... Calculate the polar radius ρ, and then use the formula... Calculate the polar angle θ;

[0032] When ρ < ρ0 × (1 + k1 × var) and θ ∈ [θ1, θ2], it is classified as light dust pollution;

[0033] When ρ0×(1+k1×var)≤ρ<ρ1×(1+k2×var), it is classified as moderate dust pollution;

[0034] When ρ≥ρ1×(1+k2×var), it is classified as severe dust pollution;

[0035] Where ρ0 and ρ1 are the baseline radius thresholds, k1 and k2 are the dynamic coefficients, θ1 and θ2 are the upper and lower limits of the polar angle when there is light dust pollution, and var is the variance of historical pollution data.

[0036] As a further aspect of the present invention, the power generation revenue Sg is calculated according to the formula Sg=∑Q0×R×ΔG×η×P, where Q0 is the daily power generation per unit area of ​​the clean photovoltaic panel, η is the dust power generation loss conversion coefficient, and P is the real-time electricity price.

[0037] The cleaning cost Sr is calculated using the formula Sr=Sr0×(1+k×α(R,ΔG)), where Sr0 is the baseline cleaning cost, k is the dust pollution level adjustment factor (0 for light dust pollution, 0.5 for moderate dust pollution, and 1 for heavy dust pollution), and α(R,ΔG) is the level coefficient.

[0038] As a further aspect of the present invention, according to the formula The grade coefficient α(R,ΔG) is calculated, where w1 and w2 represent the weighting coefficients of coverage R and average absolute gray difference ΔG, ΔGmin and ΔGmax are the minimum and maximum gray differences in historical data, and Rmask is the coverage mask of the non-detection area.

[0039] An adaptive intelligent cleaning control method for photovoltaic panels includes:

[0040] S1. Use dual cameras to collect visible light and near-infrared images of the photovoltaic panel at a frequency f within a dynamic time window;

[0041] S2. Average the multiple visible light images to obtain the average visible light image, average the multiple near-infrared light images to obtain the average near-infrared light image, use the average visible light image to obtain the coverage R, use the average near-infrared light image to obtain the average absolute gray level difference ΔG, construct a dynamic feature space mapping method to realize the adaptive classification of dust pollution levels.

[0042] S3. Calculate the power generation revenue Sg and cleaning cost Sr of the photovoltaic panel group under different pollution levels, and determine whether to perform the cleaning operation.

[0043] S4. Construct a track-type robotic arm on top of the photovoltaic panel array and equip it with a high-pressure water gun. Based on the Depth Deterministic Strategy Gradient Algorithm (DDPG), generate three continuous control parameters: robotic arm speed, water pressure, and water volume, which are used to regulate the robotic arm and the high-pressure water gun in real time.

[0044] This invention provides an adaptive intelligent cleaning control system and method for photovoltaic panels, which has the following advantages compared with the prior art:

[0045] (1) This invention uses a dynamic time window and dual cameras for acquisition, combined with dynamic adjustment of solar altitude angle and light intensity, to avoid shadows and light interference, improve image quality, and provide a reliable data basis for subsequent pollution detection.

[0046] (2) Based on the multi-dimensional features of visible light and near-infrared images, this invention constructs a dynamic feature space mapping method to classify dust pollution levels, adaptively matches different pollution scenarios, and improves the accuracy and adaptability of dust pollution detection.

[0047] (3) This invention integrates dynamic analysis of power generation revenue and cleaning cost, and uses the DDPG algorithm to generate robotic arm speed, water pressure and water volume, thereby achieving adaptive control of cleaning strategy execution parameters, improving cleaning efficiency and reducing operation and maintenance costs. Attached Figure Description

[0048] Figure 1 This is the system principle block diagram of the present invention;

[0049] Figure 2 This is a flowchart of the steps of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] like Figure 1 This invention provides an adaptive intelligent cleaning control system for photovoltaic panels, comprising:

[0052] The image acquisition module captures visible light and near-infrared images of the photovoltaic panel through a camera;

[0053] One set of dual cameras is deployed for every 50 photovoltaic panels, installed diagonally above the photovoltaic panel array, at a distance of 3 meters and a downward angle of 45°, to avoid the photovoltaic panels' shadows obscuring the lenses;

[0054] The dual cameras can be either industrial RGB cameras or near-infrared cameras;

[0055] Industrial RGB cameras are equipped with infrared cut-off filters to ensure that the RGB channels only receive visible light;

[0056] The near-infrared camera is equipped with an 850nm narrowband filter to block visible light interference, making the light transmittance attenuation characteristics of thick dust more significant.

[0057] A light sensor is installed on the side of the camera bracket at the same height as the lens to measure light intensity in real time.

[0058] When acquiring real-time images of photovoltaic panels, it is first necessary to determine the time period for acquisition. Traditional methods often use a fixed time window. However, a fixed time window cannot cover the shadow patterns of different scenes because the solar altitude angle varies significantly with latitude, season, and date. Therefore, a dynamic time window can be used.

[0059] The specific steps for using a dynamic time window are as follows:

[0060] Enter the latitude and longitude of the photovoltaic power station (which can be automatically obtained via GPS), the current date, and the orientation of the photovoltaic panels;

[0061] Use the astronomical algorithm library to calculate the time period when the solar altitude angle is less than 30°.

[0062] Choosing a time period with a solar altitude angle of less than 30° is primarily about achieving a balance between light quality and shadow control, for the following reasons:

[0063] When the sun's altitude angle is low, the light enters the photovoltaic panel surface at an oblique angle. Compared with direct sunlight at noon, this can significantly reduce specular reflection, allowing the camera to clearly capture minute defects on the panel surface, such as hidden cracks and dust accumulation.

[0064] Photovoltaic panels are usually arranged in rows. When the solar altitude angle is too high, the shadow of the front row of panels may be projected onto the back row of panels. When the solar altitude angle is less than 30°, with a reasonable array spacing design, it can be ensured that the self-shadow of a single panel (such as the shadow of the bracket) and the shadow length of adjacent panels are less than 1 / 3 of the panel length, so as to avoid the shadow from interfering with the detection, such as mistakenly judging the shadow as a stain.

[0065] Ambient light is monitored in real time using a light sensor, and only time periods with light intensity >3000 lux are retained;

[0066] The core logic of retaining periods with light intensity >3000 lux is to find the optimal balance between image acquisition quality and environmental interference control in engineering practice. Periods with light intensity below 3000 lux may have stray light, such as ground reflection and reflection from surrounding buildings, which can lead to uneven image brightness, similar to stage lighting that flickers. It is easy to mistake changes in light and shadow for defects. Sufficient natural light can unify the direction of illumination and reduce background interference.

[0067] The time period at the intersection of solar altitude angle <30° and light intensity >3000 lux is selected to form two candidate windows (morning + afternoon). The first effective window in the morning is given priority, while avoiding possible cloud cover in the afternoon to ensure stable data collection once a day.

[0068] If the shadow length is greater than 1 / 3 of the board length during 3 consecutive days of data collection, the window will be automatically moved forward / delayed by 15 minutes.

[0069] Three consecutive days of excessively long shadows are essentially a signal that the solar altitude angle is consistently lower than expected: this may be due to changes in the sunrise / sunset azimuth angle caused by seasonal changes, or the local latitude causing the solar altitude angle corresponding to the time window to gradually deviate from the design threshold.

[0070] A single instance of prolonged shadowing may be due to temporary weather (such as cloud cover) or occasional obstruction (such as birds flying overhead), but three consecutive days of shadowing indicates a mismatch between the time window and the sun's position. If no adjustment is made, the shadow will continue to cover more than one-third of the photovoltaic panel's surface, leading to a higher probability of misjudgment during testing.

[0071] A 15-minute advance or delay can change the sun's altitude angle by about 3.75° (the sun moves 1° every 4 minutes). This change in angle is enough to shorten the shadow length by about 1 / 4. The shadow length is inversely proportional to the sun's altitude angle, and it is highly likely that the shadow can be kept within 1 / 3 of the board length.

[0072] For example, a power station is located at 114°E, 37°N, on October 1, 2024. The calculated time periods when the solar angle is less than 30° are: 6:30-8:30 AM and 3:30-5:30 PM. The actual light intensity measured by the light sensor is stable at >3000 lux after 7:00 AM, and begins to decrease at 3:30 PM, then is <3000 lux after 4:00 PM. Therefore, the time period with an intensity greater than 3000 lux is 7:00-4:00 PM. Finding the intersection of these two values, we get 7:00-8:30 AM and 3:30-4:00 PM. The final selected window is 7:00-8:30 AM.

[0073] After determining the dynamic time window, according to the formula Set the acquisition frequency f of the dual cameras, where T is the length of the dynamic time window;

[0074] Numerator 100 - 40 × e -0.05T This represents the total number of captured frames, which combines the exponential decay characteristic with the base number of frames.

[0075] e -0.05T The time decay reflects the dust deposition rate. The denominator 60T converts the time window into seconds, ensuring that the frequency unit is frames per second.

[0076] When T < 10 minutes, the frequency decays rapidly, which matches the high-frequency sampling requirements of sudden pollution.

[0077] When T ≥ 10 minutes, the frequency decay slows down, adapting to the slow-changing characteristics of dust deposition and avoiding data redundancy.

[0078] The classification module averages multiple visible light images captured by industrial RGB cameras within a dynamic time window to obtain an average visible light image, and averages multiple near-infrared light images captured by near-infrared cameras within a dynamic time window to obtain an average near-infrared light image.

[0079] The core purpose of averaging multiple images is to suppress random noise. Due to the randomness of factors such as sensor noise and environmental interference, when averaging multiple images of the same scene, random noise will cancel each other out due to superposition, while the real scene signal will be preserved and enhanced, thereby improving the image signal-to-noise ratio and stability, and providing a more reliable input for subsequent processing.

[0080] Dust in visible light images usually appears as a gray level or color different from the background. The dust area extracted by threshold segmentation and morphological processing can directly quantify the "breadth" of surface contamination by its coverage area ratio.

[0081] The specific steps for determining the coverage R based on the average visible light image are as follows:

[0082] The average visible light image was converted to a grayscale image, and salt-and-pepper noise was removed using a 3×3 median filter.

[0083] Grayscale is used to reduce interference from color dimensions and focus on brightness features (dust usually manifests as grayscale differences), while median filtering can eliminate light spots and noise while preserving edges, avoiding the misidentification of noise as dust during subsequent segmentation;

[0084] The grayscale image is converted to the HSV color space, the brightness V channel is extracted, and the V channel is binarized using the Otsu adaptive thresholding method. The image is then segmented into a dusty region (foreground, low grayscale value) and a clean region (background, high grayscale value).

[0085] In the HSV space, the brightness channel V is more robust to changes in illumination, which can reduce the impact of reflections or shadows on the photovoltaic panel surface; the Otsu thresholding method automatically determines the optimal segmentation threshold through statistical grayscale histograms, adapting to the brightness differences between dusty and clean areas under different illumination conditions.

[0086] The binary image is opened (erosion followed by dilation) to remove small noise areas, and then closed (dilation followed by erosion) to fill the small holes in the dust area. The number of pixels in the dust area Ndust and the total number of pixels in the image Ntotal are counted, and the coverage is calculated according to the formula R = Ndust / Ntotal.

[0087] Morphological operations can eliminate isolated noise generated during segmentation and fill voids in dust areas, ensuring that dust areas are continuous and complete connected domains, making coverage calculations more accurate.

[0088] Near-infrared light has different penetrability and reflection characteristics than visible light. Dust can change the intensity of near-infrared light reflection on a surface, resulting in changes in grayscale values. Calculating the absolute grayscale difference can quantify this change, and its value directly reflects the "depth" of surface contamination.

[0089] The specific steps for calculating the average absolute gray level difference ΔG based on the average near-infrared image are as follows:

[0090] A 5×5 window median filter is applied to the average near-infrared image;

[0091] The filtered image is divided into several sub-blocks. A local mean threshold is calculated for each sub-block. Based on the local threshold, the image is binarized into dust areas (high grayscale) and background areas (low grayscale).

[0092] The surface temperature distribution of photovoltaic panels may be uneven due to uneven illumination and differences in internal circuits. The global threshold is prone to missegmentation. Dynamically adjusting the threshold according to the local grayscale characteristics of each sub-block can more accurately segment high-temperature areas covered by dust.

[0093] The above two steps are performed on the reference infrared image of the same group of photovoltaic panels in a clean state and the current image to be detected to obtain Iref and Icurr, respectively.

[0094] Iref refers to the near-infrared light image collected when the photovoltaic panel surface is free of dust and the environmental conditions (such as light intensity, ambient temperature, and shooting angle) are consistent with the state to be tested.

[0095] The absolute grayscale difference of Icurr and Iref is calculated pixel by pixel according to the formula ΔI(x,y)=|Icurr(x,y)-Iref(x,y)|.

[0096] According to the formula Calculate the average absolute grayscale difference for the dusty area;

[0097] By directly comparing the grayscale difference between clean and dirty states, common-mode interference from ambient temperature and light can be eliminated. The average absolute grayscale difference in the dusty area can directly reflect the thickness—the greater the thickness, the stronger the heat insulation effect, the more the grayscale of the current image increases relative to the reference image, and the larger the ΔG value.

[0098] A dynamic feature space mapping method is constructed to map the coverage area ratio R and the average absolute gray level difference ΔG to a polar coordinate system. The adaptive classification of dust pollution levels is achieved by dynamically expanding the sector region. The specific operation is as follows:

[0099] After normalizing R and ΔG, map them to the polar coordinate system:

[0100] The polar radius ρ represents the severity of pollution, and is directly calculated from the geometric mean of R and ΔG.

[0101] The polar angle θ represents the pollution type bias, determined by the ratio of R to ΔG.

[0102] When θ≈0 (R>>ΔG), it indicates a large area of ​​thin dust.

[0103] when At this time, it indicates a localized thick layer of dust;

[0104] In the polar coordinate system, define three dynamically expanding sector regions, corresponding to the pollution levels:

[0105] When ρ < ρ0 × (1 + k1 × var) and θ ∈ [θ1, θ2], it is classified as light dust pollution;

[0106] If only the polar radius is used to determine light dust pollution, local thick dust with small ρ but large θ may be missed. By setting the polar angle range, local thick dust with large θ can be filtered out, leaving only large areas of thin dust as light pollution, thus avoiding the misjudgment of "small area thick pollution" as not requiring treatment.

[0107] When ρ0×(1+k1×var)≤ρ<ρ1×(1+k2×var), it is classified as moderate dust pollution;

[0108] When ρ≥ρ1×(1+k2×var), it is classified as severe dust pollution;

[0109] Where ρ0 and ρ1 are the baseline radius thresholds, k1 and k2 are dynamic coefficients, θ1 and θ2 are the upper and lower limits of the polar angle when there is light dust pollution, and var is the variance of historical pollution data, reflecting the degree of environmental fluctuation.

[0110] The cleaning decision module calculates the power generation revenue Sg and cleaning cost Sr of the photovoltaic panel group under different pollution levels, and determines whether to perform the cleaning operation.

[0111] Triggering a cleaning command by comparing the power generation revenue Sg with the cleaning cost Sr is essentially an economic decision-making logic based on dynamic cost-benefit analysis. Its core lies in transforming the operation and maintenance strategy into a quantifiable economic trade-off.

[0112] Dust covering photovoltaic panels reduces light transmittance, directly leading to power generation loss. Sg transforms this loss into calculable economic value—the power generation revenue that can be recovered after cleaning. Furthermore, long-term dust accumulation can cause hidden damage such as hot spot effects; Sg also implies preventative protection for equipment lifespan.

[0113] Traditional fixed-cycle cleaning (such as monthly or quarterly cleaning) has significant limitations: it may over-clean when pollution is light, wasting resources; and it may delay cleaning when pollution is heavy, exacerbating power generation losses. However, by comparing Sg and Sr in real time, the strategy can be dynamically adjusted according to the actual degree of pollution—the operation is started when the power generation benefit from cleaning exceeds the cost, and postponed otherwise, achieving precise management of maintenance on demand.

[0114] The power generation revenue Sg is calculated according to the formula Sg=∑Q0×R×ΔG×η×P, where Q0 is the daily power generation per unit area of ​​the clean photovoltaic panel, calibrated by historical data, η is the dust power generation loss conversion coefficient, representing the power generation loss caused by one unit (R×ΔG), and P is the real-time electricity price.

[0115] The cleaning cost Sr is calculated using the formula Sr=Sr0×(1+k×α(R,ΔG)), where Sr0 is the baseline cleaning cost.

[0116] k is the dust pollution level adjustment factor. If it is light dust pollution, k is 0; if it is moderate dust pollution, k is 0.5; if it is heavy dust pollution, k is 1.

[0117] α(R,ΔG) is the rank coefficient, and its specific calculation formula is as follows:

[0118] Where w1+w2=1 represents the weighting coefficient of coverage R and average absolute gray difference ΔG, ΔGmin and ΔGmax are the minimum and maximum gray differences in historical data, and Rmask is the coverage mask of the non-detection area, with a value of 0-1.

[0119] The coverage R directly determines the physical area to be cleaned. The larger R is, the wider the area to be cleaned, and the cost of consumables and time increases linearly.

[0120] The average absolute gray difference ΔG reflects the thickness and adhesion strength of dust. The larger ΔG is, the stronger the heat insulation effect of dust. It usually corresponds to a thicker or more adherent contamination layer, which requires higher pressure cleaning, resulting in a non-linear increase in energy consumption and equipment wear costs.

[0121] 1-Rmask is used to remove areas in the image that do not need cleaning, such as borders and bracket shadows, so that the grade coefficient only reflects the contamination of the effective detection area, where Rmask < 0.9;

[0122] w1 focuses on cost per unit area: it is suitable for scenarios where dust is evenly distributed but thin, such as floating dust, where cleaning cost is positively correlated with area.

[0123] w2 focuses on depth cost: it is suitable for localized thick dust, such as bird droppings and scabs, where the thickness has a greater impact on the difficulty of cleaning;

[0124] When Sg≥1.2×Sr, the cleaning operation is triggered; otherwise, the cleaning operation is not triggered.

[0125] The cleaning cost Sr usually only includes direct inputs, while the actual power generation revenue Sg needs to cover implicit costs, such as downtime losses and temporary equipment commissioning costs during the cleaning process. Setting Sg to 1.2 times Sr is essentially to ensure that the revenue can not only cover the explicit cleaning costs, but also offset the implicit losses through a proportional coefficient, thus avoiding decision-making losses due to incomplete cost accounting.

[0126] Pollution levels, equipment status, and environmental factors can all cause dynamic changes in Sg and Sr. If only Sg = Sr is used as the trigger condition, when the data fluctuates, there may be a situation where the revenue just covers the cost but there is no actual surplus. The 1.2-fold coefficient is essentially to introduce a risk buffer for decision-making. By reserving a 20% benefit space, the probability of decision-making errors caused by data uncertainty is reduced.

[0127] The parameter control module constructs a track-type robotic arm on top of the photovoltaic panel array and is equipped with a high-pressure water gun. Based on the deep deterministic policy gradient algorithm (DDPG), it constructs a state space S, an action space A, and a reward function R. Through learning, it generates three continuous control parameters: robotic arm speed, water pressure, and water volume, which are used to control the robotic arm and the high-pressure water gun in real time.

[0128] The track layout fits the dense arrangement and tilt design of the photovoltaic panels. The robotic arm achieves coverage without blind spots through translation and joint rotation, solving the problem of blind spots in spraying by manual / ground equipment and ensuring that each panel is cleaned thoroughly.

[0129] High-pressure water guns remove stubborn stains such as bird droppings and accumulated dust through impact, making them more efficient than manual wiping and avoiding damage to the photovoltaic panel coating from hard scratches. Combined with flow control, water-saving cleaning is achieved.

[0130] The specific steps for generating continuous control parameters based on DDPG are as follows:

[0131] Define a state space and construct a multidimensional state vector S containing pollution characteristics, environmental parameters, and equipment states. t =[R t ,ΔG t ,T t H t ,P t ,L t ], where R t ΔG represents the dust coverage of the average visible light image. t T represents the average absolute gray level difference of the average near-infrared image. t For real-time temperature (affecting photovoltaic panel efficiency and cleaning solution evaporation), H t For air humidity (affecting dust stickiness), P t At the current electricity price, L t The lifespan of the component;

[0132] Design the motion space and define the discrete motion set A = {robotic arm speed, water pressure, water volume};

[0133] Construct the reward function, specifically the formula R(S) t ,a t )=γ×(Su-λ×Sr)+(1-γ)×Q t Where Su is the incremental power generation revenue after cleaning, Sr is the cleaning cost, γ is the recent revenue weight (which can be taken as 0.7), λ is the cost penalty coefficient (which can be taken as 1.2), and Q... t To reward component lifespan protection, 10 points are added for each light cleaning and 5 points are deducted for each heavy cleaning, in order to avoid excessive wear and tear.

[0134] The Actor-Critic network is constructed with the following parameters:

[0135] For the Actor network, the input layer is a 6-dimensional vector St, the fully connected layer 1 has 256 neurons, the fully connected layer 2 has 128 neurons, the dropout is 0.2, the activation function is ReLU, the output layer has 4 neurons, corresponding to 3 actions, and the learning rate is 1e-4.

[0136] For the Critic network, the input layer is a 9-dimensional vector, i.e., state St + action A. The fully connected layer 1 has 384 neurons, the fully connected layer 2 has 192 neurons, the dropout is 0.3, the activation function is LeakyReLU, the output layer has 1 neuron, and the learning rate is 3e-4.

[0137] Collect cleaning records from the past three years, acquire new states in real time through edge computing, and store them in the experience pool D. Each time, randomly select several samples from D and use mean squared error (MSE) to optimize the network parameters.

[0138] In addition, γ and λ in the reward function can be updated monthly based on the latest cleaned data.

[0139] like Figure 2 An adaptive intelligent cleaning control method for photovoltaic panels, comprising the following steps:

[0140] S1. Use dual cameras to collect visible light and near-infrared images of the photovoltaic panel at a frequency f within a dynamic time window;

[0141] S2. Average the multiple visible light images to obtain the average visible light image, average the multiple near-infrared light images to obtain the average near-infrared light image, use the average visible light image to obtain the coverage R, use the average near-infrared light image to obtain the average absolute gray level difference ΔG, construct a dynamic feature space mapping method to realize the adaptive classification of dust pollution levels.

[0142] S3. Calculate the power generation revenue Sg and cleaning cost Sr of the photovoltaic panel group under different pollution levels, and determine whether to perform the cleaning operation.

[0143] S4. Construct a track-type robotic arm on top of the photovoltaic panel array and equip it with a high-pressure water gun. Based on the Depth Deterministic Strategy Gradient Algorithm (DDPG), generate three continuous control parameters: robotic arm speed, water pressure, and water volume, which are used to regulate the robotic arm and the high-pressure water gun in real time.

[0144] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0145] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A photovoltaic panel adaptive intelligent cleaning control system, characterized in that, include: The image acquisition module uses dual cameras to acquire visible light and near-infrared light images of the photovoltaic panel at a frequency f within a dynamic time window; The classification module averages multiple acquired visible light images to obtain an average visible light image, averages multiple acquired near-infrared light images to obtain an average near-infrared light image, calculates the coverage R using the average visible light image, and calculates the average absolute grayscale difference using the average near-infrared light image. A dynamic feature space mapping method is constructed to achieve adaptive classification of dust pollution levels. Specifically, R is mapped to... After normalization, map to polar coordinates: according to the formula Calculate the polar diameter According to the formula Calculate the polar angle ;when It is classified as light dust pollution; when It is classified as moderate dust pollution; when It is classified as severely dusty; among them, , As the baseline radius threshold, , For dynamic coefficients, , The upper and lower limits of the polar angle are for light dust pollution, and var is the variance of historical pollution data; The cleaning decision module calculates the power generation revenue Sg and cleaning cost Sr of the photovoltaic panel group under different pollution levels, and determines whether to perform the cleaning operation. The parameter control module constructs a track-type robotic arm on top of the photovoltaic panel array and is equipped with a high-pressure water gun. Based on the Deep Deterministic Strategy Gradient Algorithm (DDPG), it generates three continuous control parameters: robotic arm speed, water pressure, and water volume, which are used to control the robotic arm and the high-pressure water gun in real time.

2. The photovoltaic panel adaptive intelligent cleaning control system according to claim 1, characterized in that, The dual-camera system uses an industrial RGB camera and a near-infrared camera. The industrial RGB camera is equipped with an infrared cutoff filter, and the near-infrared camera is equipped with an 850nm narrowband filter.

3. The photovoltaic panel adaptive intelligent cleaning control system according to claim 1, characterized in that, The specific steps for setting a dynamic time window are as follows: Obtain the latitude and longitude of the photovoltaic power station, the current date, and the orientation of the photovoltaic panels; Use the astronomical algorithm library to calculate the time period when the solar altitude angle is less than 30°. Ambient light intensity is monitored in real time using a light sensor to determine the time periods when the light intensity is greater than 3000 lux; The time period at the intersection of solar altitude angle <30° and illumination >3000 lux is used to form two candidate windows, and the first candidate window is selected as the effective window; If the shadow length is greater than 1 / 3 of the board length during 3 consecutive days of data collection, the window will be automatically moved forward / backward by 15 minutes.

4. The photovoltaic panel adaptive intelligent cleaning control system according to claim 1, characterized in that, After determining the dynamic time window, according to the formula Set the acquisition frequency f of the dual cameras, where T is the length of the dynamic time window.

5. The photovoltaic panel adaptive intelligent cleaning control system according to claim 1, characterized in that, The specific steps for determining the coverage R based on the average visible light image are as follows: The average visible light image was converted to a grayscale image, and salt-and-pepper noise was removed using a 3×3 median filter. The grayscale image is converted to the HSV color space, the brightness V channel is extracted, and the V channel is binarized using the Otsu adaptive thresholding method to segment the image into dusty and clean areas. The binary image is opened to remove small noise areas, and then closed to fill the holes in the dust area. The number of pixels in the dust area, Ndust, is counted and the total number of pixels in the image, Ntotal. The coverage is calculated using the formula R=Ndust / Ntotal.

6. The photovoltaic panel adaptive intelligent cleaning control system according to claim 1, characterized in that, Based on the average near-infrared light image The specific steps are as follows: A 5×5 window median filter is applied to the average near-infrared image; The filtered image is divided into several sub-blocks, and a local mean threshold is calculated for each sub-block. Based on the local threshold, the image is binarized into dust area and background area. The above two steps are performed on the reference infrared image of the same group of photovoltaic panels in a clean state and the current image to be detected to obtain Iref and Icurr, respectively. According to the formula Calculate the absolute grayscale difference pixel by pixel for Icurr and Iref; According to the formula Calculate the average absolute gray level difference for the dusty area.

7. The photovoltaic panel adaptive intelligent cleaning control system according to claim 1, characterized in that, According to the formula Calculate the power generation revenue Sg, where, To determine the daily power generation per unit area of ​​clean photovoltaic panels, Here, P represents the conversion factor for dust-to-energy power generation losses, and P represents the real-time electricity price. According to the formula The cleaning cost Sr is calculated, where Sr0 is the baseline cleaning cost, and k is a dust contamination level adjustment factor: k is 0 for light dust contamination, 0.5 for moderate dust contamination, and 1 for heavy dust contamination. This is the grade coefficient.

8. The photovoltaic panel adaptive intelligent cleaning control system according to claim 7, characterized in that, According to the formula Calculate the grade coefficient Where w1 and w2 represent the coverage R and the average absolute gray level difference, respectively. The weighting coefficients, , Rmask represents the minimum and maximum grayscale difference in historical data, and Rmask is the coverage mask for the non-detection area.

9. A photovoltaic panel adaptive intelligent cleaning control method, used in the photovoltaic panel adaptive intelligent cleaning control system according to any one of claims 1-8, characterized in that, include: S1. Use dual cameras to collect visible light and near-infrared light images of the photovoltaic panel at a frequency f within a dynamic time window; S2. Average the acquired visible light images to obtain the average visible light image. Average the acquired near-infrared images to obtain the average near-infrared image. Calculate the coverage R using the average visible light image and the average absolute grayscale difference using the average near-infrared image. A dynamic feature space mapping method is constructed to achieve adaptive classification of dust pollution levels; S3. Calculate the power generation revenue Sg and cleaning cost Sr of the photovoltaic panel group under different pollution levels, and determine whether to perform the cleaning operation. S4. Construct a track-type robotic arm on top of the photovoltaic panel array and equip it with a high-pressure water gun. Based on the Depth Deterministic Strategy Gradient Algorithm (DDPG), generate three continuous control parameters: robotic arm speed, water pressure, and water volume, which are used to regulate the robotic arm and the high-pressure water gun in real time.

Images