Intelligent flushing system for solar cell panel of photovoltaic power station

Abstract

The invention relates to the technical field of solar panel maintenance, in particular to a photovoltaic power station solar cell panel intelligent flushing system, which comprises an image analysis module, an intelligent control module, an image processing module, a data processing module, a data processing module, a data processing module, a data processing module and a data processing module, and is characterized in that the image analysis module is used for monitoring smudginess data of the surface of a solar cell panel and transmitting the smudginess data to the intelligent control module in real time; the multispectral camera collects surface images of a current cell panel and a next cell panel, the surface images comprise cell panel surface images collected under different spectral bands or angles, and smudginess distribution and smudginess degree of the next cell panel are obtained; a cleaning strategy is generated according to the cleaning requirement through a preset intelligent algorithm, a control instruction is output and sent to the intelligent control module through the communication unit, and the intelligent control module adjusts the working state of the flushing device. According to the system, the precise recognition capability of the system on the dirty area is improved, it is ensured that water flow can effectively cover the part needing to be cleaned in the cleaning process, and waste of the non-dirty area is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of solar panel maintenance, and particularly to an intelligent flushing system for solar panels in a photovoltaic power station. Background Art

[0002] In a photovoltaic power station, as a key component, the efficiency of solar panels is closely related to their cleanliness. Currently, most panel cleaning systems on the market adopt traditional methods such as manual cleaning, mechanical brushing, and spray cleaning. Although manual cleaning is effective in certain specific scenarios, due to the large workload, low efficiency, difficulty in precisely controlling the cleaning intensity, and safety risks, it can no longer meet the requirements of efficient operation and maintenance of modern photovoltaic power stations. Although mechanical brushing improves the automation level, due to the existence of fixed brush heads and the difficulty in adapting to different degrees of dirt and cleaning intensities, it may cause mechanical damage to the panel surface and affect its long-term stability.

[0003] In recent years, spray cleaning has become the mainstream method for cleaning photovoltaic power stations. By flushing the panel surface with high-pressure water flow, however, most existing spray systems are based on preset cleaning strategies and lack real-time monitoring and intelligent adjustment of the dirt level on the panel surface. Traditional spray systems cannot adjust the nozzle position, pressure, and flushing sequence according to the actual dirt condition of each panel, often resulting in waste of resources and uneven cleaning effects. In addition, with the continuous expansion of the scale of photovoltaic power stations, cleaning efficiency and resource utilization have become urgent problems to be solved. Especially during the cleaning process of large-area panels, how to accurately clean according to the specific dirt level of each panel and avoid ineffective cleaning and damage to the panels has become a difficult point in the current technology. Summary of the Invention

[0004] The present invention provides an intelligent flushing system for solar panels in a photovoltaic power station.

[0005] An intelligent flushing system for solar panels in a photovoltaic power station includes an image analysis module, a central processing unit, a communication unit, an intelligent control module, and a flushing device;

[0006] The image analysis module is used to monitor the dirt data on the surface of the solar panel and transmit the dirt data to the intelligent control module in real time. The image analysis module includes a multispectral camera, which collects the surface images of the current panel and the next panel. The surface images include collecting the surface images of the panel at different spectral bands or angles. Through reflection area recognition, the dirt distribution and dirt level of the next panel are obtained;

[0007] The central processing unit generates a cleaning strategy according to the dirt distribution and degree of dirt of the next solar panel provided by the image analysis module, using a preset intelligent algorithm, outputs a control instruction, and the control instruction is sent to the intelligent control module through the communication unit, and the intelligent control module adjusts the working state of the flushing device;

[0008] The flushing device includes a high-pressure water gun and an adjustable nozzle. Based on the control instruction, the intelligent control module adjusts the position and pressure of the adjustable nozzle of the flushing device in advance, where:

[0009] Adjustable nozzle position: Automatically adjust the spraying position of the adjustable nozzle according to the dirt distribution of the next solar panel;

[0010] Water flow pressure adjustment: Automatically adjust the water pressure of the high-pressure water gun according to the degree of dirt, and at the same time avoid damage to the solar panel caused by excessive water pressure.

[0011] Optionally, the image analysis module specifically includes:

[0012] Multi-spectral image acquisition: The image analysis module uses a multi-spectral camera to collect images of the solar panel surface at different spectral bands or angles, including at least visible light and infrared light images;

[0013] Construct a reflection area prediction model: The image analysis module trains a neural network specifically for reflection area recognition and prediction. This neural network predicts and marks the reflection area according to the image features under different lighting conditions, including reflection intensity, spectral differences between visible light and infrared light images, and morphological features of the reflection area, generates a reflection area layer, and based on the obtained reflection area map, removes it from the original image to obtain a dirt area layer;

[0014] Differential processing: Perform differential processing with the dirt area layer to suppress the influence of reflection on the dirt area, obtain a dirt distribution map of the solar panel, and at the same time calculate the dirt degree of each dirt area through image intensity information to ensure accurate evaluation of the dirt.

[0015] Optionally, the multi-spectral camera collects images of the solar panel surface at multiple spectral bands by adjusting the angle and spectral band of the camera, including visible light images and infrared light images. The multi-spectral camera includes an adjustable band selector that switches within different spectral band ranges to capture the surface reflection characteristics of the solar panel under different spectra, providing multi-dimensional data support for dirt area recognition and reflection area suppression.

[0016] Optionally, the reflection area prediction model adopts a convolutional neural network structure and is trained with the annotated surface image data of the solar panel. During the training process, the network learns the image features under different lighting conditions, including the reflection intensity, the spectral differences between visible light and infrared light images, and the morphological features of the reflection area. Among them:

[0017] Reflection intensity feature: The neural network extracts the reflection intensity features in the visible light and infrared light images, and identifies potential reflection areas through the brightness and intensity differences of the reflected light;

[0018] Spectral difference feature: Using the spectral differences between visible light and infrared light images, the neural network distinguishes the reflection area and the dirty area. The reflection area shows different brightness or color differences in the infrared light image;

[0019] Morphological feature of the reflection area: The neural network analyzes the morphological features of the reflection area to distinguish the reflection area from other interference areas;

[0020] Based on the reflection intensity, the spectral differences between visible light and infrared light images, and the morphological features of the reflection area, the neural network outputs the prediction result of the reflection area and generates the reflection area layer R reflect (x,y). Among them, R reflect (x,y)=1 indicates that this position is the reflection area, and R reflect (x,y)=0 indicates a non-reflection area. The generated reflection area layer is processed differently from the dirty area layer M(x,y). In the dirty area layer M(x,y), M(x,y)=1 indicates that this position is the dirty area, and M(x,y)=0 indicates a non-dirty area. Thus, the dirt distribution is obtained. M(x,y) is the representation of the dirt distribution. The differential processing includes suppressing the influence of reflection on the dirty area by superimposing the reflection area layer and the dirty area layer, and masking the area R reflect (x,y) marked by the reflection area to remove the interference of the reflected light on the identification of the dirty area and ensure the accuracy of the identification of the dirty area M(x,y). After removing the reflection influence, the finally output dirty area layer M(x,y) presents the true dirt distribution of the solar panel, providing accurate data support for subsequent cleaning strategies.

[0021] Optionally, the reflection area prediction model is trained using a convolutional neural network. The input of the neural network is multi-spectral data including visible light and infrared light images, and the output is the reflection area layer. The training process is carried out by minimizing the following loss function:

[0022] L reflect =∑ (x,y) |R reflect (x,y)-R groundtruth (x,y)| 2 , where Rreflect (x, y) is the layer of the reflection area predicted by the neural network, R groundtruth (x, y) is the layer of the true reflection area obtained by manual annotation. Using the prediction result of the reflection area output by the neural network, the layer of the reflection area R reflect (x, y) is generated.

[0023] Optionally, the differential processing is expressed as: M final (x, y) = M(x, y) · (1 - R reflect (x, y));

[0024] For the layer of the reflection area R reflect (x, y), 1 indicates that this position is the reflection area, 0 indicates the non - reflection area. In the layer of the dirt area M(x, y), 1 indicates the dirt area, 0 indicates the non - dirt area. Through differential processing, if R reflect (x, y) = 1, it means that this position is the reflection area. At this time, 1 - R reflect (x, y) = 0, so the layer of the dirt area M(x, y) will be suppressed at the position of the reflection area, ensuring that the interference of the reflection area on the dirt area is removed; if R reflect (x, y) = 0, it means that this position is not the reflection area. At this time, 1 - R reflect (x, y) = 1, and the layer of the dirt area M(x, y) remains unchanged.

[0025] Optionally, calculating the dirt degree of each dirt area through the image intensity information specifically includes:

[0026] The intensity value of each pixel in the multi - spectral image is I(x, y), where I(x, y) is the brightness value of the current pixel position. The lower the intensity, the more dirt accumulates on the surface area;

[0027] Dirt will cause a decrease in the image intensity value (the dirt area is usually darker). Calculate the dirt degree D(x, y):

[0028] where I(x, y) is the image intensity value of the current pixel position, I max is the maximum intensity value in the whole image (for example, 255 for an 8 - bit image), I min is the minimum intensity value in the whole image. The dirt degree is quantified by the difference in image intensity. The lower the intensity, the higher the dirt degree. The value range of D(x, y) is [0, 1].

[0029] Optionally, the cleaning strategy specifically includes:

[0030] Flushing sequence: Adjust the spraying position of the nozzle according to the dirt distribution, and determine the flushing sequence according to the dirt degree and dirt distribution;

[0031] Water pressure adjustment: Adjust the water pressure of the high-pressure water gun according to the degree of dirt to ensure the cleaning effect in different areas.

[0032] Optionally, the flushing sequence includes priority calculation, expressed as:

[0033] P 优 (x, y) = D(x, y)·M(x, y), where P 优 (x, y) is the cleaning priority of area (x, y). The higher the cleaning priority, the more priority should be given to cleaning this area. Based on the cleaning priority, the cleaning sequence is sorted.

[0034] Optionally, the water pressure P flow (x, y) adjustment is expressed as: P flow (x, y) = P min +(P max -P min )·D(x, y), where P min and P max are the minimum and maximum values of the water pressure respectively, and D(x, y) is the degree of dirt, which determines the range of water pressure change.

[0035] Advantages of the present invention:

[0036] In the present invention, by combining multi-spectral image acquisition and a reflection area prediction model, the interference of reflected light on the surface of the solar panel is effectively suppressed, and the reflection area and the dirty area are accurately distinguished. Traditional flushing systems are often affected by the reflected light source, resulting in misjudgment of the dirty area. However, in the present invention, a dedicated reflection area prediction network is trained through a deep learning model, combined with information in different spectral bands, which can efficiently identify the reflected light intensity, spectral differences, and morphological characteristics of the reflection area, thereby accurately generating a reflection area layer, avoiding the interference of reflection artifacts on the identification of the dirty area, improving the accurate identification ability of the system for the dirty area, ensuring that the water flow can effectively cover the part to be cleaned during the cleaning process, and avoiding waste of the non-dirty area.

[0037] In the present invention, based on the real-time monitoring of the degree of dirt on the surface of the battery panel, the water flow pressure and sequence of the flushing device are dynamically adjusted through the intelligent control module, so as to accurately match the cleaning requirements. Through multi-spectral image analysis and deep learning algorithms, the degree of dirt on the next battery panel can be predicted in advance during the cleaning process of the battery panel, and the working state of the flushing device is automatically adjusted. In particular, the dynamic adjustment of the water flow position is optimized according to the dirt distribution, and the adjustment of the water flow pressure is refined according to the degree of dirt, improving the cleaning efficiency, reducing water resource waste, and avoiding over-cleaning and damage to the battery panel caused by water flow. Description of the Drawings

[0038] To more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only those of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0039] Figure 1 Schematic diagram of each module of the flushing system according to an embodiment of the present invention;

[0040] Figure 2 Schematic diagram of the image processing module according to an embodiment of the present invention. Detailed implementation manners

[0041] The present invention will be described in detail below in conjunction with the drawings and specific embodiments. At the same time, it should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well-known technologies, those skilled in the art can also adopt other alternative methods for implementation; moreover, the drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0042] It should be pointed out that in the specification, it is mentioned that "an embodiment", "embodiments", "exemplary embodiments", "some embodiments", etc. indicate that the described embodiments may include specific features, structures or characteristics, but not necessarily every embodiment includes such specific features, structures or characteristics. Additionally, when combining embodiments to describe specific features, structures or characteristics, implementing such features, structures or characteristics in combination with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the relevant art.

[0043] Generally, terms can be understood at least in part from their use in the context. For example, at least in part depending on the context, the term "one or more" used herein can be used to describe any feature, structure or characteristic in a singular sense, or can be used to describe a combination of features, structures or characteristics in a plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey a set of exclusive factors, but rather, at least in part depending on the context, allowing for the existence of other factors that may not be explicitly described.

[0044] As Figure 1 - Figure 2 shown, a smart flushing system for solar panels in a photovoltaic power station includes an image analysis module, a central processing unit, a communication unit, a smart control module, and a flushing device;

[0045] The image analysis module is used to monitor the dirt data on the surface of the solar panel and transmit the dirt data to the intelligent control module in real time. The image analysis module includes a multispectral camera, which collects the surface images of the current panel and the next panel. The surface images include the surface images of the panel collected at different spectral bands or angles. Through reflection area recognition, the dirt distribution and dirt degree of the next panel are obtained;

[0046] Based on the dirt distribution and dirt degree of the next panel provided by the image analysis module, the central processing unit uses a preset intelligent algorithm to generate a cleaning strategy according to the cleaning requirements, outputs a control instruction, and the control instruction is sent to the intelligent control module through the communication unit. The intelligent control module adjusts the working state of the flushing device;

[0047] The flushing device includes a high-pressure water gun and an adjustable nozzle. Based on the control instruction, the intelligent control module adjusts the position and pressure of the adjustable nozzle of the flushing device in advance, where:

[0048] Adjustable nozzle position: Automatically adjust the spraying position of the adjustable nozzle according to the dirt distribution of the next panel;

[0049] Water flow pressure adjustment: Automatically adjust the water pressure of the high-pressure water gun according to the dirt degree, and at the same time avoid damage to the panel caused by excessive water pressure.

[0050] The image analysis module specifically includes:

[0051] Multispectral image acquisition: The image analysis module uses a multispectral camera to collect the surface images of the panel at different spectral bands or angles, including at least visible light and infrared light images;

[0052] Construct a reflection area prediction model: The image analysis module trains a neural network specifically for reflection area recognition and prediction. This neural network predicts and marks the reflection area based on the image features under different lighting conditions, including reflection intensity, spectral differences between visible light and infrared light images, and morphological features of the reflection area, generates a reflection area layer, and based on the obtained reflection area map, removes it from the original image to obtain a dirt area layer;

[0053] Differential processing: Perform differential processing with the dirt area layer to suppress the influence of reflection on the dirt area, obtain the dirt distribution map of the panel, and at the same time calculate the dirt degree of each dirt area through image intensity information to ensure accurate assessment of the dirt;

[0054] The image analysis module transmits the fused dirt distribution data and dirt degree data to the intelligent control module in real time. The dirt data includes the pollution distribution, dirt degree of the next panel, and preliminary determination information of the cleaning requirements, for the intelligent control module to further analyze and process.

[0055] The multispectral camera captures images of the surface of the solar panel at multiple spectral bands by adjusting the angle and spectral band of the camera, including visible light images and infrared light images. The multispectral camera includes an adjustable band selector that switches within different spectral band ranges to capture the surface reflection characteristics of the panel under different spectra, providing multi-dimensional data support for the identification of dirty areas and the suppression of reflection areas.

[0056] The reflection area prediction model adopts a convolutional neural network structure and is trained with the annotated image data of the solar panel surface. During the training process, the network learns the image features under different lighting conditions, including the reflection intensity, the spectral differences between visible light and infrared light images, and the morphological features of the reflection area. Among them:

[0057] Reflection intensity feature: The neural network extracts the reflection intensity features in visible light and infrared light images, and identifies potential reflection areas through the brightness and intensity differences of the reflected light.

[0058] Spectral difference feature: Using the spectral differences between visible light and infrared light images, the neural network distinguishes between reflection areas and dirty areas. The reflection area appears as different brightness or color differences in the infrared light image.

[0059] Reflection area morphological feature: The neural network analyzes the morphological features of the reflection area to distinguish the reflection area from other interference areas.

[0060] Based on the reflection intensity, the spectral differences between visible light and infrared light images, and the morphological features of the reflection area, the neural network outputs the prediction result of the reflection area, generating a reflection area layer R reflect (x, y), where R reflect (x, y) = 1 indicates that this position is a reflection area, and R reflet (x, y) = 0 indicates a non-reflection area. The generated reflection area layer is differentially processed with the dirty area layer M(x, y). In the dirty area layer M(x, y), M(x, y) = 1 indicates that this position is a dirty area, and M(x, y) = 0 indicates a non-dirty area. Thus, the dirty distribution is obtained. M(x, y) is the representation of the dirty distribution. The differential processing includes suppressing the influence of reflection on the dirty area by overlaying the reflection area layer and the dirty area layer, masking the area R reflect (x, y) marked by the reflection area, removing the interference of the reflected light on the identification of the dirty area, ensuring the accuracy of the identification of the dirty area M(x, y). Finally, the output dirty area layer M(x, y) presents the true dirty distribution of the solar panel after removing the reflection influence, providing accurate data support for subsequent cleaning strategies.

[0061] Reflection area intensity R intensity(x, y) represents the reflection intensity at a certain position (x, y), which is calculated by comparing the brightness differences between visible light and infrared light images, and is expressed by the following formula for reflection intensity:

[0062] R intensity (x, y) = I visible (x, y) - I infrared (x, y)|, where I visible (x, y) is the pixel value of the visible light image at the position (x, y), and I infrared (x, y) is the pixel value of the infrared light image at the position (x, y), and R intensity (x, y) reflects the intensity difference of the reflected light;

[0063] Spectral difference R spectral (x, y) measures the spectral difference by calculating the pixel value differences between the visible light and infrared light bands. The specific calculation is as follows: where I visible (x, y) and I infrared (x, y) are the pixel values of the visible light image and the infrared light image at the position (x, y) respectively, and R spectral (x, y) is the spectral difference value at this position, reflecting the relative difference between the two spectra;

[0064] Extraction of morphological features of the reflection area: Morphological feature R shape (x, y) extracts the morphological features of the reflection area through morphological operations. By calculating the morphological features of this area, the reflection area is detected. The morphological operations include erosion, dilation, opening operation, closing operation, highlighting the boundary of the reflection area.

[0065] The reflection area prediction model is trained using a convolutional neural network. The input of the neural network is multi-spectral data including visible light and infrared light images, and the output is the reflection area layer. The training process is carried out by minimizing the following loss function:

[0066] L reflect = Σ (x,y) |R reflect (x, y) - R groundtruth (x, y)| 2 , where R reflect (x, y) is the reflection area layer predicted by the neural network, and R groundtruth (x, y) is the labeled true reflection area layer, which is obtained through manual annotation. Using the reflection area prediction result output by the neural network, the reflection area layer R reflect (x, y) is generated.

[0067] The differential processing is expressed as: M final (x, y) = M(x, y)·1 - Rreflect (x,y));

[0068] Reflection area layer R reflect In (x,y), 1 indicates that this position is a reflection area, 0 indicates a non-reflection area. In the dirt area layer M(x,y), 1 indicates a dirt area, 0 indicates a non-dirt area. Through differential processing, if R reflect (x,y) = 1, it means that this position is a reflection area. At this time, 1 - R reflect (x,y) = 0. Therefore, the dirt area layer M(x,y) will be suppressed at the reflection area position, ensuring that the interference of the reflection area on the dirt area is removed; if R reflect (x,y) = 0, it means that this position is not a reflection area. At this time, 1 - R reflect (x,y) = 1, and the dirt area layer M(x,y) remains unchanged. M final (x,y) is the dirt area layer after suppressing the reflection area, where the influence of the reflection area is suppressed, ensuring the accurate identification of the dirt area.

[0069] Calculating the dirt degree of each dirt area through image intensity information specifically includes:

[0070] The intensity value of each pixel in the multi-spectral image is I(x,y), where I(x,y) is the brightness value of the current pixel position. The lower the intensity, the more dirt accumulates on the surface;

[0071] Dirt will cause a decrease in the image intensity value (the dirt area is usually darker). Calculate the dirt degree D(x,y):

[0072] Among them, I(x,y) is the image intensity value of the current pixel position, I max is the maximum intensity value in the entire image (for example, 255 for an 8-bit image), I min is the minimum intensity value in the entire image. The dirt degree is quantified through the difference in image intensity. The lower the intensity, the higher the dirt degree;

[0073] To ensure that the dirt degree D(x,y) is within the range of 0 to 1, it is normalized according to the minimum and maximum values of the image. I in the formula max and I min are dynamically determined according to the brightness distribution of the image, ensuring that the calculated D(x,y) value range is [0, 1].

[0074] The cleaning strategy specifically includes:

[0075] Flushing order: Adjust the spraying position of the nozzle according to the dirt distribution, and determine the flushing order according to the dirt degree and dirt distribution;

[0076] Water pressure adjustment: Adjust the water pressure of the high-pressure water gun according to the degree of dirt to ensure the cleaning effect in different areas.

[0077] The flushing sequence includes priority calculation, expressed as:

[0078] P 优 (x, y) = D(x, y) · M(x, y), where P 优 (x, y) is the cleaning priority of area (x, y). The higher the cleaning priority, the area should be cleaned first. Based on the cleaning priority, the cleaning sequence is sorted.

[0079] Water pressure P flow (x, y) adjustment is expressed as: P flow (x, y) = P min +(P max -P min )·D(x, y), where P min and P max are the minimum and maximum values of the water pressure respectively, and D(x, y) is the degree of dirt, which determines the range of water pressure change.

[0080] The present invention covers any substitutions, modifications, equivalent methods and solutions made on the essence and scope of the present invention. In order to enable the public to have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention. However, those skilled in the art can also fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion to the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0081] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. An intelligent flushing system for solar panels in a photovoltaic power station, characterized in that: It includes an image analysis module, a central processing unit, a communication unit, an intelligent control module, and a flushing device; The image analysis module is used to monitor the dirt data on the surface of the solar panel and transmit the dirt data to the intelligent control module in real time. The image analysis module includes a multi-spectral camera, which collects surface images of the current solar panel and the next solar panel. The surface image includes collecting the surface image of the solar panel at different spectral bands or angles, and obtaining the dirt distribution and dirt degree of the next solar panel through reflection area recognition; The central processing unit generates a cleaning strategy based on the cleaning requirements using a preset intelligent algorithm according to the dirt distribution and dirt degree of the next battery panel provided by the image analysis module, and outputs a control instruction, which is sent to the intelligent control module through the communication unit, and the intelligent control module adjusts the working state of the flushing device; The flushing device includes a high-pressure water gun and an adjustable nozzle. Based on the control instruction, the intelligent control module adjusts the position and pressure of the adjustable nozzle of the flushing device in advance, wherein: Adjustable nozzle position: Automatically adjust the spraying position of the adjustable nozzle according to the dirt distribution of the next solar panel; Water flow pressure regulation: Automatically adjust the water pressure of the high-pressure water gun according to the degree of dirtiness, while avoiding damage to the solar panels caused by excessive water pressure.

2. The photovoltaic power station solar panel intelligent flushing system according to claim 1, characterized in that: The image analysis module specifically includes: Multispectral image acquisition: The image analysis module uses a multispectral camera to collect images of the panel surface at different spectral bands or angles, including at least visible light and infrared light images; Constructing a reflection area prediction model: The image analysis module trains a neural network specifically for reflection area recognition and prediction. The neural network predicts and marks the reflection area based on the image features under different lighting conditions, including reflection intensity, visible light and infrared light image spectrum differences, and morphological features of the reflection area. The generated reflection area layer is removed from the original image based on the obtained reflection area map to obtain a dirty area layer; Differentiation processing: Differentiate it with the dirty area layer to suppress the influence of reflection on the dirty area, obtain the dirt distribution map of the solar panel, and calculate the degree of dirtiness of each dirty area through image intensity information.

3. The photovoltaic power station solar panel intelligent flushing system according to claim 2, characterized in that: The multi-spectral camera collects solar panel surface images in multiple spectral bands, including visible light images and infrared light images, by adjusting the angle and spectral band of the camera. The multi-spectral camera includes an adjustable band selector that switches within different spectral bands to capture the surface reflection characteristics of the solar panel under different spectra.

4. The photovoltaic power station solar panel intelligent flushing system according to claim 2, characterized in that: The reflection area prediction model adopts a convolutional neural network structure and is trained through labeled panel surface image data. During the training process, the network learns image features under different lighting conditions, including reflection intensity, spectral differences between visible light and infrared light images, and morphological features of the reflection area, where: Reflection intensity features: The neural network extracts reflection intensity features from visible and infrared images and identifies potential reflection areas based on differences in brightness and intensity of reflected light. Spectral difference feature: Using the spectral difference between visible light and infrared light images, the neural network distinguishes between reflective areas and dirty areas. Reflective areas appear as different brightness or color differences in infrared light images. Reflection area morphological features: The neural network analyzes the morphological features of the reflection area to distinguish the reflection area from other interference areas; Through the reflection intensity, the spectral difference between visible light and infrared light images, and the morphological characteristics of the reflection area, the neural network outputs the prediction result of the reflection area and generates the reflection area layer R reflect (x,y), where R reflect (x, y) = 1 means that the position is a reflection area, R reflect (x, y) = 0 represents a non-reflective area. The generated reflective area layer is differentiated from the dirty area layer M(x, y). In the dirty area layer M(x, y), M(x, y) = 1 indicates that the position is a dirty area, and M(x, y) = 0 indicates a non-dirty area. Thus, the dirt distribution is obtained. M(x, y) is a representation of the dirt distribution. The differentiation process includes superimposing the reflective area layer and the dirty area layer to suppress the influence of reflection on the dirty area and mark the area R of the reflective area. reflect (x, y) is shielded to remove the interference of reflected light on the identification of the dirty area, ensuring the accurate identification of the dirty area M(x, y).

5. The photovoltaic power station solar panel intelligent flushing system according to claim 4, characterized in that: The reflection area prediction model is trained using a convolutional neural network. The input of the neural network is multispectral data including visible light and infrared light images, and the output is a reflection area layer. The training process is carried out by minimizing the following loss function: L reflect =∑ (x,y) |R reflect (x,y)-R groundtruth (x,y)| 2 , where R reflect (x, y) is the reflection area layer predicted by the neural network, R groundtruth (x, y) is the annotated real reflection area layer, which is obtained through manual annotation. The reflection area prediction result output by the neural network is used to generate the reflection area layer R reflect (x,y).

6. The photovoltaic power station solar panel intelligent flushing system according to claim 5, characterized in that: The differentiation process is expressed as: final (x,y)=M(x,y)·(1-R reflect (x,y)); Reflection Area Layer R reflect In (x, y), 1 indicates that the position is a reflective area, and 0 indicates a non-reflective area. In the dirty area layer M(x, y), 1 indicates a dirty area, and 0 indicates a non-dirty area. Through differential processing, if R reflect (x, y) = 1, indicating that the position is a reflection area, at this time 1-R reflect (x, y) = 0, so the dirty area layer M(x, y) will be suppressed at the reflection area position to ensure that the interference of the reflection area on the dirty area is removed; if R reflect (x, y) = 0, indicating that the position is not a reflection area. reflect (x, y) = 1, the dirty area layer M(x, y) remains unchanged.

7. The photovoltaic power station solar panel intelligent flushing system according to claim 4, characterized in that: Calculating the degree of dirtiness of each dirty area by using the image intensity information specifically includes: The intensity value of each pixel in the multispectral image is I(x,y), where I(x,y) is the brightness value of the current pixel position. The lower the intensity, the more dirt is accumulated on the surface. Dirt will cause the image intensity value to decrease. Calculate the degree of dirt D(x,y): Among them, I(x,y) is the image intensity value at the current pixel position, I max is the maximum intensity value in the entire image, I min It is the minimum intensity value in the entire image. The degree of dirtiness is quantified by the difference in image intensity. The lower the intensity, the higher the degree of dirtiness. The value range of D(x,y) is [0, 1].

8. The photovoltaic power station solar panel intelligent flushing system according to claim 4, characterized in that: The cleaning strategy specifically includes: Flushing sequence: adjust the spray position of the nozzle according to the dirt distribution, and determine the flushing sequence according to the degree of dirt and dirt distribution; Water pressure adjustment: According to the degree of dirt, adjust the water pressure of the high-pressure water gun to ensure the cleaning effect of different areas.

9. The photovoltaic power station solar panel intelligent flushing system according to claim 8, characterized in that: The flushing order includes a priority calculation, expressed as: P 优 (x,y)=D(x,y)·M(x,y), where, P 优 (x,y) is the cleaning priority of area (x,y). The higher the cleaning priority, the higher the priority of the area to be cleaned. The cleaning order is sorted based on the cleaning priority.

10. The photovoltaic power station solar panel intelligent flushing system according to claim 9, characterized in that: The water pressure P flow (x,y) adjustment is expressed as: P flow (x,y)=P min +(P max -P min )·D(x,y), where P min and P max are the minimum and maximum values ​​of water pressure respectively, and D(x,y) is the degree of dirtiness, which determines the range of water pressure variation.

Images