Photovoltaic panel surface dust accumulation detection method and device and storage medium

Abstract

The invention provides a photovoltaic panel surface accumulated dust detection method and device and a storage medium, and belongs to the technical field of photovoltaic panel accumulated dust detection, and the method comprises the steps: dividing a to-be-detected photovoltaic power station into a plurality of accumulated dust detection regions; performing coordinate marking on the accumulated dust detection area to obtain original photovoltaic panel coordinate information; obtaining an original entity photovoltaic panel image from an infrared imager; a virtual photovoltaic power station is constructed through the dust deposition data, the meteorological data and the photovoltaic panel state data; and carrying out image extraction on the virtual photovoltaic power station to obtain an original virtual photovoltaic panel image. The photovoltaic panel surface dust detection device can accurately detect the surface dust on the surface of the photovoltaic panel, effectively improves the detection efficiency of the surface dust on the surface of the photovoltaic panel, realizes fault diagnosis and health management of the photovoltaic power station to be detected, effectively prolongs the service life of the photovoltaic power station to be detected, saves a large amount of manpower, material resources and time cost, and improves the detection efficiency of the photovoltaic power station to be detected. And the labor intensity of power station staff is reduced.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of photovoltaic panel dust accumulation detection, and specifically relates to a photovoltaic panel surface dust accumulation detection method, device and storage medium. Background Art

[0002] In recent years, solar photovoltaic power generation, as a widely used clean energy, plays an important role in the sustainable development of the environment and economy. However, during the operation of solar photovoltaic power stations, due to long-term exposure to complex external environments, a large amount of dust will inevitably accumulate on the surface of photovoltaic panels. The deposition of a large amount of dust will cause shading and temperature effects on the surface of photovoltaic panels, resulting in abnormal heating of the photovoltaic panel surface and obvious hot spots, which will reduce the power generation efficiency of photovoltaic panels and the service life of the corresponding power station. Therefore, it is particularly important to detect dust accumulation on the surface of photovoltaic panels in a timely, accurate and efficient manner.

[0003] The current existing detection method is to manually use an infrared thermal imager and a lifting platform to regularly detect the photovoltaic panels of the power station to determine whether there is a hot spot phenomenon, so as to take corresponding cleaning measures, or to collect images of the photovoltaic panel surface by drones, use image recognition technology to obtain the amount of dust on the photovoltaic panel surface, and determine whether the photovoltaic panel needs to be cleaned. However, the above two detection methods have the following problems: (1) Manual regular detection of the photovoltaic panel surface may fail to clean the abnormal photovoltaic panel surface in time, causing damage to the photovoltaic panel before cleaning. At the same time, the cost of manpower, material resources and time is too high; (2) The use of drones to collect images of the photovoltaic panel surface and analyze its dust content may cause inaccurate identification of the dust content at non-hot spots on the photovoltaic panel surface, resulting in excessive cleaning and waste of resources; (3) The above two detection methods cannot accurately locate each photovoltaic panel, resulting in poor cleaning effect and low efficiency when cleaning the abnormal photovoltaic panel surface. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method, device and storage medium for detecting dust accumulation on the surface of a photovoltaic panel in view of the deficiencies in the prior art.

[0005] The technical solution of the present invention to solve the above technical problem is as follows: A method for detecting dust accumulation on the surface of a photovoltaic panel comprises the following steps:

[0006] S1: Divide the photovoltaic power station to be inspected into multiple dust detection areas;

[0007] S2: Marking the coordinates of each of the dust detection areas respectively to obtain original photovoltaic panel coordinate information corresponding to each of the dust detection areas;

[0008] S3: obtaining an original physical photovoltaic panel image corresponding to each of the dust detection areas from an infrared imager;

[0009] S4: importing a plurality of dust deposition data, a plurality of meteorological data and a plurality of photovoltaic panel status data, and constructing a virtual photovoltaic power station corresponding to the photovoltaic power station to be detected through all the dust deposition data, all the meteorological data and all the photovoltaic panel status data;

[0010] S5: extracting images of the virtual photovoltaic power station to obtain original virtual photovoltaic panel images corresponding to each of the dust detection areas;

[0011] S6: constructing a training model, and performing model analysis on the training model according to all the original photovoltaic panel coordinate information, all the original virtual photovoltaic panel images, and all the original physical photovoltaic panel images to obtain a dust accumulation detection model;

[0012] S7: Using the dust accumulation detection model, respectively detect the original photovoltaic panel coordinate information corresponding to each of the dust accumulation detection areas and the original virtual photovoltaic panel image corresponding to each of the dust accumulation detection areas to obtain detection results of each of the dust accumulation detection areas.

[0013] Another technical solution of the present invention to solve the above technical problem is as follows: a photovoltaic panel surface dust detection device, comprising:

[0014] A division module, used to divide the photovoltaic power station to be inspected into multiple dust accumulation detection areas;

[0015] A coordinate marking module, used to mark the coordinates of each of the dust accumulation detection areas respectively, and obtain original photovoltaic panel coordinate information corresponding to each of the dust accumulation detection areas;

[0016] A physical image acquisition module, used to obtain an original physical photovoltaic panel image corresponding to each of the dust accumulation detection areas from an infrared imager;

[0017] Import module, used to import multiple dust deposition data, multiple meteorological data and multiple photovoltaic panel status data;

[0018] A virtual power station construction module, used to construct a virtual photovoltaic power station corresponding to the photovoltaic power station to be detected through all the dust deposition data, all the meteorological data and all the photovoltaic panel status data;

[0019] A virtual image extraction module, used to extract images of the virtual photovoltaic power station to obtain original virtual photovoltaic panel images corresponding to each of the dust detection areas;

[0020] A model analysis module is used to construct a training model, and to perform model analysis on the training model according to all the original photovoltaic panel coordinate information, all the original virtual photovoltaic panel images, and all the original physical photovoltaic panel images to obtain a dust accumulation detection model;

[0021] The detection result acquisition module is used to detect the original photovoltaic panel coordinate information corresponding to each dust detection area and the original virtual photovoltaic panel image corresponding to each dust detection area through the dust detection model to obtain the detection result of each dust detection area.

[0022] Based on the above-mentioned method for detecting dust accumulation on the surface of a photovoltaic panel, the present invention also provides a system for detecting dust accumulation on the surface of a photovoltaic panel.

[0023] Another technical solution of the present invention to solve the above-mentioned technical problem is as follows: a photovoltaic panel surface dust detection system comprises a memory, a processor and a computer program stored in the memory and executable on the processor; when the processor executes the computer program, the photovoltaic panel surface dust detection method as described above is implemented.

[0024] Based on the above-mentioned method for detecting dust accumulation on the surface of a photovoltaic panel, the present invention also provides a computer-readable storage medium.

[0025] Another technical solution of the present invention to solve the above technical problem is as follows: a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned method for detecting dust on the surface of a photovoltaic panel is implemented.

[0026] The beneficial effects of the present invention are as follows: by dividing the photovoltaic power station to be inspected into dust accumulation detection areas, the coordinates of the dust accumulation detection areas are marked to obtain the original photovoltaic panel coordinate information, the original physical photovoltaic panel image is obtained from the infrared imager, and a virtual photovoltaic power station is constructed by dust deposition data, meteorological data and photovoltaic panel status data. The image of the virtual photovoltaic power station is extracted to obtain the original virtual photovoltaic panel image, and a dust accumulation detection model is obtained by model analysis of the training model according to the original photovoltaic panel coordinate information, the original virtual photovoltaic panel image and the original physical photovoltaic panel image. The detection result is obtained by detecting the original photovoltaic panel coordinate information and the original virtual photovoltaic panel image with the dust accumulation detection model, which can accurately realize the detection of dust on the surface of the photovoltaic panel, effectively improve the detection efficiency of dust on the surface of the photovoltaic panel, realize fault diagnosis and health management of the photovoltaic power station to be inspected, effectively improve the service life of the photovoltaic power station to be inspected, and at the same time save a lot of manpower, material and time costs, and reduce the labor intensity of power station employees. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1A schematic diagram of a process for detecting dust accumulation on the surface of a photovoltaic panel provided by an embodiment of the present invention;

[0028] Figure 2 A schematic diagram of the coordinate division of each dust accumulation detection area in a photovoltaic power station to be detected in a method for detecting dust accumulation on the surface of a photovoltaic panel provided by an embodiment of the present invention;

[0029] Figure 3 A schematic diagram of force analysis of dust particles on the surface of a photovoltaic panel in a method for detecting dust accumulation on the surface of a photovoltaic panel provided by an embodiment of the present invention;

[0030] Figure 4 A schematic diagram of a model structure of a method for detecting dust accumulation on the surface of a photovoltaic panel provided by an embodiment of the present invention;

[0031] Figure 5 A schematic diagram of a model construction of a method for detecting dust accumulation on the surface of a photovoltaic panel provided by an embodiment of the present invention;

[0032] Figure 6 A module block diagram of a photovoltaic panel surface dust detection device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0034] Figure 1 A schematic flow chart of a method for detecting dust accumulation on the surface of a photovoltaic panel provided in an embodiment of the present invention.

[0035] like Figure 1 and 2 As shown, a method for detecting dust accumulation on the surface of a photovoltaic panel comprises the following steps:

[0036] S1: Divide the photovoltaic power station to be inspected into multiple dust detection areas;

[0037] S2: Marking the coordinates of each of the dust detection areas respectively to obtain original photovoltaic panel coordinate information corresponding to each of the dust detection areas;

[0038] S3: obtaining an original physical photovoltaic panel image corresponding to each of the dust detection areas from an infrared imager;

[0039] S4: importing a plurality of dust deposition data, a plurality of meteorological data and a plurality of photovoltaic panel status data, and constructing a virtual photovoltaic power station corresponding to the photovoltaic power station to be detected through all the dust deposition data, all the meteorological data and all the photovoltaic panel status data;

[0040] S5: extracting images of the virtual photovoltaic power station to obtain original virtual photovoltaic panel images corresponding to each of the dust detection areas;

[0041] S6: constructing a training model, and performing model analysis on the training model according to all the original photovoltaic panel coordinate information, all the original virtual photovoltaic panel images, and all the original physical photovoltaic panel images to obtain a dust accumulation detection model;

[0042] S7: Using the dust accumulation detection model, respectively detect the original photovoltaic panel coordinate information corresponding to each of the dust accumulation detection areas and the original virtual photovoltaic panel image corresponding to each of the dust accumulation detection areas to obtain detection results of each of the dust accumulation detection areas.

[0043] It should be understood that the virtual photovoltaic power station is obtained by constructing a model for all the meteorological data and all the photovoltaic panel status data using the ANSYS Fluent tool.

[0044] It should be understood that the image of the virtual photovoltaic power station is extracted by using the infrared imaging technology principle.

[0045] It should be understood that the dust deposition data may be a dust deposition criterion.

[0046] Specifically, sensor technology is used to collect real-time status data of photovoltaic power stations, such as meteorological data such as light intensity, ambient temperature, relative humidity, wind speed, and current and voltage of photovoltaic panels (i.e., photovoltaic panel status data). And through 5G technology combined with the data transmission method of message queue, it is transmitted in real time to the virtual entity of the photovoltaic power station to be built. Secondly, based on the physical structure and working principle of the physical entity of the photovoltaic power station itself and the collision-adhesion dynamic behavior criterion between the surface of the photovoltaic panel and the dust particles, a virtual entity of the photovoltaic power station (i.e., a virtual photovoltaic power station) identical to the physical entity of the photovoltaic power station is constructed through platforms such as COMSOL Multiphysics, ANSYS Fluent, PVsyst, Twins, and SolarWinds.

[0047] It should be understood that the photovoltaic panels of the physical entity and virtual entity of the target power station (i.e. the photovoltaic power station to be detected) are divided into dust accumulation identification areas (i.e. dust accumulation detection areas) in a plane grid format, and each dust accumulation identification area is a single photovoltaic panel. Figure 2 As shown, for example: the coordinates of photovoltaic panel No. 1 are (1,1), the coordinates of photovoltaic panel No. 2 are (2,1), and so on, and each photovoltaic panel in the power station is assigned the corresponding coordinates (that is, the original photovoltaic panel coordinate information).

[0048] It should be understood that the infrared imager may be arranged above the photovoltaic power station to be inspected.

[0049] It should be understood that, at each sampling moment, infrared image data (ie, original physical photovoltaic panel image) of each dust accumulation identification area is collected by infrared imaging technology.

[0050] Specifically, the coordinate information (i.e., the original photovoltaic panel coordinate information) and infrared image data (i.e., the original physical photovoltaic panel image and the original virtual photovoltaic panel image) of each dust identification area corresponding to the physical entity and the virtual entity at each sampling moment are respectively sent to their respective CCSAR-Net models (i.e., the training model).

[0051] In the above embodiment, the photovoltaic power station to be inspected is divided into dust detection areas, the coordinates of the dust detection areas are marked to obtain the original photovoltaic panel coordinate information, the original physical photovoltaic panel image is obtained from the infrared imager, and a virtual photovoltaic power station is constructed by dust deposition data, meteorological data and photovoltaic panel status data. The image of the virtual photovoltaic power station is extracted to obtain the original virtual photovoltaic panel image, and the dust detection model is obtained by analyzing the training model according to the original photovoltaic panel coordinate information, the original virtual photovoltaic panel image and the original physical photovoltaic panel image. The detection result is obtained by detecting the original photovoltaic panel coordinate information and the original virtual photovoltaic panel image with the dust detection model, which can accurately realize the detection of dust on the surface of the photovoltaic panel, effectively improve the detection efficiency of dust on the surface of the photovoltaic panel, realize fault diagnosis and health management of the photovoltaic power station to be inspected, effectively improve the service life of the photovoltaic power station to be inspected, and at the same time save a lot of manpower, material and time costs, and reduce the labor intensity of power station employees.

[0052] Optionally, as an embodiment of the present invention, the training model includes a backbone network and a head network.

[0053] The S6 process includes:

[0054] The backbone network is used to extract features of the original photovoltaic panel coordinate information and the original physical photovoltaic panel image corresponding to each dust detection area, respectively, to obtain the coordinate features of the physical photovoltaic panel to be processed corresponding to each dust detection area and the image features of the physical photovoltaic panel to be processed corresponding to each dust detection area;

[0055] The backbone network is used to extract features of each original photovoltaic panel coordinate information and an original virtual photovoltaic panel image corresponding to each dust detection area, so as to obtain a to-be-processed virtual photovoltaic panel coordinate feature corresponding to each dust detection area and a to-be-processed virtual photovoltaic panel image feature corresponding to each dust detection area;

[0056] Through the head network, mapping processing is performed on the coordinate features of each of the to-be-processed physical photovoltaic panels and the image features of the to-be-processed physical photovoltaic panels corresponding to each of the dust detection areas, so as to obtain a physical photovoltaic panel feature matrix corresponding to each of the dust detection areas and a physical photovoltaic panel prediction classification result corresponding to each of the dust detection areas;

[0057] Through the head network, each of the coordinate features of the virtual photovoltaic panel to be processed and the image features of the virtual photovoltaic panel to be processed corresponding to each of the dust detection areas are mapped and processed to obtain a virtual photovoltaic panel feature matrix corresponding to each of the dust detection areas and a virtual photovoltaic panel prediction classification result corresponding to each of the dust detection areas;

[0058] Importing the real classification results corresponding to each of the dust detection areas, calculating the target loss value for all the physical photovoltaic panel feature matrices, all the virtual photovoltaic panel feature matrices, all the physical photovoltaic panel prediction classification results, all the virtual photovoltaic panel prediction classification results and all the real classification results, and obtaining the target loss value;

[0059] Determine whether the target loss value is greater than or equal to a preset threshold value. If not, return to S2; if so, update the parameters of the training model according to the target loss value to obtain a dust accumulation detection model.

[0060] It should be understood that the preset threshold may be a threshold for hot spot formation.

[0061] It should be understood that the data processing steps of performing feature extraction on each of the original photovoltaic panel coordinate information and the original virtual photovoltaic panel image corresponding to each of the dust detection areas through the backbone network are the same as the data processing steps of performing feature extraction on each of the original photovoltaic panel coordinate information and the original physical photovoltaic panel image corresponding to each of the dust detection areas through the backbone network, with only the processed data being different.

[0062] It should be understood that the data processing steps of mapping the coordinate features of each virtual photovoltaic panel to be processed and the image features of the virtual photovoltaic panel to be processed corresponding to each dust detection area through the head network are the same as the data processing steps of mapping the coordinate features of each physical photovoltaic panel to be processed and the image features of the physical photovoltaic panel to be processed corresponding to each dust detection area through the head network, except that the processed data are different.

[0063] Specifically, the model is divided into two parts: the backbone network and the head network. The backbone network is used to extract high-level features from the input data, and the head network maps the high-level features extracted by the backbone network into the final output, thereby realizing the identification of dust on the surface of photovoltaic panels.

[0064] In the above embodiment, a dust accumulation detection model is obtained by analyzing the model of the training model according to the original photovoltaic panel coordinate information, the original virtual photovoltaic panel image and the original physical photovoltaic panel image, thereby realizing the recognition of dust on the surface of the photovoltaic panel, effectively improving the detection efficiency of dust on the surface of the photovoltaic panel, realizing fault diagnosis and health management of the photovoltaic power station to be detected, and effectively improving the service life of the photovoltaic power station to be detected.

[0065] Optionally, as an embodiment of the present invention, the backbone network includes a plurality of residual blocks arranged in sequence,

[0066] The process of extracting features from the original photovoltaic panel coordinate information and the original physical photovoltaic panel images corresponding to the dust detection areas through the backbone network to obtain the physical photovoltaic panel coordinate features to be processed corresponding to the dust detection areas and the physical photovoltaic panel image features to be processed corresponding to the dust detection areas includes:

[0067] The first residual block is used to perform feature extraction on the original photovoltaic panel coordinate information and the original physical photovoltaic panel image corresponding to each dust detection area, and the result of the feature extraction is input into the second residual block until the last residual block is passed, thereby obtaining the coordinate features of the physical photovoltaic panel to be processed corresponding to each dust detection area and the image features of the physical photovoltaic panel to be processed corresponding to each dust detection area.

[0068] It should be understood that the backbone network includes multiple residual blocks.

[0069] In the above embodiment, the backbone network is used to perform feature extraction on the original photovoltaic panel coordinate information and the original entity photovoltaic panel image to obtain the coordinate features of the entity photovoltaic panel to be processed and the image features of the entity photovoltaic panel to be processed, thereby realizing the recognition of dust on the surface of the photovoltaic panel, effectively improving the detection efficiency of dust on the surface of the photovoltaic panel, realizing fault diagnosis and health management of the photovoltaic power station to be detected, and effectively improving the service life of the photovoltaic power station to be detected.

[0070] Optionally, as an embodiment of the present invention, the residual block includes a convolution block, an attention mechanism layer and a residual connection layer.

[0071] The process of extracting features from the original photovoltaic panel coordinate information and the original physical photovoltaic panel image corresponding to each dust detection area by using the first residual block comprises:

[0072] The convolution block is used to extract features of each of the original photovoltaic panel coordinate information and the original physical photovoltaic panel image corresponding to each of the dust detection areas, so as to obtain first physical photovoltaic panel coordinate features corresponding to each of the dust detection areas and first physical photovoltaic panel image features corresponding to each of the dust detection areas;

[0073] The attention mechanism layer is used to extract the first physical photovoltaic panel coordinate features and the first physical photovoltaic panel image features corresponding to the dust detection areas, respectively, to obtain the second physical photovoltaic panel coordinate features corresponding to the dust detection areas and the second physical photovoltaic panel image features corresponding to the dust detection areas;

[0074] The residual connection layer is used to extract the coordinate features of each of the second physical photovoltaic panels and the second physical photovoltaic panel image features corresponding to each of the dust detection areas, so as to obtain the coordinate features of the third physical photovoltaic panel corresponding to each of the dust detection areas and the third physical photovoltaic panel image features corresponding to each of the dust detection areas;

[0075] Respectively concatenate each of the second physical photovoltaic panel coordinate features with the corresponding physical third photovoltaic panel coordinate features to obtain fourth physical photovoltaic panel coordinate features corresponding to each of the dust accumulation detection areas;

[0076] Each of the second physical photovoltaic panel image features is spliced ​​with the corresponding third physical photovoltaic panel image feature to obtain a fourth physical photovoltaic panel image feature corresponding to each of the dust accumulation detection areas.

[0077] It should be understood that each residual block consists of a convolution block (Conv Block), an attention mechanism (Shuffle Attention) (i.e., an attention mechanism layer), a residual connection (Residual Connection) (i.e., a residual connection layer) and an addition operation (Addition).

[0078] Specifically, the present invention adopts Shuffle Attention (SA), which effectively combines the spatial attention mechanism and the channel attention mechanism. The specific operation is: SA first divides the channel dimension into multiple sub-features, and then processes them in parallel. For each sub-feature, SA uses Shuffle Unit (Shuffle Unit uses both the spatial attention mechanism and the channel attention mechanism for the sub-features) to extract the feature dependencies in the spatial and channel dimensions, and then aggregates all sub-features. Finally, Channel Shuffle is used to realize information communication between different sub-features. The application of SA uses two attention mechanisms at the same time, which improves the semantic expression ability of dust accumulation features, helps the CCSAR-Net model (i.e., the training model) extract the attention area, and focuses on the dust accumulation features.

[0079] In the above embodiment, the first residual block is used to extract features of the original photovoltaic panel coordinate information and the original physical photovoltaic panel image, which effectively combines the spatial attention mechanism and the channel attention mechanism, improves the semantic expression ability of dust accumulation features, and can help the model extract attention areas and focus on dust accumulation features.

[0080] Optionally, as an embodiment of the present invention, the convolution block includes a first coordinate convolution layer, a first convolution layer, a batch normalization layer and an activation layer,

[0081] The process of extracting features from the original photovoltaic panel coordinate information and the original physical photovoltaic panel images corresponding to the dust detection areas through the convolution block to obtain first physical photovoltaic panel coordinate features corresponding to the dust detection areas and first physical photovoltaic panel image features corresponding to the dust detection areas includes:

[0082] Extracting the features of the coordinate information of each of the original photovoltaic panel coordinate information through the first coordinate convolution layer, to obtain fifth physical photovoltaic panel coordinate features corresponding to each of the dust detection areas;

[0083] Performing image feature extraction on each of the original physical photovoltaic panel images through the first convolution layer to obtain fifth physical photovoltaic panel image features corresponding to each of the dust detection areas;

[0084] Normalizing each of the fifth physical photovoltaic panel coordinate features and the fifth physical photovoltaic panel image features corresponding to each of the dust accumulation detection areas through the batch normalization layer to obtain a sixth physical photovoltaic panel coordinate feature corresponding to each of the dust accumulation detection areas and a sixth physical photovoltaic panel image feature corresponding to each of the dust accumulation detection areas;

[0085] The activation layer is used to map each of the sixth physical photovoltaic panel coordinate features and the sixth physical photovoltaic panel image features corresponding to each of the dust detection areas to obtain the first physical photovoltaic panel coordinate features corresponding to each of the dust detection areas and the first physical photovoltaic panel image features corresponding to each of the dust detection areas.

[0086] It should be understood that in addition to the traditional convolution layer (Conv Layer) (i.e., the first convolution layer), batch normalization layer (Batchnorm Layer) and activation layer (Relu) in the convolution block, a coordinate convolution (CoordConv Layer) (i.e., the first coordinate convolution layer) is introduced.

[0087] Specifically, the coordinate convolution (i.e., the first coordinate convolution layer) adds coordinate information (i, j) to the input data (infrared images of each dust accumulation identification area), so that the network can consider the spatial structure and positional relationship of the input data during the convolution process. The application of coordinate convolution makes better use of the position information of the input data, thereby enhancing the CCSAR-Net model (i.e., the training model)'s ability to perceive the dust accumulation location.

[0088] In the above embodiment, the convolution block is used to perform feature extraction on the original photovoltaic panel coordinate information and the original physical photovoltaic panel image to obtain the first physical photovoltaic panel coordinate feature and the first physical photovoltaic panel image feature, thereby better utilizing the position information of the input data, thereby enhancing the model's perception of the dust accumulation location, effectively improving the detection efficiency of dust on the surface of the photovoltaic panel, realizing fault diagnosis and health management of the photovoltaic power station to be detected, and effectively improving the service life of the photovoltaic power station to be detected.

[0089] Optionally, as an embodiment of the present invention, the residual connection layer includes a second coordinate convolution layer and a second convolution layer,

[0090] The process of extracting the features of the second physical photovoltaic panel coordinates and the second physical photovoltaic panel image features corresponding to the dust detection areas through the residual connection layer to obtain the third physical photovoltaic panel coordinate features corresponding to the dust detection areas and the third physical photovoltaic panel image features corresponding to the dust detection areas includes:

[0091] Extracting the feature information of the coordinate information of each of the second physical photovoltaic panel coordinate features through the second coordinate convolution layer, and obtaining the third physical photovoltaic panel coordinate features corresponding to each of the dust detection areas;

[0092] The second convolutional layer is used to extract image features of each of the second physical photovoltaic panel images to obtain third physical photovoltaic panel image features corresponding to each of the dust detection areas.

[0093] It should be understood that the residual connection (i.e., the residual connection layer) is a mode of coordinate convolution (i.e., the second coordinate convolution layer) plus a traditional convolution layer (i.e., the second convolution layer).

[0094] In the above embodiment, feature extraction is performed on the second entity photovoltaic panel coordinate feature and the second entity photovoltaic panel image feature through the residual connection layer to obtain the third entity photovoltaic panel coordinate feature and the third entity photovoltaic panel image feature, which effectively improves the detection efficiency of dust on the surface of the photovoltaic panel, realizes the fault diagnosis and health management of the photovoltaic power station to be detected, and effectively improves the service life of the photovoltaic power station to be detected.

[0095] Optionally, as an embodiment of the present invention, the head network includes a maximum pooling layer, a discarding layer, a fully connected layer and a classification layer,

[0096] The process of mapping the coordinate features of each of the to-be-processed physical photovoltaic panels and the image features of the to-be-processed physical photovoltaic panels corresponding to each of the dust detection areas through the head network to obtain a physical photovoltaic panel feature matrix corresponding to each of the dust detection areas and a physical photovoltaic panel prediction classification result corresponding to each of the dust detection areas includes:

[0097] The maximum pooling layer is used to perform pooling processing on each of the to-be-processed physical photovoltaic panel coordinate features and the to-be-processed physical photovoltaic panel image features corresponding to each of the dust accumulation detection areas, respectively, to obtain a seventh physical photovoltaic panel coordinate feature corresponding to each of the dust accumulation detection areas and a seventh physical photovoltaic panel image feature corresponding to each of the dust accumulation detection areas;

[0098] Through the discarded layer, feature optimization processing is performed on each of the seventh physical photovoltaic panel coordinate features and the seventh physical photovoltaic panel image features corresponding to each of the dust accumulation detection areas to obtain the eighth physical photovoltaic panel coordinate features corresponding to each of the dust accumulation detection areas and the eighth physical photovoltaic panel image features corresponding to each of the dust accumulation detection areas;

[0099] The eighth physical photovoltaic panel coordinate features corresponding to each of the dust accumulation detection areas and the eighth physical photovoltaic panel image features corresponding to each of the dust accumulation detection areas are spliced ​​through the fully connected layer to obtain a physical photovoltaic panel feature matrix corresponding to each of the dust accumulation detection areas;

[0100] The classification layer is used to classify each of the physical photovoltaic panel feature matrices to obtain a predicted classification result of the physical photovoltaic panel corresponding to each of the dust accumulation detection areas.

[0101] It should be understood that the head network is composed of a maximum pooling layer (Max-Pooling Layer), a dropout layer (DropoutLayer), a fully connected layer (Fully Connected Layer) and a classification layer (Softmax Layer).

[0102] In the above embodiment, the coordinate features of the entity photovoltaic panel to be processed and the image features of the entity photovoltaic panel to be processed are mapped through the head network to obtain the entity photovoltaic panel feature matrix and the entity photovoltaic panel prediction classification results, which effectively improves the detection efficiency of dust on the surface of the photovoltaic panel, realizes the fault diagnosis and health management of the photovoltaic power station to be detected, and effectively improves the service life of the photovoltaic power station to be detected.

[0103] Optionally, as an embodiment of the present invention, the process of calculating the target loss value of all the physical photovoltaic panel feature matrices, all the virtual photovoltaic panel feature matrices, all the physical photovoltaic panel prediction classification results, all the virtual photovoltaic panel prediction classification results and all the real classification results to obtain the target loss value includes:

[0104] The domain adaptive loss value is calculated for each physical photovoltaic panel feature matrix and the virtual photovoltaic panel feature matrix corresponding to each dust accumulation detection area through the first formula, so as to obtain the domain adaptive loss value corresponding to each dust accumulation detection area. The first formula is:

[0105]

[0106] in, is the domain adaptation loss value corresponding to the a-th dust detection area, C is the C-th dust category, m is the total number of dust categories, N is the total number of rows and columns of the physical photovoltaic panel feature matrix, and W i a-s is the photovoltaic panel weight of the i-th row in the physical photovoltaic panel feature matrix corresponding to the a-th dust detection area, W j a-s is the photovoltaic panel weight of the jth column in the physical photovoltaic panel feature matrix corresponding to the ath dust detection area, K() is the Gaussian kernel function, is the physical photovoltaic panel feature of the i-th row in the physical photovoltaic panel feature matrix corresponding to the a-th dust detection area, is the physical photovoltaic panel feature in the jth column of the physical photovoltaic panel feature matrix corresponding to the ath dust detection area, W j a-t is the photovoltaic panel weight of the jth column in the virtual photovoltaic panel feature matrix corresponding to the ath dust detection area, is the jth column of virtual photovoltaic panel features in the physical photovoltaic panel feature matrix corresponding to the ath dust detection area, W ia-t is the photovoltaic panel weight of the i-th row in the virtual photovoltaic panel feature matrix corresponding to the a-th dust detection area, is the virtual photovoltaic panel feature of the i-th row in the physical photovoltaic panel feature matrix corresponding to the a-th dust detection area;

[0107] The cross entropy loss value is calculated for all the predicted classification results of the physical photovoltaic panels, all the predicted classification results of the virtual photovoltaic panels, and all the real classification results by the second formula to obtain the target cross entropy loss value. The second formula is:

[0108] L MRCE =L MRCEs -L MRCEt ,

[0109] in,

[0110] in,

[0111] Among them, L MRCE is the target cross entropy loss value, L MRCEs is the entity cross entropy loss value, L MRCEt is the virtual cross entropy loss value, M is the total number of dust detection areas, is the predicted classification result of the physical photovoltaic panel corresponding to the a-th dust detection area, is the true classification result corresponding to the a-th dust detection area, E C is the regularization penalty coefficient of the Cth dust category, C is the Cth dust category, m is the total number of dust categories, Predict the classification result of the virtual photovoltaic panel corresponding to the a-th dust detection area;

[0112] All the domain adaptation loss values ​​are added to the cross entropy loss value to obtain a target loss value.

[0113] It should be understood that the domain adaptation loss (i.e., the domain adaptation loss value) is calculated from the features extracted from the source domain and the target domain (i.e., the predicted value of the photovoltaic panel) and the corresponding true label (i.e., the true value of the photovoltaic panel), all of which encourage the model to align the features of the source domain and the target domain and reduce the distribution difference between the source domain and the target domain. The boundary regularized cross entropy loss (i.e., the cross entropy loss value) is calculated from the classification results (i.e., the predicted classification results) output by the CCSAR-Net model (i.e., the training model) and the corresponding true label (i.e., the true classification result). The parameters of the CCSAR-Net_1 model and the CCSAR-Net_2 model (i.e., the training model) are continuously adjusted by the domain adaptation loss based on the Gaussian kernel function and the boundary regularized cross entropy loss, so that the CCSAR-Net_2 model (i.e., the training model) can better adapt to the CCSAR-Net_1 model and reach the performance level of the CCSAR-Net_1 model.

[0114] Specifically, the present invention adopts a domain adaptive algorithm based on the Gaussian kernel function, which minimizes the difference between the means of the kernel matrices after dimensionality reduction between different domains, so as to make the source domain s (the feature information of the dust accumulation identification area corresponding to the physical entity) and the target domain t (the feature information of the dust accumulation identification area corresponding to the virtual entity) more consistent in the feature space. The specific process of the algorithm is: calculate the sample weight matrix and the Gaussian kernel matrix, extract the kernel matrices inside the source domain, inside the target domain, and between the source domain and the target domain from the Gaussian kernel matrix, perform dimensionality reduction operations on these matrices and calculate the means respectively, and calculate the final loss according to the sample weights and the kernel matrix after dimensionality reduction. The dust accumulation type defined in the present invention belongs to the binary classification (whether the threshold of hot spot formation is reached). Assuming X s is the sample of the source domain (i.e. the true value of the photovoltaic panel), X t is the sample of the target domain (i.e., the predicted value of the photovoltaic panel), and their weights are W s and W t , H is the feature mapping space of Hilbert, f(·) represents the data feature mapping process, and the domain adaptation loss L based on the Gaussian kernel function GKDA The definitions are as follows:

[0115]

[0116] Where C represents the number of dust categories (i.e. the Cth dust category), N s Represents the number of samples in the source domain, N t represents the number of samples in the target domain, and K(·) represents the Gaussian kernel function used for data feature mapping.

[0117] Specifically, the present invention introduces a boundary regularized cross entropy loss function, which improves the robustness of the decision boundary and enhances the distinction between categories by applying a significant regularization penalty to the boundary of the dust type data. Assuming that the classification network is represented by S(·), then the sample x i The boundary of is expressed as:

[0118]

[0119] In the formula Represents the category y in the classification network output i The output value of category c is expressed as:

[0120]

[0121] Where N c Represents the number of samples of category c, when it satisfies When , the generalization error is the smallest, that is, the larger the sample size of a category, the smaller the distance between the category and the decision boundary. Therefore, the cross entropy loss L with boundary regularization is introduced based on the true label distribution. MRCE The definition is as follows:

[0122] L MRCE =L MRCEs -L MRCEt ,

[0123] L MRCEs or

[0124]

[0125] Where N represents the total number of samples, E c represents the regularization penalty coefficient for category c, Represents the category y in the classification network output i The output value of (i.e., the prediction classification result of the physical photovoltaic panel or the prediction classification result of the virtual photovoltaic panel).

[0126] In the above embodiment, the target loss value is calculated for the physical photovoltaic panel feature matrix, the virtual photovoltaic panel feature matrix, the physical photovoltaic panel prediction classification results, the virtual photovoltaic panel prediction classification results and the real classification results to obtain the target loss value, which can accurately realize the detection of dust on the surface of the photovoltaic panel, effectively improve the detection efficiency of dust on the surface of the photovoltaic panel, realize the fault diagnosis and health management of the photovoltaic power station to be detected, effectively improve the service life of the photovoltaic power station to be detected, and at the same time save a lot of manpower, material and time costs, and reduce the labor intensity of power station employees.

[0128] Optionally, as another embodiment of the present invention, after completing the construction of the virtual entity of the photovoltaic power station and the learning of the CCSAR-Net_2 model (i.e., the dust accumulation detection model), the dust on the surface of the physical entity photovoltaic panel can be cleaned through the digital twin of the photovoltaic power station. The overall framework is as follows: at each sampling moment, the coordinate information of each photovoltaic panel of the virtual entity photovoltaic power station (i.e., the original photovoltaic panel coordinate information) and its corresponding infrared image data (i.e., the original virtual photovoltaic panel image) are obtained, and transmitted to the CCSAR-Net_2 model (i.e., the dust accumulation detection model) in real time through the data transmission method of 5G technology and message queue. The CCSAR-Net_2 model (i.e., the dust accumulation detection model) extracts features from it and outputs the corresponding recognition result (i.e., the detection result) to determine whether the dust on the surface of the photovoltaic panel has reached the threshold for the formation of a hot spot, thereby determining whether the photovoltaic panel needs to be cleaned. The specific contents are as follows: If the output result is "yes", the CCSAR-Net_2 model (i.e., the dust detection model) will also output the coordinate information corresponding to the photovoltaic panel. At the same time, the coordinate information will be passed to the photovoltaic panel surface dust cleaning system through 5G technology and message queues. After receiving the coordinate information, the system will execute the corresponding program to the location specified by the coordinates to complete the cleaning work. If the output result is "no", the CCSAR-Net_2 model (i.e., the dust detection model) will not output the coordinate information of the photovoltaic panel, which means that there is no need to clean the photovoltaic panel. Through the above method, the digital twin of the photovoltaic power station will complete the dust accumulation on the surface of each photovoltaic panel at each sampling time, and at the same time, complete the corresponding cleaning tasks for the photovoltaic panels that need to be cleaned.

[0129] Optionally, as another embodiment of the present invention, the technical problem solved by the present invention is as follows:

[0130] 1. Accurately obtain the location information of abnormal photovoltaic panels. At present, when detecting dust on the surface of photovoltaic panels, due to the high cost of manpower, material resources and time, the detection accuracy is often not accurate to each photovoltaic panel, which will lead to poor cleaning effect and low efficiency when cleaning abnormal photovoltaic panels. Therefore, the present invention is based on the residual network, introduces coordinate convolution, and combines domain adaptive loss and boundary regularized cross entropy loss to construct a CCSAR-Net model. The model can accurately detect dust on the surface of photovoltaic panels and output the location information of abnormal photovoltaic panels, which effectively improves the cleaning effect and cleaning efficiency of dust on the surface of photovoltaic panels.

[0131] 2. Timely and efficient detection of abnormal photovoltaic panels. The existing photovoltaic panel surface dust detection method cannot timely and efficiently complete the detection of dust on the surface of each photovoltaic panel. Based on digital twin technology, the present invention uses sensor technology and message queue data transmission in a 5G environment to obtain various data of photovoltaic power stations in real time and transmit them to the photovoltaic power station virtual entity. Combined with the collision-adhesion dynamic behavior criterion between the photovoltaic panel surface and dust particles and the CCSAR-Net model, a digital twin of the photovoltaic power station is built, and the real-time status of the physical entity of the photovoltaic power station is mapped through the digital twin of the photovoltaic power station. In the digital twin of the photovoltaic power station, the detection of dust on the surface of all photovoltaic panels can be completed in a timely and efficient manner, and abnormal photovoltaic panels can be screened out.

[0132] 3. Accurately and efficiently complete the cleaning of abnormal photovoltaic panels. The coordinate information of abnormal photovoltaic panels output by the digital twin of the photovoltaic power station is transmitted to the photovoltaic panel surface dust cleaning system in real time using 5G technology combined with the data transmission method of the message queue. The system is based on the photovoltaic panel coordinate division rule proposed by the present invention, and accurately locates the abnormal photovoltaic panel through the received coordinate information, completes the cleaning of the abnormal photovoltaic panel surface dust, and effectively improves the cleaning efficiency of the photovoltaic panel surface dust cleaning system.

[0133] Optionally, as another embodiment of the present invention, the inventive point of the present invention is as follows:

[0134] 1. The present invention is based on a residual network, introduces coordinate convolution, and integrates domain adaptive loss based on Gaussian kernel function and cross entropy loss of boundary regularization to construct a CCSAR-Net deep learning model. The model can accurately detect dust on the surface of photovoltaic panels and output the location information of abnormal photovoltaic panels, effectively improving the detection efficiency of dust on the surface of photovoltaic panels. At the same time, the detection accuracy is accurate to each photovoltaic panel, which greatly improves the cleaning effect and efficiency of dust on the surface of photovoltaic panels.

[0135] 2. Based on the digital twin technology, the present invention realizes the integration of the virtual entity of the photovoltaic power station and the CCSAR-Net deep learning model to construct a digital twin of the photovoltaic power station. The digital twin of the photovoltaic power station can timely and efficiently complete the detection of dust on the surface of all photovoltaic panels in the photovoltaic power station, screen out abnormal photovoltaic panels, and transmit their coordinate information to the photovoltaic panel surface dust cleaning system in real time through 5G technology combined with the data transmission method of the message queue, so as to complete the cleaning of the abnormal photovoltaic panel surface dust in a timely, efficient and accurate manner. The digital twin of the photovoltaic power station realizes the fault diagnosis and health management of the photovoltaic power station, effectively improves the service life of the photovoltaic power station, and saves a lot of manpower, material and time costs, and reduces the labor intensity of power station employees.

[0136] Optionally, as another embodiment of the present invention, the present invention is specifically:

[0137] First, sensor technology is used to collect real-time status data of the photovoltaic power station, such as meteorological data such as light intensity, ambient temperature, relative humidity, wind speed, and the current and voltage of the photovoltaic panels. And it is transmitted to the virtual entity of the photovoltaic power station to be built in real time through 5G technology combined with the data transmission method of the message queue. Secondly, based on the physical structure and working principle of the physical entity of the photovoltaic power station itself and the collision-adhesion dynamic behavior criterion between the surface of the photovoltaic panel and the dust particles, a virtual entity of the photovoltaic power station that is the same as the physical entity of the photovoltaic power station is constructed through platforms such as COMSOL Multiphysics, ANSYS Fluent, PVsyst, Twins, and SolarWinds.

[0138] Optionally, as another embodiment of the present invention, Figure 3 As shown, the specific content of the criterion of the collision-adhesion dynamic behavior between the floor panel surface and the dust particles of the present invention is as follows:

[0139] When dust particles move in gas-solid two-phase flow, the fluid drag force F d It is the main force affecting the movement of dust particles. If dust particles are to be deposited on the surface of photovoltaic panels, they will be subject to the collision force F of the photovoltaic panel surface. p In addition, dust particles are also affected by their own gravity F during movement. g and air buoyancy F b The effect is as follows:

[0140]

[0141] In the formula, ρ a and ρ are the density of dust particles and air respectively, in kg / m 3 ; d is the particle diameter, unit is m; m is the mass of the dust particle, unit is kg; u and v are the velocity of the dust particle and the velocity of the flow field, unit is m / s; τ p is the velocity response time of dust particles in the flow field, in seconds; μ is the aerodynamic viscosity, in P a ·s;C D is the resistance coefficient of the airflow around the particle; Re r is the Reynolds number; is the collision coefficient; L is the compression displacement, in μm; L av and L maxare the average and maximum values ​​of L, respectively; x is the compression deformation period; r and R are the radius of the particle where the collision occurs and the radius of the photovoltaic panel surface (for photovoltaic panels, R→∞), respectively, in μm; α1, v1, and E1 are the elastic deformation coefficient, Poisson's ratio, and Young's elastic modulus of the particle, respectively (in Pa); α2, v2, and E2 are the elastic deformation coefficient, Poisson's ratio, and Young's elastic modulus of the photovoltaic panel surface, respectively (in Pa); m * and M are the equivalent mass and the surface mass of the photovoltaic panel respectively (since M>>m, m * ≈m).

[0142] When dust particles collide and adhere to the surface of the photovoltaic panel, they will be subjected to electrostatic force F e , friction force F f and adhesion force F a Effect, among which, the adhesion force F a The van der Waals force F w and capillary force F c The calculation formulas for each force are as follows:

[0143]

[0144] Where ρ a is the density of dust particles, in kg / m 3 ; r is the radius of the dust particle, in μm; e is the relative dielectric constant of air, generally e = 1; e0 is the absolute dielectric constant of air, generally e0 = 8.85×10 -12 F / m; Q is the charge of the particle, in C; α is the distance between the two contacting surfaces, in nm; ξ is the specific charge, which is generally -7×10 for micron particles. -6 C / g;c RH is the relative humidity of air; h is the thickness of water film, in nm; T is the indoor temperature, in K; R g is the molar gas constant, in J / (mol·K); V0 is the molar volume of water at room temperature, in m 3 / mol; H1 and H2 are the Hamaker constants of the two objects in contact in air and water respectively; e m is the equivalent thickness of a monolayer of water molecules when saturated with adsorption, in nm; C B is the water molecule adsorption coefficient.

[0145] When dust particles collide with the surface of the photovoltaic panel, a spatial rectangular coordinate system xyz is established with the collision point as the origin and the z-axis as the normal direction of the photovoltaic panel surface. The forces on the dust particles are as follows: Figure 3 As shown. Where θ is the particle velocity The angle between the direction of and the normal z-axis; is the particle speed The angle between the projection on the xoy plane and the y-axis; δ is the installation inclination angle of the photovoltaic panel.

[0146] According to the kinetic energy theorem, if dust particles are deposited on the surface of a photovoltaic panel, the following conditions must be met simultaneously in the normal and tangential directions of the collision surface:

[0147] Normal:

[0148] F c L c +F w L w +γ+F e L e +F g L g cosδ+E p cos 2 θ-F p L p -F b L b cosδ-F d L d cosθ≥0,

[0149] Tangential:

[0150]

[0151] Where: L c , L w , L e , L g , L p , L b , L d and L f The capillary force F c 、Van der Waals force F w , electrostatic force F e , gravity F g , collision force F p 、Air buoyancy F b , fluid drag F d and friction F f The action distance is in meters; γ is the interface energy when dust particles contact the surface of photovoltaic panels, where γ1 and γ2 are the surface energies of dust particles and the glass cover of the photovoltaic panel, respectively. Their values ​​can be obtained by referring to relevant literature and are in J / m 2 ; A is the projected area of ​​dust particles on the photovoltaic panel surface, unit: m 2 ; E p is the kinetic energy of the particles before collision, in J.

[0152] The above conditions are the basis for judging whether dust particles adhere to the surface of photovoltaic panels when constructing a virtual entity of a photovoltaic power station.

[0153] Optionally, as another embodiment of the present invention, Figures 2 to 5 As shown in FIG. 1 , the present invention establishes a CCSAR-Net (Coordinate Convolutional Shuffle Attention Residual Network) model to realize the recognition of dust on the surface of photovoltaic panels. The specific process is as follows: the photovoltaic panels of the physical entity and the virtual entity of the target power station are divided into dust recognition areas according to a plane grid format. Each dust recognition area is a single photovoltaic panel, such as Figure 2 As shown in the figure, for example: the coordinates of photovoltaic panel No. 1 are (1,1), the coordinates of photovoltaic panel No. 2 are (2,1), and so on, and each photovoltaic panel in the power station is assigned the corresponding coordinates. Subsequently, infrared image data of each dust identification area are collected by infrared imaging technology at each sampling time. Finally, the coordinate information and infrared image data of each dust identification area corresponding to the physical entity and the virtual entity at each sampling time are respectively sent to their respective CCSAR-Net models (the CCSAR-Net model corresponding to the physical entity is the CCSAR-Net_1 model; the CCSAR-Net model corresponding to the virtual entity is the CCSAR-Net_2 model). The CCSAR-Net model is used to extract and classify dust features, and the parameters of the two models are updated by combining domain adaptive loss and boundary regularized cross entropy loss. The domain adaptive loss is calculated from the features extracted from the source domain and the target domain and the corresponding true labels, which encourages the model to align the features of the source domain and the target domain and reduce the distribution difference between the source domain and the target domain. The boundary regularized cross entropy loss is calculated from the classification results output by the CCSAR-Net model and the corresponding true labels. The parameters of CCSAR-Net_1 model and CCSAR-Net_2 model are continuously adjusted by domain adaptive loss based on Gaussian kernel function and cross entropy loss with boundary regularization, so that the CCSAR-Net_2 model can better adapt to the CCSAR-Net_1 model and reach the performance level of the CCSAR-Net_1 model.

[0154] Figure 6 A module block diagram of a photovoltaic panel surface dust detection device provided in an embodiment of the present invention.

[0155] Optionally, as another embodiment of the present invention, Figure 6 As shown, a photovoltaic panel surface dust detection device comprises:

[0156] A division module, used to divide the photovoltaic power station to be inspected into multiple dust accumulation detection areas;

[0157] A coordinate marking module, used to mark the coordinates of each of the dust accumulation detection areas respectively, and obtain original photovoltaic panel coordinate information corresponding to each of the dust accumulation detection areas;

[0158] A physical image acquisition module, used to obtain an original physical photovoltaic panel image corresponding to each of the dust accumulation detection areas from an infrared imager;

[0159] Import module, used to import multiple dust deposition data, multiple meteorological data and multiple photovoltaic panel status data;

[0160] A virtual power station construction module, used to construct a virtual photovoltaic power station corresponding to the photovoltaic power station to be detected through all the dust deposition data, all the meteorological data and all the photovoltaic panel status data;

[0161] A virtual image extraction module, used to extract images of the virtual photovoltaic power station to obtain original virtual photovoltaic panel images corresponding to each of the dust detection areas;

[0162] A model analysis module is used to construct a training model, and to perform model analysis on the training model according to all the original photovoltaic panel coordinate information, all the original virtual photovoltaic panel images, and all the original physical photovoltaic panel images to obtain a dust accumulation detection model;

[0163] The detection result acquisition module is used to detect the original photovoltaic panel coordinate information corresponding to each dust detection area and the original virtual photovoltaic panel image corresponding to each dust detection area through the dust detection model to obtain the detection result of each dust detection area.

[0164] Optionally, another embodiment of the present invention provides a photovoltaic panel surface dust detection system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the photovoltaic panel surface dust detection method described above is implemented. The system may be a computer or other system.

[0165] Optionally, another embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned method for detecting dust accumulation on the surface of a photovoltaic panel is implemented.

[0166] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0167] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0168] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0169] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of the present invention.

[0170] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0171] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0172] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting dust on the surface of a photovoltaic panel, characterized in that: The steps include: S1: Divide the photovoltaic power station to be inspected into multiple dust detection areas; S2: Marking the coordinates of each of the dust detection areas respectively to obtain original photovoltaic panel coordinate information corresponding to each of the dust detection areas; S3: obtaining an original physical photovoltaic panel image corresponding to each of the dust detection areas from an infrared imager; S4: importing a plurality of dust deposition data, a plurality of meteorological data and a plurality of photovoltaic panel status data, and constructing a virtual photovoltaic power station corresponding to the photovoltaic power station to be detected through all the dust deposition data, all the meteorological data and all the photovoltaic panel status data; S5: extracting images of the virtual photovoltaic power station to obtain original virtual photovoltaic panel images corresponding to each of the dust detection areas; S6: constructing a training model, and performing model analysis on the training model according to all the original photovoltaic panel coordinate information, all the original virtual photovoltaic panel images, and all the original physical photovoltaic panel images to obtain a dust accumulation detection model; S7: Using the dust accumulation detection model, respectively detect the original photovoltaic panel coordinate information corresponding to each of the dust accumulation detection areas and the original virtual photovoltaic panel image corresponding to each of the dust accumulation detection areas to obtain detection results of each of the dust accumulation detection areas.

2. The method for detecting dust accumulation on the surface of a photovoltaic panel according to claim 1, characterized in that: The training model includes a backbone network and a head network. The S6 process includes: The backbone network is used to extract features of the original photovoltaic panel coordinate information and the original physical photovoltaic panel image corresponding to each dust detection area, respectively, to obtain the coordinate features of the physical photovoltaic panel to be processed corresponding to each dust detection area and the image features of the physical photovoltaic panel to be processed corresponding to each dust detection area; The backbone network is used to extract features of each original photovoltaic panel coordinate information and an original virtual photovoltaic panel image corresponding to each dust detection area, so as to obtain a to-be-processed virtual photovoltaic panel coordinate feature corresponding to each dust detection area and a to-be-processed virtual photovoltaic panel image feature corresponding to each dust detection area; Through the head network, mapping processing is performed on the coordinate features of each of the to-be-processed physical photovoltaic panels and the image features of the to-be-processed physical photovoltaic panels corresponding to each of the dust detection areas, so as to obtain a physical photovoltaic panel feature matrix corresponding to each of the dust detection areas and a physical photovoltaic panel prediction classification result corresponding to each of the dust detection areas; Through the head network, each of the coordinate features of the virtual photovoltaic panel to be processed and the image features of the virtual photovoltaic panel to be processed corresponding to each of the dust detection areas are mapped and processed to obtain a virtual photovoltaic panel feature matrix corresponding to each of the dust detection areas and a virtual photovoltaic panel prediction classification result corresponding to each of the dust detection areas; Importing the real classification results corresponding to each of the dust detection areas, calculating the target loss value for all the physical photovoltaic panel feature matrices, all the virtual photovoltaic panel feature matrices, all the physical photovoltaic panel prediction classification results, all the virtual photovoltaic panel prediction classification results and all the real classification results, and obtaining the target loss value; Determine whether the target loss value is greater than or equal to a preset threshold value. If not, return to S2; if so, update the parameters of the training model according to the target loss value to obtain a dust accumulation detection model.

3. The method for detecting dust accumulation on the surface of a photovoltaic panel according to claim 2, characterized in that: The backbone network includes a plurality of sequentially arranged residual blocks. The process of extracting features from the original photovoltaic panel coordinate information and the original physical photovoltaic panel images corresponding to the dust detection areas through the backbone network to obtain the physical photovoltaic panel coordinate features to be processed corresponding to the dust detection areas and the physical photovoltaic panel image features to be processed corresponding to the dust detection areas includes: The first residual block is used to perform feature extraction on the original photovoltaic panel coordinate information and the original physical photovoltaic panel image corresponding to each dust detection area, and the result of the feature extraction is input into the second residual block until the last residual block is passed, thereby obtaining the coordinate features of the physical photovoltaic panel to be processed corresponding to each dust detection area and the image features of the physical photovoltaic panel to be processed corresponding to each dust detection area.

4. The method for detecting dust accumulation on the surface of a photovoltaic panel according to claim 3, characterized in that: The residual block includes a convolution block, an attention mechanism layer and a residual connection layer. The process of extracting features from the original photovoltaic panel coordinate information and the original physical photovoltaic panel image corresponding to each dust detection area by using the first residual block comprises: The convolution block is used to extract features of each of the original photovoltaic panel coordinate information and the original physical photovoltaic panel image corresponding to each of the dust detection areas, so as to obtain first physical photovoltaic panel coordinate features corresponding to each of the dust detection areas and first physical photovoltaic panel image features corresponding to each of the dust detection areas; The attention mechanism layer is used to extract the first physical photovoltaic panel coordinate features and the first physical photovoltaic panel image features corresponding to the dust detection areas, respectively, to obtain the second physical photovoltaic panel coordinate features corresponding to the dust detection areas and the second physical photovoltaic panel image features corresponding to the dust detection areas; The residual connection layer is used to extract the coordinate features of each of the second physical photovoltaic panels and the second physical photovoltaic panel image features corresponding to each of the dust detection areas, so as to obtain the coordinate features of the third physical photovoltaic panel corresponding to each of the dust detection areas and the third physical photovoltaic panel image features corresponding to each of the dust detection areas; Respectively concatenate each of the second physical photovoltaic panel coordinate features with the corresponding physical third photovoltaic panel coordinate features to obtain fourth physical photovoltaic panel coordinate features corresponding to each of the dust accumulation detection areas; Each of the second physical photovoltaic panel image features is spliced ​​with the corresponding third physical photovoltaic panel image feature to obtain a fourth physical photovoltaic panel image feature corresponding to each of the dust accumulation detection areas.

5. The method for detecting dust accumulation on the surface of a photovoltaic panel according to claim 4, characterized in that: The convolution block includes a first coordinate convolution layer, a first convolution layer, a batch normalization layer and an activation layer, The process of extracting features from the original photovoltaic panel coordinate information and the original physical photovoltaic panel images corresponding to the dust detection areas through the convolution block to obtain first physical photovoltaic panel coordinate features corresponding to the dust detection areas and first physical photovoltaic panel image features corresponding to the dust detection areas includes: Extracting the features of the coordinate information of each of the original photovoltaic panel coordinate information through the first coordinate convolution layer, to obtain fifth physical photovoltaic panel coordinate features corresponding to each of the dust detection areas; Performing image feature extraction on each of the original physical photovoltaic panel images through the first convolution layer to obtain fifth physical photovoltaic panel image features corresponding to each of the dust detection areas; Normalizing each of the fifth physical photovoltaic panel coordinate features and the fifth physical photovoltaic panel image features corresponding to each of the dust accumulation detection areas through the batch normalization layer to obtain a sixth physical photovoltaic panel coordinate feature corresponding to each of the dust accumulation detection areas and a sixth physical photovoltaic panel image feature corresponding to each of the dust accumulation detection areas; The activation layer is used to map each of the sixth physical photovoltaic panel coordinate features and the sixth physical photovoltaic panel image features corresponding to each of the dust detection areas to obtain the first physical photovoltaic panel coordinate features corresponding to each of the dust detection areas and the first physical photovoltaic panel image features corresponding to each of the dust detection areas.

6. The method for detecting dust accumulation on the surface of a photovoltaic panel according to claim 4, characterized in that: The residual connection layer includes a second coordinate convolution layer and a second convolution layer, The process of extracting the features of the second physical photovoltaic panel coordinates and the second physical photovoltaic panel image features corresponding to the dust detection areas through the residual connection layer to obtain the third physical photovoltaic panel coordinate features corresponding to the dust detection areas and the third physical photovoltaic panel image features corresponding to the dust detection areas includes: Extracting the feature information of the coordinate information of each of the second physical photovoltaic panel coordinate features through the second coordinate convolution layer, and obtaining the third physical photovoltaic panel coordinate features corresponding to each of the dust detection areas; The second convolutional layer is used to extract image features of each of the second physical photovoltaic panel images to obtain third physical photovoltaic panel image features corresponding to each of the dust detection areas.

7. The method for detecting dust accumulation on the surface of a photovoltaic panel according to claim 2, characterized in that: The head network includes a maximum pooling layer, a dropout layer, a fully connected layer, and a classification layer. The process of mapping the coordinate features of each of the to-be-processed physical photovoltaic panels and the image features of the to-be-processed physical photovoltaic panels corresponding to each of the dust detection areas through the head network to obtain a physical photovoltaic panel feature matrix corresponding to each of the dust detection areas and a physical photovoltaic panel prediction classification result corresponding to each of the dust detection areas includes: The maximum pooling layer is used to perform pooling processing on each of the to-be-processed physical photovoltaic panel coordinate features and the to-be-processed physical photovoltaic panel image features corresponding to each of the dust accumulation detection areas, respectively, to obtain a seventh physical photovoltaic panel coordinate feature corresponding to each of the dust accumulation detection areas and a seventh physical photovoltaic panel image feature corresponding to each of the dust accumulation detection areas; Through the discarded layer, feature optimization processing is performed on each of the seventh physical photovoltaic panel coordinate features and the seventh physical photovoltaic panel image features corresponding to each of the dust accumulation detection areas to obtain the eighth physical photovoltaic panel coordinate features corresponding to each of the dust accumulation detection areas and the eighth physical photovoltaic panel image features corresponding to each of the dust accumulation detection areas; The eighth physical photovoltaic panel coordinate features corresponding to each of the dust accumulation detection areas and the eighth physical photovoltaic panel image features corresponding to each of the dust accumulation detection areas are spliced ​​through the fully connected layer to obtain a physical photovoltaic panel feature matrix corresponding to each of the dust accumulation detection areas; The classification layer is used to classify each of the physical photovoltaic panel feature matrices to obtain a physical photovoltaic panel prediction classification result corresponding to each of the dust accumulation detection areas.

8. The method for detecting dust accumulation on the surface of a photovoltaic panel according to claim 2, characterized in that: The process of calculating the target loss value of all the physical photovoltaic panel feature matrices, all the virtual photovoltaic panel feature matrices, all the physical photovoltaic panel prediction classification results, all the virtual photovoltaic panel prediction classification results and all the real classification results to obtain the target loss value includes: The domain adaptive loss value is calculated for each physical photovoltaic panel feature matrix and the virtual photovoltaic panel feature matrix corresponding to each dust accumulation detection area through the first formula, so as to obtain the domain adaptive loss value corresponding to each dust accumulation detection area. The first formula is: in, is the domain adaptation loss value corresponding to the a-th dust detection area, c is the C-th dust category, m is the total number of dust categories, N is the total number of rows and columns of the physical photovoltaic panel feature matrix, is the photovoltaic panel weight of the i-th row in the physical photovoltaic panel feature matrix corresponding to the a-th dust detection area, is the photovoltaic panel weight of the jth column in the physical photovoltaic panel feature matrix corresponding to the ath dust detection area, K() is the Gaussian kernel function, is the physical photovoltaic panel feature of the i-th row in the physical photovoltaic panel feature matrix corresponding to the a-th dust detection area, is the physical photovoltaic panel feature in the jth column of the physical photovoltaic panel feature matrix corresponding to the ath dust detection area, is the photovoltaic panel weight of the jth column in the virtual photovoltaic panel feature matrix corresponding to the ath dust detection area, is the jth column of virtual photovoltaic panel features in the physical photovoltaic panel feature matrix corresponding to the ath dust detection area, is the photovoltaic panel weight of the i-th row in the virtual photovoltaic panel feature matrix corresponding to the a-th dust detection area, is the virtual photovoltaic panel feature of the i-th row in the physical photovoltaic panel feature matrix corresponding to the a-th dust detection area; The cross entropy loss value is calculated for all the predicted classification results of the physical photovoltaic panels, all the predicted classification results of the virtual photovoltaic panels, and all the real classification results by the second formula to obtain the target cross entropy loss value. The second formula is: in, in, C∈{1,…,m}, Among them, L MRCE is the target cross entropy loss value, L MRCEs is the entity cross entropy loss value, L MRCEt is the virtual cross entropy loss value, M is the total number of dust detection areas, is the predicted classification result of the physical photovoltaic panel corresponding to the a-th dust detection area, is the true classification result corresponding to the a-th dust detection area, E C is the regularization penalty coefficient of the Cth dust category, C is the Cth dust category, m is the total number of dust categories, Predict the classification result of the virtual photovoltaic panel corresponding to the a-th dust detection area; All the domain adaptation loss values ​​are added to the cross entropy loss value to obtain a target loss value.

9. A photovoltaic panel surface dust detection device, characterized in that: include: A division module, used to divide the photovoltaic power station to be inspected into multiple dust accumulation detection areas; A coordinate marking module, used to mark the coordinates of each of the dust accumulation detection areas respectively, and obtain original photovoltaic panel coordinate information corresponding to each of the dust accumulation detection areas; A physical image acquisition module, used to obtain an original physical photovoltaic panel image corresponding to each of the dust accumulation detection areas from an infrared imager; Import module, used to import multiple dust deposition data, multiple meteorological data and multiple photovoltaic panel status data; A virtual power station construction module, used to construct a virtual photovoltaic power station corresponding to the photovoltaic power station to be detected through all the dust deposition data, all the meteorological data and all the photovoltaic panel status data; A virtual image extraction module, used to extract images of the virtual photovoltaic power station to obtain original virtual photovoltaic panel images corresponding to each of the dust detection areas; A model analysis module is used to construct a training model, and to perform model analysis on the training model according to all the original photovoltaic panel coordinate information, all the original virtual photovoltaic panel images, and all the original physical photovoltaic panel images to obtain a dust accumulation detection model; The detection result acquisition module is used to detect the original photovoltaic panel coordinate information corresponding to each dust detection area and the original virtual photovoltaic panel image corresponding to each dust detection area through the dust detection model to obtain the detection result of each dust detection area.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for detecting dust accumulation on the surface of a photovoltaic panel as claimed in any one of claims 1 to 8 is implemented.

Images