Visual detection system for dust deposition of photovoltaic panel based on deep learning and cloud monitoring platform

Abstract

The invention belongs to the technical field of photovoltaic power generation, and discloses a photovoltaic panel dust deposition visual detection system based on deep learning and a cloud monitoring platform, and the detection system comprises a data collection and preprocessing module which is used for collecting and cleaning a photovoltaic panel image, and carrying out the processing of image correction, silver wire removal and dust deposition image synthesis, generating a training test data set; the model construction module is used for constructing a photovoltaic dust deposition detection model; the model training and evaluation module is used for training the photovoltaic dust deposition detection model based on the training test data set and setting evaluation indexes to evaluate the trained photovoltaic dust deposition detection model to obtain an optimal photovoltaic dust deposition detection model; and the detection application module is used for inputting the photovoltaic dust deposition image in the actual working condition into the optimal photovoltaic dust deposition detection model and monitoring the dust deposition condition of the photovoltaic panel in real time. Dust on the surface of the photovoltaic panel can be accurately detected, dust distribution visualization is achieved, and meanwhile real-time monitoring and analysis of data are achieved through the cloud platform.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic power generation, and particularly relates to a photovoltaic panel dust accumulation visualization detection system and a cloud monitoring platform based on deep learning. Background Art

[0002] Photovoltaic panels are usually exposed to the outdoor environment, and the atmosphere is filled with many dust particles. In actual engineering, the surface of photovoltaic panels is prone to dust accumulation, and these dusts will affect the power generation efficiency and safety of the components to varying degrees. First, the dust layer covering the battery panel will hinder the sunlight irradiation, resulting in a decrease in the light transmittance of the glass on the surface of the photovoltaic panel, thereby reducing the amount of solar radiation received by the photovoltaic panel. Second, the dust will also absorb heat and convert it into its own heat energy, further affecting the heat dissipation performance of the cover glass of the photovoltaic panel, and thus affecting the overall working efficiency.

[0003] The existing photovoltaic dust accumulation detection technologies are as follows:

[0004] Direct observation method: This method is the most direct, but it requires a large amount of manual participation and is difficult to carry out in bad weather or at night. And manual observation is easily affected by subjective factors and cannot accurately judge the subtle dust changes.

[0005] Laser scanning method: This method can quickly and accurately obtain the dust coverage of the photovoltaic panel, but it requires the use of high-precision laser scanning instruments, with high costs and high requirements for equipment.

[0006] Image processing method: This method can detect the dust coverage by processing and analyzing the surface image of the photovoltaic panel, but it is easily interfered by factors such as light and shadow, and the processing speed is slow.

[0007] In summary, the existing photovoltaic panel dust accumulation degree detection technologies have the defects of high requirements for equipment, poor analysis accuracy, and low efficiency. Therefore, it is urgent to propose a photovoltaic panel dust accumulation visualization detection system and a cloud monitoring platform based on deep learning. Summary of the Invention

[0008] To solve the above technical problems, the present invention proposes a photovoltaic panel dust accumulation visualization detection system and a cloud monitoring platform based on deep learning, which are used to accurately detect the dust on the surface of the photovoltaic panel and realize the visualization of the dust distribution. At the same time, the real-time monitoring and analysis of data are realized through the cloud platform to solve the problems existing in the above-mentioned prior art.

[0009] To achieve the above object, the present invention provides a photovoltaic panel dust accumulation visualization detection system based on deep learning, including: a data acquisition and preprocessing module, a model construction module, a model training and evaluation module, and a detection application module that are connected in sequence;

[0010] The data acquisition and preprocessing module is used to acquire images of clean photovoltaic panels, and perform image correction, silver wire removal, and synthesis of dust accumulation pictures to generate a training and test data set.

[0011] The model construction module is used to construct a photovoltaic dust accumulation detection model, and the photovoltaic dust accumulation detection model includes a feature extraction layer, four residual blocks, and a fully connected layer.

[0012] The model training and evaluation module is used to train the photovoltaic dust accumulation detection model based on the training and test data set, and set evaluation metrics to evaluate the trained photovoltaic dust accumulation detection model to obtain the optimal photovoltaic dust accumulation detection model.

[0013] The detection application module is used to input the photovoltaic dust accumulation image in the actual working condition into the optimal photovoltaic dust accumulation detection model, and after preprocessing, transmittance calculation, and image post-processing, output the corresponding medium transmission map and average transmittance to monitor the dust accumulation situation of the photovoltaic panel in real time.

[0014] Optionally, the data acquisition and preprocessing module includes an image correction unit, a silver wire removal unit, and a dust accumulation picture synthesis unit.

[0015] The image correction unit is used to select four corresponding points on the source image and target image of the clean photovoltaic panel respectively, construct and solve the perspective transformation equation system to realize image correction.

[0016] The silver wire removal unit is used to convert the corrected image from the RGB color space to the HSV color space, create a binary mask to identify the silver wire according to the color characteristics of the silver wire in the HSV space, and then fill in the missing parts based on the morphological dilation operation and the Navier-Stokes algorithm to remove the silver wire.

[0017] The dust accumulation picture synthesis unit is used to segment the clean photovoltaic panel image into ash-free blocks of preset pixels, randomly generate transmittance, and calculate the ash-containing blocks based on the ash-free blocks through the atmospheric scattering model to generate images with different dust accumulation degrees and their corresponding transmittance data labels.

[0018] Optionally, the feature extraction layer uses multi-scale convolution, and the convolution kernel sizes are 3×3, 5×5, and 7×7 respectively, and 32-channel filters are used for each scale.

[0019] The multi-scale convolution calculation process of the feature extraction layer includes: first performing convolution operations, then successively passing through batch normalization processing and ReLU activation function processing, and finally adding the channel outputs corresponding to the three different-scale convolution kernels to obtain the output of the feature extraction layer.

[0020] Optionally, each residual block includes two 3×3 convolutional kernels, each followed by batch normalization and the ReLU activation function; each residual block also contains a skip connection, a residual mapping, and an SEBlock module;

[0021] The SEBlock module is used to adaptively recalibrate the cross-channel feature responses through Squeeze operation and Excitation operation.

[0022] Optionally, the fully connected layer includes an average pooling layer, two linear transformations, and the ReLU activation function;

[0023] The fully connected layer is used to compress the feature maps extracted from the feature extraction layer and the residual blocks into a scalar based on the average pooling layer, and after being processed by two linear transformations and the ReLU activation function, finally outputs the average transmittance and the medium transmission map.

[0024] Optionally, the model training and evaluation module is used to set the parameters for model training, select an optimizer and a loss function, and use the training and test data sets to train and evaluate the model, and then quantify the model performance through the mean square error, the mean absolute error, and the relative error, and save the model with the minimum loss during the training process as the optimal photovoltaic dust accumulation detection model.

[0025] The present invention also provides a cloud monitoring platform for deploying the above-mentioned deep learning-based photovoltaic panel dust accumulation visualization detection system, including: a data transmission module, a data processing and analysis module, a warning module, and a remote monitoring module;

[0026] The data transmission module is used to connect the detection device with the cloud monitoring platform through Internet of Things technology to realize real-time data transmission and storage;

[0027] The data processing and analysis module is equipped with the above-mentioned deep learning-based photovoltaic panel dust accumulation visualization detection system, and is used to process and analyze the collected data, evaluate the dust accumulation degree of the photovoltaic panel, and generate a dust distribution map and a trend map;

[0028] The warning module is used to automatically send a warning signal when the dust accumulation degree exceeds the set threshold;

[0029] The remote monitoring module is used to view the dust accumulation situation of the photovoltaic panel and the system operation status at any time through the Web end or the mobile application of the cloud monitoring platform.

[0030] The present invention also provides a computer device, including a memory, a processor, and a computer program stored on the memory, and the processor executes the computer program to implement the above-mentioned system.

[0031] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above-described system is implemented.

[0032] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, the above-described system is implemented.

[0033] Compared with the prior art, the present invention has the following advantages and technical effects:

[0034] 1. The prior art solution does not quantitatively evaluate the degree of dust accumulation in photovoltaic strings. The present invention can quantitatively evaluate the degree of dust accumulation on photovoltaic panels and its specific distribution.

[0035] 2. The prior art solution selects an actual control group using actual site resources to measure the degree of dust accumulation on components. The implementation method is relatively complex, and the remote diagnosis method is limited. The present invention uses a deep learning method to achieve remote analysis, with good portability.

[0036] 3. The prior art solution uses a deep learning method to identify the degree of dust accumulation on photovoltaic panels, which requires a large amount of labeled data for training. However, in actual working conditions, it is usually very difficult to obtain enough labeled data, especially photovoltaic panel image data with dust distribution information. The dependence on a large amount of data not only increases the time and cost of data preparation, but also the scarcity of data may lead to a decline in the performance of the model, affecting its generalization ability. The present invention can obtain training set data by adding dust to clean pictures, reducing the dependence on large-scale labeled data.

[0037] 4. The monitoring range of the prior art solution is limited and cannot be monitored in real time. The present invention uses a cloud platform to achieve large-area and real-time monitoring, providing remote monitoring and warning functions, and improving the efficiency of operation and maintenance management. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0039] Figure 1 is the specific implementation flowchart of the monitoring system according to the embodiment of the present invention;

[0040] Figure 2 is the schematic diagram of the data set of the photovoltaic dust detection model according to the embodiment of the present invention;

[0041] Figure 3 is the network structure diagram of the photovoltaic dust detection model according to the embodiment of the present invention;

[0042] Figure 4Input and output images of the photovoltaic dust accumulation detection model according to the embodiments of the present invention. Among them, (a) and (e) are original dust accumulation images, (b) and (f) are images after increasing contrast, (c) and (g) are medium transmission images estimated by the model, and (d) and (h) are result images after Gaussian smoothing processing of the estimated medium transmission maps. Detailed implementation manners

[0043] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0044] It should be noted that the steps shown in the flowchart of the accompanying drawings may be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than here.

[0045] Embodiment 1

[0046] Photovoltaic panels are all used in outdoor environments, and the atmospheric environment is filled with dust particles of various shapes. In actual engineering, the surface of the battery panels of photovoltaic panels is prone to dust accumulation, and these dusts all have different degrees of influence on the power generation performance and use safety of the batteries of photovoltaic panels. On the one hand, the dust falling on the surface of the photovoltaic panel will block sunlight, reduce the transmittance of the glass on the surface of the photovoltaic panel, and reduce the solar radiation on the surface of the battery of the photovoltaic panel; on the other hand, the dust will absorb solar radiation, convert it into its own heat energy, and block the external heat dissipation of the cover glass of the photovoltaic panel. There are many influencing factors for photovoltaic panel dust accumulation. Although the current dust accumulation detection methods for photovoltaic panels can achieve real-time monitoring and are cost-effective, it is difficult to obtain good detection effects in the face of complex dust accumulation patterns and complex environmental conditions.

[0047] As Figure 1 shown, in view of the above problems, this embodiment provides a deep learning-based photovoltaic panel dust accumulation visualization detection system for accurately detecting the dust on the surface of the photovoltaic panel and realizing the visualization of the dust distribution. The model adopts a lightweight design, reduces the complexity of training, optimization and maintenance, and realizes real-time monitoring and warning functions. The system specifically includes: a data acquisition and preprocessing module, a model construction module, a model training and evaluation module, and a detection application module that are connected in sequence;

[0048] The data acquisition and preprocessing module is used to collect clean photovoltaic panel images, and perform image correction, silver wire removal and synthetic dust accumulation picture processing to generate a training and test data set;

[0049] The model construction module is used to construct a photovoltaic dust accumulation detection model, and the photovoltaic dust accumulation detection model includes a feature extraction layer, four residual blocks and a fully connected layer;

[0050] The model training and evaluation module is used to train the photovoltaic dust accumulation detection model based on the training and test data set, and set evaluation indicators to evaluate the trained photovoltaic dust accumulation detection model to obtain the optimal photovoltaic dust accumulation detection model;

[0051] The detection application module is used to input the photovoltaic dust accumulation image in the actual working condition into the optimal photovoltaic dust accumulation detection model, and after preprocessing, transmittance calculation and image post-processing, output the corresponding medium transmission map and average transmittance to monitor the dust accumulation situation of the photovoltaic panel in real time.

[0052] As an implementable way, it specifically includes the following process:

[0053] Step S21: The data acquisition and preprocessing module is used to collect visible light images of clean photovoltaic panels, perform image correction, remove silver wires, and synthesize dust accumulation pictures based on the light attenuation principle of photovoltaic panels as the training and test data set of the model provided in this embodiment.

[0054] Implementably, the data acquisition and preprocessing module includes an image correction unit, a silver wire removal unit, and a dust accumulation picture synthesis unit; among them, the image correction unit is used to select four corresponding points on the source image and the target image of the clean photovoltaic panel respectively, construct and solve the perspective transformation equation set to realize image correction; the silver wire removal unit is used to convert the corrected image from the RGB color space to the HSV color space, create a binary mask to identify silver wires according to the color characteristics of silver wires in the HSV space, and then fill in the missing parts based on the morphological dilation operation and the Navier-Stokes algorithm to remove silver wires; the dust accumulation picture synthesis unit is used to divide the clean photovoltaic panel image into ash-free blocks of preset pixels, randomly generate transmittance, and calculate the ash-containing blocks based on the ash-free blocks through the atmospheric scattering model to generate images with different dust accumulation degrees and their corresponding transmittance data labels.

[0055] Specifically include:

[0056] Step S211: Use the TIR-10 type camera produced by Pinling Company to obtain clean images of photovoltaic panels. The TIR-10 type camera has functions such as 2 million pixels and 10-fold optical autofocus, and the output image resolution is 1920×1080, and the visible light image is transmitted in real time through the network interface.

[0057] Step S212: Perform image preprocessing, including image correction and silver wire removal:

[0058] (1) Image correction is performed using perspective transformation. By selecting four corresponding points on the source image and the target image, a perspective transformation equation system containing eight unknowns is constructed and solved to obtain the transformation parameters, thereby achieving distortion correction of the image. Specifically:

[0059] The image is projected onto a new viewing plane using perspective transformation to correct the image.

[0060]

[0061] (X, Y, Z) are the coordinate points on the original image plane, and the corresponding coordinate points on the transformed image plane are (X 0 , Y 0 , Z 0 ).

[0062] Since the processed image is two-dimensional, we can set z = 1, divide the transformed image coordinates by Z, reduce the image from three dimensions to two dimensions, and then obtain the following equation:

[0063]

[0064] Generally, in this embodiment, m 33 = 1 is set to facilitate obtaining x and y, making the denominator on the left side of equation (3) equal to 1. Expanding the above formula, we get the situation of one point:

[0065]

[0066] There are 8 unknowns (m i j) in equation (3). If we want to solve for these unknowns, we need to list eight sets of equations, that is, four points are manually selected on the source image and the target image respectively, usually the four vertices of the picture.

[0067] Four coordinate points are selected on the source image, which are A: (m 0 , n 0 ), (m 1 , n 1 ), (m 2 , n 2 ), (m 3 , n 3 ).

[0068] Four coordinate points are selected on the target image, which are B: (M 0 ′, N 0 ′), (M 1 ′, N 1 ′), (M 2 ′, N 2 ′), (M 3 ′, N 3 ′).

[0069] Substituting into Equation (3), Equation (4) can be obtained as follows:

[0070]

[0071] (2) Remove the silver lines. Create a binary mask and use morphological methods to remove them:

[0072] The silver lines are part of the electrodes or connection lines in the photovoltaic panel. When extracting image features, the silver lines may be misidentified as important features in the image, thus interfering with the recognition of the true features. Secondly, the silver lines may introduce noise or artifacts, thereby reducing the overall quality of the image. Therefore, it is necessary to remove the silver lines to improve the visual effect of the image and make the subsequent output results more reliable.

[0073] The specific process of removing the silver lines includes: converting the image from the RGB color space to the HSV color space, creating a binary mask based on the color characteristics of the silver lines in the HSV space to identify the silver lines, then performing a morphological dilation operation, and finally filling in the missing parts based on the Navier-Stokes algorithm and the fast marching method.

[0074] The specific operation steps for removing the silver lines are as follows:

[0075] Step 1: Convert the image from the RGB color space to the HSV color space;

[0076] Step 2: According to the specific color characteristics of the silver lines in the HSV space, create a binary mask to accurately identify the pixels within this color range ([0,0,150]-[255,250,255]);

[0077] Step 3: Perform a morphological dilation operation to enhance the effect of the binary mask and ensure that it can completely cover the silver lines to be removed;

[0078] Step 4: Based on the Navier-Stokes algorithm, propagate the information of the damaged area to the surrounding areas. Gradually expand the known area based on the fast marching method to fill in the missing parts.

[0079] Step S213: Prepare a training and test data set, which includes the dust-covered images of photovoltaic panels with different dust accumulation degrees and the corresponding transmittance data labels.

[0080] Furthermore, the dust-covered images are obtained by performing pixel-level operations on the images of clean photovoltaic panels based on the light attenuation principle of photovoltaic panels.

[0081] Even further, the light attenuation principle of photovoltaic panels is obtained by improving the atmospheric scattering model in combination with the actual situation of dust accumulation on the surface area of photovoltaic panels.

[0082] Implementable. Based on the principle of light attenuation of photovoltaic panels, synthetic dust accumulation images are generated. The clean photovoltaic panel images are segmented into small blocks of 16×16 pixels, the transmittance is randomly generated, the dust-free blocks and the global atmospheric light a (set to 1) are given, and the dusty blocks are obtained through formula calculation, so as to obtain the dust accumulation images of photovoltaic panels with different dust accumulation degrees and their corresponding transmittance data labels, and a training and test data set is constructed.

[0083] Implementable. During the process of photographing the dust-accumulated photovoltaic panel images, due to the existence of tiny dust particles in the light transmission path, the light reflected by the photovoltaic panel will be scattered and absorbed. The dust accumulation causes the attenuation of the incident light, which is similar to the influence of haze, dust, etc. in the atmosphere on light. The atmospheric scattering model describes the imaging mechanism under the combined action of haze and light. Therefore, in this embodiment, the atmospheric scattering model is used in combination with the photovoltaic dust accumulation images to describe the attenuation process of light passing through the dust-covered surface.

[0084] The mathematical expression of the atmospheric scattering model is:

[0085] I(m) = J(m)t(m) + a(1 - t(m)) (5)

[0086] t(m) = e -θd(m) (6)

[0087] m is the position of the pixel point, I(m) is the pixel value of the dust-accumulated photovoltaic panel image at point m, J(m) is the pixel value of the clean photovoltaic panel image at point m, t(m) is the medium transmittance at point m, a is the global atmospheric light, and θ is the scattering coefficient of the transmission medium. The transmittance t(m) represents the degree to which light is scattered and absorbed when passing through the dust layer, and the global atmospheric light a represents the overall brightness of the image.

[0088] In formula (6), when d(m) approaches infinity, t(m) approaches 0. Combining with formula (5), we have:

[0089] a = I(m) d(m)→∞ (7)

[0090] In long-distance imaging, d(m) is not infinite, but a very large value. According to formula (6), this makes the transmittance t 1 very small. Therefore, an accurate estimated value of the global atmospheric light a can be obtained based on the following expression:

[0091]

[0092] The principle of light attenuation of photovoltaic panels shows that as long as the medium transmittance is obtained, the conversion between the complete clean image and the dust accumulation image can be achieved.

[0093] Since the image content is independent of the medium transmittance, and the medium transmittance is locally constant, that is, the image pixels in the small block tend to have similar medium transmittances.

[0094] Assume that the transmittance of a single image patch is random. Given a dust-free patch J p (m), the global atmospheric light a, and a random transmittance t ∈ (0, 1), the dusty patch is I p (m) = J p (m)t(m) + a(1 - t(m)), and the atmospheric light a is set to 1.

[0095] The specific operation is to segment the image of the photovoltaic panel into small patches of 16×16 pixels, obtaining 45,522 pictures. Randomly generate t(m) (0 ≤ t(m) ≤ 1) to add dust to each segmented photovoltaic panel image, generating dusty pictures and their corresponding transmittances, thereby obtaining dataset images as Figure 2 shown.

[0096] Step S22: The specific process of the model construction module includes:

[0097] (1) Use the RGB color dusty photovoltaic panel image as the model input, and take the dust concentration deposited on the photovoltaic surface in the form of average transmittance and medium transmission map as the model output.

[0098] (2) The dust detection model consists of a feature extraction layer, four residual blocks, and a fully connected layer.

[0099] Furthermore, the main role of the feature extraction layer is to extract low-level to mid-level feature information from the input dusty photovoltaic panel image, such as edges, textures, and local gray-scale distributions, etc. This layer adopts a multi-scale convolution mechanism, which processes the input image simultaneously through convolution kernels of different scales to capture information in different spatial ranges. This fusion helps to form a more comprehensive and rich feature representation.

[0100] Specifically, multi-scale convolution is adopted in the first layer of the network structure, where the convolution kernel sizes are 3×3, 5×5, and 7×7 respectively, and filters with 32 channels are used for all three scales.

[0101] The multi-scale convolution calculation process of the feature extraction layer is to first perform a convolution operation, then sequentially go through batch normalization processing and ReLU activation function processing, and finally add the channel outputs corresponding to the three different-scale convolution kernels to obtain the output of this layer.

[0102] The specific calculation process is as follows:

[0103] Assume the input feature map X, the output of the convolutional layer (without bias term) is:

[0104] F k×k = W k×k *X (9)

[0105] where Convk×k They respectively represent the convolutional kernel sizes of 3×3, 5×5, and 7×7 convolutions, and * represents the convolution operation.

[0106] Next, the output of batch normalization is:

[0107]

[0108] Among them, ε is a very small positive number used to prevent division by zero. μ and σ 2 are respectively the mean and variance of the output of the convolutional layer. γ k×k , β k×k are learnable parameters, and ⊙ represents element-wise multiplication.

[0109] Then, passing through the ReLU activation layer, the output is:

[0110]

[0111] Finally, the outputs of the three channels are added together to obtain the output of the first layer as:

[0112]

[0113] The second layer of the network consists of four residual blocks. In the field of deep learning, increasing the number of network layers beyond a certain point usually leads to worse training results than expected, and this phenomenon is called network degradation. To solve the problem of network degradation, this embodiment introduces a residual structure:

[0114] The said residual block uses two 3×3 convolutional kernels to extract features, and each convolutional kernel is followed by batch normalization (BN) and the ReLU activation function; in addition, each residual block also includes a skip connection, a residual mapping, and an SEBlock module. The SEBlock attention mechanism adaptively recalibrates the cross-channel feature responses by explicitly modeling the interdependencies between channels.

[0115] This module can construct an arbitrary given transformation F tr : X→U,

[0116] For simplicity, in the following notations, F tr is a standard convolution. Let V = [v 1 , v 2 , …, v c represent the set of learned filter parameters, where v c refers to the parameters of the c-th filter.

[0117] Then the output of F tr can be written as U = [u 1 , u 2 , …, u c, where Here, * represents convolution,

[0118] The SEBlock module introduces two steps after the convolution operation: Squeeze and Excitation. The Squeeze operation uses global average pooling of channels to compress the feature map of W×H×C into a feature vector of 1×1×C; the Excitation operation adopts a gating mechanism combined with the Sigmoid activation function, and parameterizes the gating mechanism through two linear transformation layers to recalibrate the feature response across channels. The specific operation process is as follows:

[0119] 1) Squeeze operation:

[0120] F sq The operation is to use global average pooling of channels to directly compress the feature map of W×H×C containing global information into a feature vector of 1×1×C, that is, to turn each two-dimensional channel into a numerical value with a global receptive field. At this time, 1 pixel represents 1 channel, shielding the spatial distribution information and making better use of the correlation between channels.

[0121] Specifically expressed as, the statistic is obtained by shrinking the spatial dimension H×W through U, and the calculation formula for the c-th element of z is:

[0122] 2) Excitation operation:

[0123] To utilize the information aggregated in the Squeeze operation, the following second operation is used to comprehensively capture channel dependencies.

[0124] To achieve this goal, a simple gating mechanism is selected and the Sigmoid activation is used:

[0125] s = F ex (z, W) = σ(W 2 δ(W 1 z)) (14)

[0126] where δ refers to the ReLU function, and

[0127] To limit model complexity and help generalization, two linear transformations are used before and after the non-linear activation function to parameterize the gating mechanism, that is, the ReLU activation function of the dimensionality reduction layer with parameters W 1 and r

[0128] Then there is a layer with parameters W 2The dimensionality increase layer. Then, through the Sigmoid activation function, the transformation U is readjusted to obtain the final output result:

[0129]

[0130] where F scale (u c , s c ) refers to the channel multiplication between the feature map and the scalar s c .

[0131] The third layer of the network is a fully connected layer, whose function is to map the features extracted from the previous feature extraction layer and residual blocks to the final outputs, namely the average transmittance and the medium transmission map.

[0132] The fully connected layer of this embodiment includes an average pooling layer, two linear transformations, and a ReLU activation function.

[0133] Average pooling is a pooling operation that reduces the size of an image or feature map by dividing the image or feature map into different regions and calculating the average value within each region.

[0134] By performing adaptive average pooling on all pixels of each channel, the two-dimensional feature map of each channel is compressed into a scalar.

[0135] Average pooling is expressed as follows:

[0136]

[0137] where the number of channels C is 32, and H and W are the height and width respectively.

[0138] After linear transformation 1, the pooled feature tensor is mapped from a 32-dimensional vector to a 16-dimensional vector. Assuming the input is The output is The specific calculation formula is:

[0139] y = w 1 x + b 1 (17)

[0140] where is the weight matrix, connecting the input and the hidden layer. is the bias vector.

[0141] The output result after passing through the ReLU activation function can be expressed as:

[0142] y′ = ReLU(w 1 x + b 1 ) (18)

[0143] The linear transformation 2 maps the 16-dimensional vector from the previous step to a scalar value (i.e., the output size is 1-dimensional). The calculation method is the same as that of the linear transformation 1, which will not be elaborated here.

[0144] Step S23: The model training and evaluation module is used to set the learning rate, batch size, number of iterations, adaptive moment estimation optimizer (Adam), loss function, training set and validation set, evaluation metrics, and save the optimal model.

[0145] The optimal model includes: recording the training loss and validation loss of each Epoch during the training process of the detection model. Save the weights of the model every 10 Epochs, and select the weights with the minimum loss to obtain the optimal model. Input the photovoltaic dust accumulation image into the optimal model, and output the average transmittance and its medium transmission map to visualize the specific distribution of dust on the photovoltaic panel surface.

[0146] During the model training, the complete dataset is divided into a training set and a validation set at a ratio of 80% and 20%. The filter weights of each layer of the model follow a Gaussian distribution, the learning rate is set to 0.000001, and 500,000 iterations of training are performed on a PC equipped with an Nvidia GeForce GTX780 GPU, with a batch size of 32. The mean squared error (MSE) is used as the loss function.

[0147] The model learns the mapping relationship function between the RGB values and the medium transmittance, and trains I p (m) and the loss function between the corresponding true medium transmittance t to minimize the loss function.

[0148] This embodiment uses the mean squared error (MSE) as the loss function:

[0149]

[0150] The complete dataset is divided into a training set and a validation set, where the training set accounts for 80% of the total dataset.

[0151] The filter weights of each layer of the model follow a Gaussian distribution (mean μ = 0, standard deviation σ = 0.001), and the learning rate is set to 0.000001.

[0152] Training is performed on a PC equipped with an Nvidia GeForce GTX780 GPU, setting 500,000 iterations with a batch size of 32.

[0153] To analyze the effectiveness of the model, MSE, MAE, and RE are used as evaluation metrics.

[0154] MAE is the average of the absolute errors between the predicted values and the true values. It reflects the average deviation between the model's prediction results and the actual results, and reflects the accuracy of the model prediction. It is defined as follows:

[0155]

[0156] MSE is the average of the squares of the errors between the predicted values and the true values. It is more sensitive to larger errors, thus helping to identify the deficiencies of the model. It is defined as follows:

[0157]

[0158] RE is the ratio of the error between the predicted value and the true value to the true value, which reflects the proportion of the error in the true value, provides information on the proportion of the error relative to the true value, and facilitates error assessment in cases of different scales. It is defined as follows:

[0159]

[0160] In the above formula, n represents the number of samples, y i represents the true value of the i-th sample, represents the predicted value of the i-th sample. The smaller the value on the left side of the equation, the higher the prediction accuracy of the model.

[0161] Step S24: The detection application module is used to input photovoltaic panels with different dust accumulation degrees in the obtained actual working conditions into the optimal model, adjust the contrast of the dust accumulation pictures of the photovoltaic panels, calculate the average transmittance of the dust accumulation pictures, and output the corresponding medium transmission maps and perform Gaussian smoothing processing to reduce high-frequency noise and visualize the specific distribution of dust on the surface of the photovoltaic panels.

[0162] Furthermore, the process of adjusting the contrast of the tested dust accumulation images includes:

[0163] Contrast adjustment is a common operation in image processing. By adjusting the difference in brightness between different regions of the image, the visual effect of the image becomes more distinct. When adjusting the contrast, the brightness values of the image are stretched or compressed according to a contrast factor. The larger the contrast factor, the stronger the contrast of the image, and vice versa.

[0164] The basic idea of adjusting the contrast is to perform a non-linear transformation on each pixel value in the image. The adjusted value can be calculated using the following formula:

[0165] I′(x,y)=(I(x,y)-128)×f+128 (23)

[0166] I(x,y): The pixel value of the original image at position I(x,y). I′(x,y) is the adjusted value. The contrast factor f = 2.5, and the contrast is adjusted to 2.5 times the original value.

[0167] Perform Gaussian smoothing on the output medium transmission map to reduce noise by weighted averaging of each pixel and its neighborhood, thereby making the image softer.

[0168] Gaussian smoothing is achieved by weighted averaging each pixel in the image using a Gaussian function. The Gaussian function is a function based on the normal distribution and has the following form:

[0169]

[0170] where (x,y) is the position of the current pixel, and σ is the standard deviation of the Gaussian function, which controls the degree of smoothing. In this embodiment, σ = 8.

[0171] It is completed by applying a Gaussian kernel (i.e., a matrix). The value of each pixel is multiplied by the weighted values of its surrounding pixels (according to the values of the Gaussian kernel) and summed. In this embodiment, the size of the convolution kernel used is 13×13. The high-frequency noise in the medium transmission map is eliminated, the visual continuity of the image is enhanced, and the dust distribution closer to the real situation is obtained.

[0172] Furthermore, the input and output images of this embodiment are given as Figure 4 shown. Among them, (a) and (e) are the original dust-accumulated images, (b) and (f) are the images after increasing the contrast, (c) and (g) are the medium transmission maps estimated by the model, and (d) and (h) are the results after Gaussian smoothing of the estimated medium transmission maps.

[0173] To comprehensively evaluate the effectiveness of the model, this embodiment uses MobileNetV2, MobileNetV2(1.4), resnet50, ShuffleNet(1.5), ShuffleNet(x2), and the model proposed in this embodiment to conduct 10 comparative experiments. The specific network structures of the models are as Figure 3 shown.

[0174] All models use the adaptive moment estimation optimizer, the learning rate is set to 0.000001, the batch size is 32, and the loss function is MSE. Subsequently, this embodiment calculates indicators such as RE, MAE, and MSE between the actual transmittance and the estimated transmittance, and conducts a detailed comparative analysis of the experimental results of several algorithms. The experimental results are shown in Table 1.

[0175] Table 1

[0176] model MAE MSE RE number of parameters MobileNetV2 0.1756 0.0308 0.2277 3504872 MobileNetV2(1.4) 0.0697 0.0048 0.2849 6963136 ResNet50 0.2688 0.0722 0.1988 25557032 ShuffleNet(1.5) 0.0347 0.1864 0.2212 1364608 ShuffleNet(x2) 0.0238 0.0056 0.2879 2385184 the said model 0.0211 0.0004 0.0468 83809

[0177] Specifically, ShuffleNet and MobileNetV2 (1.4) perform better than some other models in terms of the MAE metric. However, compared with the said models, the MAE of these models is still relatively high, indicating that the said models are more prominent in the accuracy of predicting the dust amount.

[0178] The performances of two versions of MobileNetV2, ShuffleNet (x2), and ResNet50 in terms of MSE are all below 0.1, which shows that these models can fit the data well to a certain extent and reduce the sum of squared errors. However, the relative errors of these models are much higher than those of the said models.

[0179] Two versions of ShuffleNet and MobileNetV2 (1.4) perform well in terms of MAE, but their RE metrics all exceed 0.2, indicating that the RE of these models is relatively large and the prediction stability and accuracy are relatively low.

[0180] In contrast, the relative error of the model proposed in this embodiment is much lower than 0.2, and the prediction results are more accurate and stable.

[0181] In terms of the three metrics of MAE, MSE, and RE, the model proposed in this embodiment significantly outperforms other neural network models. At the same time, the number of parameters is much lower than that of other algorithms, and the calculation speed is improved.

[0182] All of the above illustrate the effectiveness and superiority of the model proposed in this embodiment in the photovoltaic panel dust detection experiment.

[0183] This embodiment also provides a cloud monitoring platform for deploying the above-mentioned deep learning-based photovoltaic panel dust accumulation visualization detection system, including: a data transmission module, a data processing and analysis module, an early warning module, and a remote monitoring module;

[0184] Step S31: The data transmission module, including: a network interface, a communication protocol (TCP / IP), data encryption, data buffering, and status monitoring.

[0185] It is used to connect the detection device to the cloud platform through Internet of Things technology, realize the real-time transmission and storage of data, and ensure the stability and security of data transmission. Specifically as follows:

[0186] Provide multiple network connection options to ensure stable data transmission to the cloud platform. Select the appropriate communication protocol according to the network environment to ensure the efficiency and stability of data transmission. Encrypt the data during transmission to prevent data leakage or tampering. Use a buffer to temporarily store data during data transmission to prevent data loss due to network problems. Monitor the status of data transmission in real time, automatically handle transmission interruptions or errors, and notify the operation and maintenance personnel in a timely manner.

[0187] Step S32: The data processing and analysis module includes: data storage, data cleaning, and model deployment.

[0188] It is used to process and analyze the collected data, evaluate the dust accumulation degree of the photovoltaic panel, and generate a dust distribution map and a trend chart. Specifically as follows:

[0189] Store the collected data in the cloud database for subsequent processing and analysis. Remove the noise and outliers in the data to improve the data quality and provide an accurate data basis for subsequent analysis. Deploy the trained deep learning model to the cloud platform to analyze the real-time collected dust accumulation images and output the quantitative results of the dust accumulation degree.

[0190] Step S33: The warning module includes: threshold setting, warning rules, notification system, and log recording.

[0191] It is used to automatically send a warning signal when the dust accumulation degree exceeds the set threshold to remind the operation and maintenance personnel to perform cleaning and maintenance in a timely manner. Specifically as follows:

[0192] Set a reasonable dust accumulation degree threshold according to the operating status and historical data of the photovoltaic panel. Define the warning rules when the dust accumulation degree exceeds the threshold to ensure timely notification of the operation and maintenance personnel. Notify the operation and maintenance personnel in a timely manner through multiple methods such as text messages, emails, and apps to ensure that problems can be handled in a timely manner. Record the warning events and handling status for subsequent analysis and auditing, providing data support for operation and maintenance management.

[0193] Step S34: The remote monitoring module includes: Web application, mobile application, user management, real-time monitoring, and historical data query

[0194] The operation and maintenance personnel can view the dust accumulation situation and the system operation status of the photovoltaic panel at any time through the Web end or mobile application of the cloud platform. Specifically as follows:

[0195] Provide an intuitive web interface to facilitate operation and maintenance personnel to view and manage data on the computer. Provide a mobile application to enable operation and maintenance personnel to view and manage data anytime, anywhere. Support multi-user login, and users with different permissions can view and manage different data, improving the flexibility of operation and maintenance management. Real-time display of the dust accumulation situation of photovoltaic panels and the system operation status to ensure that operation and maintenance personnel can promptly discover problems. Provide a historical data query function to facilitate operation and maintenance personnel to analyze and review data and provide support for operation and maintenance decision-making.

[0196] Through the design and implementation of the above modules, the cloud monitoring platform can achieve efficient and accurate monitoring and analysis of the dust accumulation situation of photovoltaic panels, providing strong technical support for the operation and maintenance management of photovoltaic power stations.

[0197] Embodiment 2

[0198] This embodiment also provides a computer device, including a memory, a processor, and a computer program stored on the memory. The processor executes the computer program to implement the system described above.

[0199] Embodiment 3

[0200] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the system described above is implemented.

[0201] Embodiment 4

[0202] This embodiment also provides a computer program product, including a computer program. When the computer program is executed by a processor, the system described above is implemented.

[0203] The above are only the preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A photovoltaic panel dust accumulation visualization detection system based on deep learning, characterized in that: include: A data acquisition and preprocessing module, a model building module, a model training and evaluation module, and a detection application module connected in sequence; The data acquisition and preprocessing module is used to acquire clean photovoltaic panel images, and perform image correction, silver line removal and dust accumulation image synthesis to generate a training test data set; The model building module is used to build a photovoltaic dust accumulation detection model, and the photovoltaic dust accumulation detection model includes a feature extraction layer, four residual blocks and a fully connected layer; The model training and evaluation module is used to train the photovoltaic dust accumulation detection model based on the training test data set, and set evaluation indicators to evaluate the trained photovoltaic dust accumulation detection model to obtain the optimal photovoltaic dust accumulation detection model; The detection application module is used to input the photovoltaic dust accumulation image in actual working conditions into the optimal photovoltaic dust accumulation detection model, and after preprocessing, transmittance calculation and image post-processing, output the corresponding medium transmission map and average transmittance, so as to monitor the dust accumulation of photovoltaic panels in real time.

2. The system according to claim 1, characterized in that The data acquisition and preprocessing module includes an image correction unit, a silver line removal unit and a dust accumulation picture synthesis unit; The image correction unit is used to select four corresponding points on the source image and the target image of the clean photovoltaic panel, respectively, to construct and solve the perspective transformation equation group, and realize image correction; The silver line removal unit is used to convert the corrected image from the RGB color space to the HSV color space, create a binary mask to identify the silver line according to the color characteristics of the silver line in the HSV space, and then fill the missing part based on the morphological dilation operation and the Navier-Stokes algorithm, so as to remove the silver line; The synthetic dust accumulation picture unit is used to divide the clean photovoltaic panel image into dust-free blocks of preset pixels and randomly generate transmittance. The dust-containing blocks are obtained by calculating the atmospheric scattering model based on the dust-free blocks, thereby generating images with different dust accumulation degrees and their corresponding transmittance data labels.

3. The system according to claim 1, characterized in that The feature extraction layer uses multi-scale convolution, and the convolution kernel sizes are 3×3, 5×5, and 7×7, respectively, and a 32-channel filter is used at each scale; The multi-scale convolution calculation process of the feature extraction layer includes: first performing a convolution operation, then sequentially performing batch normalization processing and ReLU activation function processing, and finally adding the channel outputs corresponding to three convolution kernels of different scales to obtain the output of the feature extraction layer.

4. The system according to claim 1, characterized in that Each residual block consists of two 3×3 convolution kernels, each of which is followed by batch normalization and ReLU activation function; each residual block also contains skip connection, residual mapping and SEBlock module; The SEBlock module is used to adaptively recalibrate feature responses across channels through Squeeze and Excitation operations.

5. The system according to claim 1, characterized in that The fully connected layer includes an average pooling layer, two linear transformations and a ReLU activation function; The fully connected layer is used to compress the feature map extracted from the feature extraction layer and the residual block into a scalar based on the average pooling layer, and finally outputs the average transmittance and the medium transmission map after two linear transformations and ReLU activation function processing.

6. The system according to claim 1, characterized in that The model training and evaluation module is used to set the parameters of model training, select the optimizer and loss function, and use the training and test data sets to perform model training and evaluation. The model performance is then quantified by the mean square error, mean absolute error, and relative error. The model with the smallest loss during the training process is saved as the optimal photovoltaic dust accumulation detection model.

7. A cloud monitoring platform, characterized in that: Used to deploy the photovoltaic panel dust accumulation visualization detection system based on deep learning as described in any one of claims 1 to 6, including: a data transmission module, a data processing and analysis module, an early warning module and a remote monitoring module; The data transmission module is used to connect the detection equipment with the cloud monitoring platform through the Internet of Things technology to achieve real-time transmission and storage of data; The data processing and analysis module is equipped with the photovoltaic panel dust accumulation visualization detection system based on deep learning, which is used to process and analyze the collected data, evaluate the degree of dust accumulation of the photovoltaic panel, and generate dust distribution diagrams and trend diagrams; The warning module is used to automatically send out a warning signal when the dust accumulation level exceeds a set threshold; The remote monitoring module is used to check the dust accumulation of photovoltaic panels and the system operation status anytime and anywhere through the web or mobile application of the cloud monitoring platform.

8. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the system according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the system according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the system according to any one of claims 1 to 6 is implemented.

Images