Method and system for evaluating power loss range of photovoltaic module
By adopting a dual-channel MobileNetV2 multimodal deep learning framework in the power loss range evaluation of photovoltaic modules, combining visible light and infrared images, the problem of imaging quality being affected by environmental factors is solved, and higher evaluation accuracy and adaptability are achieved.
Patent Information
- Application Number
- CN202411523087.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-29
- Publication Date
- 2025-02-18
AI Technical Summary
The accuracy of existing photovoltaic module power loss range evaluation algorithms is difficult to ensure when the imaging quality is affected by light and weather and environmental factors, especially when the installed capacity of photovoltaic systems increases and the application scenarios are diversified.
Using a method based on the dual-channel MobileNetV2 multi-modal deep learning framework, we can build a dual-modal image by acquiring the visible light and infrared images of photovoltaic modules, and use the dual-channel MobileNetV2 feature extraction network to extract visible light and infrared features, and then span through multi-cascade. The modal fusion network is characterized by a multi-scale fusion network, and finally the power loss range is mapped through the classifier.
It improves the accuracy of the power loss range evaluation of photovoltaic modules, can effectively judge the power loss range of photovoltaic modules under different degrees of dust accumulation, and is suitable for diversified photovoltaic system application scenarios.
Smart Images
Figure CN119478332A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of photovoltaic module power loss range assessment, and in particular to a photovoltaic module power loss range assessment method and system based on a dual-channel MobileNetV2 multimodal deep learning framework. Background Art
[0002] The impact of PV module pollution on PV power generation is the third most important factor after irradiance and temperature. Moreover, the pollution of PV modules is increasing over time. In the long run, the PV output power and power generation efficiency may drop to 0. The pollution of PV modules is mainly due to the deposition of pollutants such as dust, surface soil, and particulate matter on the panels of PV modules, which blocks the PV modules from receiving solar radiation energy, thereby significantly reducing the performance of the PV system. Therefore, the pollution of PV modules will lead to a decrease in PV power generation. In addition to performance loss, dust accumulation may also cause other damages to PV panels, such as damage to the surface of PV panels and formation of hot spots on PV modules. These damages will further reduce the performance of PV modules and affect the reliability and service life of PV modules. Therefore, how to quickly and accurately determine the power loss caused by pollution on PV modules and plan cleaning procedures in time according to the power loss has become an important issue in the operation and maintenance of PV systems. In recent years, various PV module power loss range assessment algorithms have emerged, especially deep learning image algorithms that can automatically extract dust distribution state features have been increasingly valued. Most of these algorithms use visible light images. However, the quality of visible light images is greatly affected by environmental factors such as light and weather, which can easily affect the performance of the algorithm. Therefore, with the increasing installed capacity of photovoltaic systems and the more diverse application scenarios of photovoltaic systems, the operation and maintenance of photovoltaic arrays has become more challenging. When photovoltaic modules are working, accurate power loss range assessment is indispensable. Summary of the invention
[0003] The object of the present invention is to provide a method and system for evaluating the power loss range of a photovoltaic module, which method and system are conducive to improving the accuracy of evaluating the power loss range of a photovoltaic module.
[0004] In order to achieve the above object, the technical solution adopted by the present invention is: a method for evaluating the power loss range of a photovoltaic module, comprising the following steps:
[0005] Step S1, obtaining a visible light image and an infrared image of a dust-accumulated component of a photovoltaic component to form a dual-modal image, and simultaneously obtaining the power loss of the corresponding dust-accumulated component, thereby constructing a sample data set for evaluating the power loss range of the photovoltaic component;
[0006] Step S2, preprocessing the dual-modal images in the sample data set, segmenting the photovoltaic module area using an image segmentation model, and then performing image correction through affine transformation;
[0007] Step S3, constructing a power loss range assessment model based on a dual-channel MobileNetV2 multimodal deep learning framework, which extracts visible light and infrared features from dual-modal images through a dual-channel MobileNetV2 feature extraction network, and then performs dual-modal fusion and multi-scale fusion of features through a multi-cascade cross-modal fusion network and a multi-scale fusion network, and then maps out the power loss range through a classifier; training the power loss range assessment model through the preprocessed dual-modal image to obtain a trained power loss range assessment model;
[0008] Step S4: evaluating the power loss range of photovoltaic modules using the trained power loss range evaluation model to accurately determine the power loss range of photovoltaic modules under different dust accumulation levels.
[0009] Furthermore, in step S1, a certain number of visible light images and infrared images of dust-accumulated components are collected from photovoltaic components by an infrared thermal imager, and each visible light image and the corresponding infrared image constitute a pair of dual-modal images; at the same time, a micro-inverter is used to collect the power of the dust-accumulated components, which is recorded as P soil , a micro-inverter of the same specification is used to collect the power of the dust-free module as the reference power, recorded as P ref , then the power loss of the dust accumulation component is expressed as:
[0010] PowerLoss=(1-P soil / P ref )×100%
[0011] Through the above operations, a sample data set with different PV module power loss ranges is constructed.
[0012] Furthermore, in step S2, the visible light image in the dual-modal image is preprocessed by:
[0013] I seg =ψ U2Net (I)
[0014] I revised =Φ AT (I seg )
[0015] Among them, ψ U2Net Indicates U 2 -Net image segmentation model, through U 2 -Net image segmentation model segments the visible light image I to obtain the segmented visible light image Iseg ; Φ AT Represents the affine transformation algorithm, which is used to transform the segmented visible light image I seg Correction is performed to obtain the corrected visible light image I revised ;
[0016] The contour coordinates of the segmented visible light image are applied to the infrared image to segment the infrared image, and then the segmented infrared image is affine transformed to obtain a corrected infrared image.
[0017] Furthermore, in step S3, the power loss range assessment model includes a dual-channel MobileNetV2 feature extraction network, n multi-cascade cross-modal fusion networks, a multi-scale fusion network and a classifier; the power loss range assessment model takes the preprocessed dual-modal image as input, and the dual-channel MobileNetV2 feature extraction network extracts n-scale visible light features and infrared features with photovoltaic component dust distribution information from visible light images and infrared images respectively; then, the extracted n-scale visible light features and infrared features are input into n multi-cascade cross-modal fusion networks in pairs, and n-scale dual-modal fusion features are output; then, the n-scale dual-modal fusion features are input into the multi-scale fusion network, and multi-scale dual-modal fusion features are output; finally, the multi-scale dual-modal fusion features are input into the classifier and mapped into m categories of power loss ranges.
[0018] Furthermore, the classifier maps the multi-scale fusion features into eight categories of power loss ranges: 0-10%, 10-20%, 20-30%, 30-40%, 40-50%, 50-60%, 60-70%, and 70-100%.
[0019] The present invention also provides a photovoltaic module power loss range assessment system, including a memory, a processor, and computer program instructions stored in the memory and executable by the processor. When the processor executes the computer program instructions, the above method can be implemented.
[0020] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a method and system for evaluating the power loss range of photovoltaic modules based on a dual-channel MobileNetV2 multimodal deep learning framework. The method uses the visible light image and infrared image of the photovoltaic module as input, and uses the dual-channel MobileNetV2 feature extraction network to extract n-scale visible light and infrared features with dust distribution information, and then uses a multi-cascade cross-modal fusion network (MCFN) and a multi-scale fusion network (MFN) to perform dual-modal and multi-scale fusion of the obtained feature information, thereby obtaining a power loss range evaluation model based on a dual-channel MobileNetV2 multimodal deep learning framework, and thereby evaluating the power loss range of photovoltaic modules. Compared with the existing algorithms, the present method can effectively realize the power loss range evaluation of photovoltaic modules under different dust accumulation levels, improve the accuracy of the power loss range evaluation of photovoltaic modules, and has strong practicality and broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 is a flow chart of a method implementation of an embodiment of the present invention;
[0022] Figure 2 is an example diagram of image preprocessing in an embodiment of the present invention;
[0023] Figure 3 is an architecture diagram of a power loss range assessment model in an embodiment of the present invention;
[0024] Figure 4 This is an example diagram of dust accumulation on photovoltaic modules in an embodiment of the present invention. DETAILED DESCRIPTION
[0025] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.
[0026] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.
[0027] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.
[0028] like Figure 1As shown, this embodiment provides a method for evaluating the power loss range of a photovoltaic module, comprising the following steps:
[0029] Step S1, obtaining a visible light image and an infrared image of a dust-accumulated component of a photovoltaic component to form a dual-modal image, and simultaneously obtaining the power loss of the corresponding dust-accumulated component, thereby constructing a sample data set for evaluating the power loss range of the photovoltaic component.
[0030] Step S2: pre-process the dual-modal images in the sample data set, segment out a part of the photovoltaic module area using an image segmentation model, and then perform image correction through affine transformation.
[0031] Step S3, constructing a power loss range assessment model based on the dual-channel MobileNetV2 multimodal deep learning framework, which extracts visible light and infrared features from the dual-modal image through the dual-channel MobileNetV2 feature extraction network, and then performs dual-modal fusion and multi-scale fusion of features through a multi-cascade cross-modal fusion network (MCFN) and a multi-scale fusion network (MFN), and then maps out the power loss range through a classifier; the power loss range assessment model is trained by the preprocessed dual-modal image to obtain a trained power loss range assessment model.
[0032] Step S4: The power loss range of the photovoltaic module is evaluated by using the trained power loss range evaluation model, so as to accurately determine the power loss range of the photovoltaic module under different dust accumulation degrees.
[0033] In this method, both the visible light image and the infrared image of the photovoltaic module are used as the input of the power loss range assessment model to avoid the imaging quality of the visible light image being affected by environmental factors such as light and weather, thereby affecting the accuracy of the model.
[0034] In step S1 of this embodiment, a certain number of visible light images and infrared images of dust-accumulated components are collected from photovoltaic components by using a FLIRT420 infrared thermal imager. Each visible light image and the corresponding infrared image form a pair of dual-modal images. At the same time, a micro-inverter is used to collect the power of the dust-accumulated components, which is recorded as P soil , a micro-inverter of the same specification is used to collect the power of the dust-free module as the reference power, recorded as P ref , then the power loss of the dust accumulation component is expressed as:
[0035] PowerLoss=(1-P soil / P ref )×100%
[0036] Through the above operations, a sample data set with different PV module power loss ranges is constructed.
[0037] In step S2 of this embodiment, the visible light image in the dual-modal image is preprocessed by:
[0038] I seg =ψ U2Net (I)
[0039] I revised =Φ AT (I seg )
[0040] Among them, ψ U2Net Indicates U 2 -Net image segmentation model, through U 2 -Net image segmentation model segments the visible light image I to obtain the segmented visible light image I seg ; Φ AT Represents the affine transformation algorithm, which is used to transform the segmented visible light image I seg Correction is performed to obtain the corrected visible light image I revised .
[0041] The contour coordinates of the segmented visible light image are applied to the infrared image to segment the infrared image, and then the segmented infrared image is affine transformed to obtain the corrected infrared image. Figure 2 shown.
[0042] Figure 3 is an architecture diagram of the power loss range assessment model based on the dual-channel MobileNetV2 multimodal deep learning framework in this embodiment. Figure 3 As shown, the power loss range assessment model includes a dual-channel MobileNetV2 feature extraction network, n multi-cascade cross-modal fusion networks, a multi-scale fusion network and a classifier. The power loss range assessment model takes the pre-processed dual-modal image as input, and the two-channel MobileNetV2 feature extraction network extracts features of multiple scales for the visible light image and the infrared image respectively, obtains the visible light features and infrared features of multiple scales containing the dust distribution information of the photovoltaic module, and extracts the visible light features and infrared features of the last n scales from them as the input of the n multi-cascade cross-modal fusion networks. Wherein, n is the number of selected scales. The extracted n-scale visible light features and infrared features are input into the n multi-cascade cross-modal fusion networks in pairs, and the n-scale dual-modal fusion features are output. Then, the n-scale dual-modal fusion features output by the n multi-cascade cross-modal fusion networks are simultaneously input into the multi-scale fusion network, and the multi-scale dual-modal fusion features are output. Finally, the multi-scale dual-modal fusion features are input into the classifier and mapped into the power loss range of m categories.
[0043] The overall range of power loss is 0-100%. In this embodiment, the classifier maps the multi-scale fusion feature into eight categories of power loss ranges: 0-10%, 10-20%, 20-30%, 30-40%, 40-50%, 50-60%, 60-70%, and 70-100%.
[0044] In this embodiment, the sample data set contains images of 8 power loss ranges, namely 0-10%, 10-20%, 20-30%, 30-40%, 40-50%, 50-60%, 60-70%, and 70-100%, with a total of 2133 groups of visible light and infrared image groups, i.e., dual-modal images. In addition, in order to simulate different dust accumulation conditions, visible light and infrared images of photovoltaic modules with two dust distribution conditions, uniform dust accumulation and local dust accumulation, were collected. The two dust accumulation distributions also contain dust coverage of different weights, such as Figure 4 As shown in the figure. From the comparison between visible light and infrared images, it can be seen that visible light images can provide detailed visual information of dust deposition, while infrared images provide thermal radiation information of dust deposition areas, and the two correspond to each other. When the amount of dust is larger and the thickness of the deposition is thicker, the change in the amount of dust can be clearly felt from the visible light image. For infrared images, the severity of dust pollution is presented through the thermal radiation comparison information between the dust deposition area and other areas. Combining the visual information of the two modal images can effectively improve the disadvantage of using visible light images alone being affected by the environment.
[0045] In order to better evaluate the performance of this method, a five-fold cross-validation statistical method was used. The data set was divided into five parts, one of which was retained as test data, and the remaining four were used to train the model. This process was repeated five times, and each subset was used as a test set once, and the rest were used as training sets. This ensures that each sample has participated in testing and training, and finally takes the average of the five results to avoid overfitting and improve the generalization ability of the model. The evaluation indicators used are as follows:
[0046]
[0047] Among them, TP represents the number of true positives, TN represents the number of true negatives, FP represents the number of false positives, and FN represents the number of false negatives. i Indicates the accuracy of the i-th test result, Acc avg Represents the average accuracy of five test results.
[0048] Through ablation experiments, the cross-modal fusion scale, MCFN, MFN and other network parameters of the dual-channel MobileNetV2 multimodal deep learning framework are selected. In order to verify the effectiveness and superiority of the power loss range evaluation model based on the dual-channel MobileNetV2 multimodal deep learning framework proposed in this method, four models with good classification performance are selected for comparative analysis, namely ResNet101, ResNeXt101-32×8d, Swin-Transformer-small, and ConvNeXt-small. Each model has been independently experimented on the visible light or infrared image dataset for 5 times, and the average of the five accuracy rates is taken as the result, as shown in Table 1.
[0049] Table 1 Classification accuracy of each model (%)
[0050]
[0051] The power loss range assessment model based on the dual-channel MobileNetV2 multimodal deep learning framework proposed by this method has an average accuracy of 88.27%, which is higher than other comparison models, indicating that this method has good power loss range assessment capabilities. In addition, the number of parameters of the model proposed by this method is 10.7M, which is smaller than other comparison models, indicating that this method is relatively lightweight and suitable for deployment and use in mobile devices.
[0052] This embodiment also provides a photovoltaic module power loss range assessment system, including a memory, a processor, and computer program instructions stored in the memory and executable by the processor. When the processor executes the computer program instructions, the above method can be implemented.
[0053] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.
[0054] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0055] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0056] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0057] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.
Claims
1. A method for evaluating the power loss range of a photovoltaic module, characterized in that: The steps include: Step S1, obtaining a visible light image and an infrared image of a dust-accumulated component of a photovoltaic component to form a dual-modal image, and simultaneously obtaining the power loss of the corresponding dust-accumulated component, thereby constructing a sample data set for evaluating the power loss range of the photovoltaic component; Step S2, preprocessing the dual-modal images in the sample data set, segmenting the photovoltaic module area using an image segmentation model, and then performing image correction through affine transformation; Step S3, constructing a power loss range assessment model based on a dual-channel MobileNetV2 multimodal deep learning framework, which extracts visible light and infrared features from dual-modal images through a dual-channel MobileNetV2 feature extraction network, and then performs dual-modal fusion and multi-scale fusion of features through a multi-cascade cross-modal fusion network and a multi-scale fusion network, and then maps out the power loss range through a classifier; training the power loss range assessment model through the preprocessed dual-modal image to obtain a trained power loss range assessment model; Step S4: evaluating the power loss range of photovoltaic modules using the trained power loss range evaluation model to accurately determine the power loss range of photovoltaic modules under different dust accumulation levels.
2. A photovoltaic module power loss range assessment method according to claim 1, characterized in that: In step S1, a certain number of visible light images and infrared images of dust-accumulated components are collected from photovoltaic components by using an infrared thermal imager, and each visible light image and the corresponding infrared image form a pair of dual-modal images; at the same time, a micro-inverter is used to collect the power of the dust-accumulated components, which is recorded as P soil , a micro-inverter of the same specification is used to collect the power of the dust-free module as the reference power, recorded as P ref , then the power loss of the dust accumulation component is expressed as: PowerLoss=(1-P soil / P ref )×100% Through the above operations, a sample data set with different PV module power loss ranges is constructed.
3. A photovoltaic module power loss range assessment method according to claim 1, characterized in that: In step S2, the visible light image in the dual-modal image is preprocessed by: I seg =ψ U2Net (I) I revised =Φ AT (I seg ) Among them, ψ U2Net Indicates U 2 -Net image segmentation model, through U 2 -Net image segmentation model segments the visible light image I to obtain the segmented visible light image I seg ; Φ AT Represents the affine transformation algorithm, which is used to transform the segmented visible light image I seg Correction is performed to obtain the corrected visible light image I revised ; The contour coordinates of the segmented visible light image are applied to the infrared image to segment the infrared image, and then the segmented infrared image is affine transformed to obtain a corrected infrared image.
4. A photovoltaic module power loss range assessment method according to claim 1, characterized in that: In step S3, the power loss range assessment model includes a dual-channel MobileNetV2 feature extraction network, n multi-cascade cross-modal fusion networks, a multi-scale fusion network and a classifier; the power loss range assessment model takes the preprocessed dual-modal image as input, and the dual-channel MobileNetV2 feature extraction network extracts n-scale visible light features and infrared features with photovoltaic component dust distribution information from the visible light image and the infrared image respectively; then, the extracted n-scale visible light features and infrared features are input into n multi-cascade cross-modal fusion networks in pairs, and n-scale dual-modal fusion features are output; then, the n-scale dual-modal fusion features are input into the multi-scale fusion network, and multi-scale dual-modal fusion features are output; finally, the multi-scale dual-modal fusion features are input into the classifier and mapped into m categories of power loss ranges.
5. A photovoltaic module power loss range assessment method according to claim 1, characterized in that: The classifier maps the multi-scale fusion features into eight categories of power loss ranges: 0-10%, 10-20%, 20-30%, 30-40%, 40-50%, 50-60%, 60-70%, and 70-100%.
6. A photovoltaic module power loss range assessment system, characterized in that: The method comprises a memory, a processor, and computer program instructions stored in the memory and executable by the processor. When the processor executes the computer program instructions, the method according to any one of claims 1 to 5 can be implemented.