Data driving and image recognition coupled photovoltaic module fault judging and positioning method

Abstract

The invention relates to a data driving and image recognition coupled photovoltaic module fault judging and positioning method, which comprises the following steps of: firstly, carrying out fault recognition on a photovoltaic array through a data driving method, and collecting a photovoltaic module infrared image of a fault photovoltaic array according to a photovoltaic array fault recognition result; and carrying out fault identification and positioning on the infrared image of the photovoltaic module through an image identification method. According to the method, the fault of the photovoltaic module can be accurately identified and positioned, the fault identification accuracy of the photovoltaic module is effectively improved, the fault photovoltaic module is quickly positioned, and the fault processing and operation and maintenance difficulty of the photovoltaic module is effectively reduced. Compared with the prior art, the method has the advantages that the proportion of fault data in a training data set is increased, the correlation of samples adopted by model training is low, the fault positioning accuracy is high, and infrared image features are fully extracted.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic module fault diagnosis, and particularly to a method for judging and locating photovoltaic module faults by coupling data driving and image recognition. Background Art

[0002] The purpose of photovoltaic module fault diagnosis is to develop efficient and accurate methods and systems to identify and locate faults in photovoltaic modules. Since photovoltaic modules are long-term exposed to complex environments and are vulnerable to various factors that cause performance degradation or failure, fault diagnosis technology is crucial for ensuring the stable operation of photovoltaic systems.

[0003] In the field of photovoltaic module fault diagnosis, the existing technologies have the following challenges: (1) Fault data is easily submerged in non-fault data, which affects the training effect of the fault diagnosis model. In practical applications, photovoltaic modules are in normal operation for most of the time, while the occurrence frequency of faults is relatively low, resulting in relatively scarce fault data. When these limited fault data are surrounded by a large amount of non-fault data, the training process of the model is easily interfered, and it is difficult to accurately capture fault features, thus affecting the accuracy and reliability of fault diagnosis.

[0004] (2) Photovoltaic modules are usually deployed in fixed positions, which leads to high correlation of training data and affects the model training effect. Due to the relatively fixed installation positions and environmental conditions of photovoltaic modules, their operation data often shows a high degree of spatio-temporal correlation. This correlation may cause overfitting in the model training process, that is, the model has too high a fitting degree for the training data, while the generalization ability for new data decreases. This will not only reduce the accuracy of the fault diagnosis model, but also may make it perform poorly in practical applications.

[0005] Chinese Patent Application Publication No. CN113139955A discloses a method and system for identifying photovoltaic module faults based on dual-light images. Although 3000 infrared images are used to train the ResNet model, the problem of unbalanced sample categories is not considered. In the actual operation of a photovoltaic power station, the occurrence frequency of faults is low. If the proportion of fault samples in the training set is too small, it is easy to cause insufficient feature learning and increase the fault undetected rate.

[0006] Therefore, there is an urgent need for a photovoltaic module diagnosis method to improve the accuracy and reliability of photovoltaic module fault diagnosis. Summary of the Invention

[0007] The object of the present invention is to overcome the defects existing in the above-mentioned prior art and provide a method for judging and locating photovoltaic module faults by coupling data driving and image recognition, so as to solve or partially solve the problems that the proportion of fault data is low and the sample correlation is too high, resulting in unsatisfactory model training effects, and further affecting fault judgment and location.

[0008] The object of the present invention can be achieved by the following technical solutions: One aspect of the present invention provides a method for judging and locating photovoltaic module faults by coupling data driving and image recognition, including the following steps: Obtain the historical power generation and environmental data of the target photovoltaic array and mark the array fault categories. Based on the obtained fault category labels, balance the power generation and environmental data, screen to obtain a sample set with low correlation and a fault-to-non-fault ratio meeting the preset requirements, and train a binary classification first-stage model and a multi-classification second-stage model respectively; Obtain the infrared image data of the target photovoltaic array and mark the component fault categories, and train an image recognition diagnosis model based on attention for multi-classification; Obtain the real-time power generation and environmental data of the target photovoltaic array, and use the first-stage model to judge whether the target photovoltaic array is faulty, realizing fault judgment; In response to the fault of the target photovoltaic array, based on the real-time power generation and environmental data, use the second-stage model to obtain the predicted fault type of the photovoltaic array; In response to the fault of the target photovoltaic array, obtain the real-time infrared image data of the target photovoltaic array, and use the image recognition diagnosis model to obtain the predicted fault type of the photovoltaic module; Based on the coupling judgment of the photovoltaic array fault type and the photovoltaic module fault type, realize fault location.

[0009] As a preferred technical solution, the steps of balancing the power generation and environmental data based on the fault category labels and screening to obtain a sample set with low correlation and a fault-to-non-fault ratio meeting the preset requirements include the following steps: Calculate the Pearson correlation coefficient between samples, screen and retain samples with a correlation coefficient less than the preset value, and adjust to make the ratio of fault samples to non-fault samples 2:1; Perform data normalization processing on the screened samples; By randomly deleting samples from the screened samples or randomly selecting samples from the screened-out samples for supplementation, obtain samples matching the preset quantity as the sample set.

[0010] As a preferred technical solution, the Pearson correlation coefficient between samples and is calculated by the following formula: , wherein, is the covariance of the sample and , and and are the variances of the samples and , respectively.

[0011] As a preferred technical solution, the training processes of the first-stage model and the second-stage model include the following steps: Divide the sample set into a binary classification sample set and a multi-classification sample set; Based on the binary classification sample set, train an XGBoost model with the goal of maximizing the prediction accuracy of whether the photovoltaic module is faulty, to obtain the first-stage model; Based on the multi-classification sample set, train an XGBoost model with the goal of maximizing the prediction accuracy of the array fault category label, to obtain the second-stage model.

[0012] As a preferred technical solution, the training process of the image recognition and diagnosis model includes the following steps: Adjust the infrared image data to a preset size and perform a normalization operation to obtain an image data set; Based on the image data set, train a ResNet50-CBAM model based on the self-attention mechanism with the goal of maximizing the prediction accuracy of the photovoltaic module fault type label, to obtain the image recognition and diagnosis model.

[0013] As a preferred technical solution, the power generation and environmental data include phase current, phase, irradiance intensity, and photovoltaic panel temperature.

[0014] As a preferred technical solution, the array fault categories include normal operation, short circuit fault, aging fault, open circuit fault, and shadow occlusion.

[0015] As a preferred technical solution, the component fault categories include component hot spot, component cracking, diode fault, module offline fault, vegetation occlusion, and no fault.

[0016] Another aspect of the present invention provides an electronic device, including: one or more processors and a memory, where the memory stores one or more programs, and the one or more programs include instructions for executing the foregoing photovoltaic module fault determination and location method that couples data-driven and image recognition.

[0017] Another aspect of the present invention provides a computer-readable storage medium comprising one or more programs for execution by one or more processors of an electronic device, wherein the one or more programs include instructions for executing the aforementioned photovoltaic component fault judgment and positioning method coupled with data drive and image recognition.

[0018] Compared with the prior art, the present invention has at least one of the following beneficial effects: (1) Increasing the proportion of fault data in the training data set: The present invention obtains a sample set with low correlation and a fault-to-non-fault ratio that meets preset requirements by screening, and trains the first-stage model and the second-stage model respectively. By ensuring that the proportion of fault data in the training data set meets the preset ratio, the problem of unsatisfactory model training effect caused by the low proportion of fault data is effectively alleviated.

[0019] (2) The samples used in model training have low correlation: The present invention calculates the Pearson coefficient between samples, screens samples with coefficients less than a preset value and ensures the number of final samples, thereby effectively improving the diversity between training samples and alleviating the problem that a large amount of homogeneous data on photovoltaic modules affects the model training effect.

[0020] (3) High fault location accuracy: The present invention first determines whether there is a fault, and then integrates the power generation and environmental data and infrared images to determine the fault type. The model is divided into two stages to perform fault judgment and location respectively. After confirming the existence of a fault, the fault type is predicted. Since there is no interference from non-fault data, the prediction will be more accurate.

[0021] (4) Fully extracting infrared image features: The present invention adopts the ResNet50-CBAM model based on the self-attention mechanism to construct an image recognition diagnosis model, which can fully extract the features in the image to achieve accurate prediction of the fault type.

[0022] (5) The hierarchical structure is clear, providing a cascade fault identification and location process algorithm from PV array failure to PV module failure, accurately locating the fault of the PV module in the faulty PV array, effectively reducing the fault location time of PV power generation operation and maintenance, and improving the efficiency of PV power generation operation and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A schematic diagram of a process of a photovoltaic module fault judgment and positioning method coupled with data drive and image recognition in an embodiment; Figure 2 A flowchart of a two-stage data-driven modeling dataset screening process in an embodiment; Figure 3 A flowchart of photovoltaic array fault diagnosis based on data-driven two-stage modeling in an embodiment; Figure 4 Schematic diagram of an electronic device in an embodiment. Specific Embodiments

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0025] In the present invention, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality" means two or more unless otherwise specifically defined.

[0026] Embodiment 1 In view of the problems existing in the foregoing prior art, this embodiment provides a method for photovoltaic module fault judgment and positioning by coupling data drive and image recognition. Refer to Figure 1 , the method includes the following steps: Step S1, obtain the historical power generation and environmental data of the target photovoltaic array and perform array fault category marking. Based on the fault category labels, balance the power generation and environmental data, and screen out a sample set with low correlation and a fault-to-non-fault ratio meeting the preset requirements. Then, train a binary classification first-stage model and a multi-classification second-stage model respectively.

[0027] Specifically, refer to Figure 2 and Figure 3 , this step includes the following sub-steps: Step S101, collect the historical operation data of the photovoltaic power station and perform fault marking on the historical data. Specifically, the collected historical power generation and environmental data of the photovoltaic array include features such as phase 1 current, phase 2 current, phase 1 voltage, phase 2 voltage, irradiance intensity, and photovoltaic panel temperature (single collection); and perform fault category annotation on the data, that is, fault marking, to obtain the array fault category , including normal operation, short-circuit fault, aging fault, open-circuit fault, and shadow occlusion and other faults.

[0028] Step S102, feature selection processing. Based on the fault category labels, balance the power generation and environmental data, and screen out a sample set with low correlation and a fault-to-non-fault ratio meeting the preset requirements.

[0029] First, the data collected in step S101 Select according to the fault labels of the labeled data, and perform balancing processing according to the fault categories. The data balancing processing uses the Pearson correlation coefficient, and the formula is , where is the sample and is the covariance of, and are the variances of the samples and respectively.

[0030] Then, select the samples with smaller correlation coefficients. Specifically, in this embodiment, balancing processing is performed for each type of fault separately, deleting the samples with a Pearson correlation coefficient greater than 0.6, and retaining all the samples with a correlation coefficient less than or equal to 0.6. Adjust the proportion of the data quantity of each fault class to an appropriate proportion, that is, a proportion similar to the fault occurrence probability (specifically, the proportions of each fault in the original dataset can be used). The ratio of fault data to non-fault data is 2:1, and the dataset is obtained.

[0031] Step S103, data preprocessing. For the dataset selected in step S102, use the formula to perform normalization processing. When the number of existing samples is greater than or equal to the threshold (such as 210867), randomly delete ( m -210867) samples from the filtered samples. If the number of existing samples is less than this threshold, randomly select (210867 - m ) samples from the deleted samples for supplementation to obtain samples matching the preset quantity as the sample set, and divide the sample set into a binary classification dataset of fault and non-fault, and a multi-classification dataset of faults, where the dataset only contains fault samples. For each dataset, it is divided into a training set for parameter selection and model training, and a test set for model evaluation and output.

[0032] Preferably, when incremental samples are collected, the steps of S101 - S103 are used to update the dataset.

[0033] Step S104, train the first-stage model for binary classification and the second-stage model for multi-classification respectively. Use the XGBoost model to train the first-stage model on the binary classification dataset of fault and non-fault preprocessed in step S103 to obtain the first-stage model . Use the XGBoost model to train the second-stage model on the multi-classification dataset of faults preprocessed in step S103 to obtain the second-stage model 。The first-stage model and the second-stage model together constitute a data-driven photovoltaic array fault diagnosis model 。

[0034] Step S2: Obtain the infrared image data of the target photovoltaic array, perform component fault category labeling, and train an image recognition diagnosis model based on attention-based multi-classification.

[0035] Specifically, this step includes the following sub-steps: Step S201: Collect the infrared image dataset of photovoltaic components , collect the infrared image data of photovoltaic components, label the images, and obtain component fault category labels , including Cell (component hot spot), Cracking (component cracking), Diode (diode fault), Offine-Module (module offline fault), Shadoing (vegetation occlusion), and No-Anomaly (no fault).

[0036] Step S202: Image preprocessing. Crop the image dataset collected in step S201 , crop the original image with a size of 40*24 to an image with a size of 36*22, and then use the formula to perform normalization on the image to obtain the image dataset 。

[0037] Step S203: Image recognition model training. Add a convolutional block attention module (CBAM) after the last convolutional layer of each basic block in ResNet50 to form a ResNet50-CBAM model. Use the ResNet50-CBAM model to perform image recognition training on the preprocessed photovoltaic component fault dataset in step S202 to obtain an image recognition diagnosis model 。

[0038] Step S3: Obtain the real-time power generation and environmental data of the target photovoltaic array, use the first-stage model to determine whether the target photovoltaic array is faulty, and achieve fault judgment. When the prediction result is that the target photovoltaic array has a fault, based on the real-time power generation and environmental data, use the second-stage model to obtain the predicted fault type of the photovoltaic array.

[0039] Specifically, this step includes the following sub-steps: Step S301: Real-time collect the power generation and environmental data of the photovoltaic array , and use the formula to normalize the data to obtain the data 。

[0040] Step S302, photovoltaic array fault judgment. Use the first-stage model obtained in step S1 to identify the photovoltaic fault array for the data obtained in S301 and perform fault judgment.

[0041] Step S303, if the prediction result is that the photovoltaic array has a fault, use the second-stage model obtained in step S104 to perform fault identification on the photovoltaic array data determined to be faulty in step S302 and perform fault identification.

[0042] Step S4, when the prediction result is that the target photovoltaic array has a fault, obtain the real-time infrared image data of the target photovoltaic array, use the image recognition diagnosis model to obtain the predicted fault type of the photovoltaic module, and perform coupling judgment based on the photovoltaic array fault type and the photovoltaic module fault type to achieve fault location.

[0043] Specifically, this step includes the following sub-steps: Step S401, photovoltaic module image acquisition of the faulty photovoltaic array. Perform photovoltaic module image acquisition on the faulty array determined in step S3, and crop the infrared image to a picture with a size of 36*22, and use the formula for normalization processing.

[0044] Step S402, photovoltaic module fault identification and location. Use the image recognition model obtained in step S2 to perform photovoltaic module fault identification and location on the photovoltaic module image preprocessed in step S401, and perform coupling judgment with the array fault type obtained in step S303.

[0045] In summary, this method first performs fault identification on the photovoltaic array through a data-driven method, collects the infrared images of the photovoltaic modules of the faulty photovoltaic array according to the photovoltaic array fault identification result, and performs fault identification and location on the infrared images of the photovoltaic modules through an image recognition method. This method can achieve accurate fault identification and location of photovoltaic modules, effectively improve the accuracy of photovoltaic module fault identification, quickly locate the faulty photovoltaic modules, and effectively reduce the difficulty of fault handling and operation and maintenance of photovoltaic modules.

[0046] Embodiment 2 Based on Embodiment 1, as Figure 4 shown, this embodiment provides an electronic device, including: one or more processors and a memory, and the memory stores one or more programs, and the one or more programs include computer program instructions for executing the photovoltaic module fault judgment and location method of coupling data driving and image recognition as described in Embodiment 1.

[0047] In a typical configuration, an electronic device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0048] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory such as read-only memory (ROM) or flash memory (flash RAM). Memory is an example of computer-readable media.

[0049] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0050] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are performed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0051] Embodiment 3 This embodiment provides a computer-readable storage medium, including one or more programs for execution by one or more processors of an electronic device, and the one or more programs include instructions for executing the photovoltaic component fault judgment and location method of coupling data driving and image recognition as described in Embodiment 1.

[0052] A computer-readable storage medium includes permanent and non-permanent, removable and non-removable media and can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of the computer's storage medium include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device. As defined herein, a computer-readable medium does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0053] As described above, the above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A photovoltaic module fault judgment and positioning method coupled with data drive and image recognition, characterized in that: The steps include: The historical power generation and environmental data of the target photovoltaic array are obtained and array fault categories are marked. Based on the obtained fault category labels, the power generation and environmental data are balanced, and a sample set with low correlation and a fault-to-non-fault ratio that meets the preset requirements is screened. The first-stage model of the binary classification and the second-stage model of the multi-classification are trained respectively. Obtain infrared image data of the target photovoltaic array and label component fault categories, and train an attention-based multi-classification image recognition diagnosis model; Acquire the real-time power generation and environmental data of the target photovoltaic array, and use the first-stage model to determine whether the target photovoltaic array is faulty to achieve fault judgment; In response to a target photovoltaic array failure, based on the real-time power generation and environmental data, using the second stage model to obtain a predicted photovoltaic array failure type; In response to a target photovoltaic array failure, real-time infrared image data of the target photovoltaic array is acquired, and the predicted photovoltaic component failure type is obtained using the image recognition diagnosis model; A coupled judgment is performed based on the photovoltaic array fault type and the photovoltaic component fault type to achieve fault location.

2. A photovoltaic module fault judgment and positioning method coupled with data drive and image recognition according to claim 1, characterized in that: The method of balancing the power generation and environmental data based on the fault category labels and screening out samples with low correlation and a fault-to-non-fault ratio that meets preset requirements comprises the following steps: Calculate the Pearson correlation coefficient between samples, select and retain samples with correlation coefficients less than the preset value, and adjust the ratio of faulty to non-faulty samples to 2:1; Perform data normalization on the screened samples; By randomly deleting samples from the screened samples, or randomly selecting samples from the screened samples for supplementation, samples matching the preset number are obtained as the sample set.

3. A photovoltaic module fault judgment and positioning method coupled with data drive and image recognition according to claim 2, characterized in that: sample and The Pearson correlation coefficient is calculated using the following formula: ， in, For sample and The covariance of and The samples are and The variance of .

4. The photovoltaic module fault judgment and positioning method coupled with data drive and image recognition according to claim 1 is characterized in that: The training process of the first stage model and the second stage model includes the following steps: Dividing the sample set into a binary classification sample set and a multi-classification sample set; Based on the binary classification sample set, the XGBoost model is trained to maximize the prediction accuracy of whether the photovoltaic module is faulty, thereby obtaining a first-stage model; Based on the multi-classification sample set, the XGBoost model is trained with the goal of maximizing the prediction accuracy of array fault category labels to obtain a second-stage model.

5. The photovoltaic module fault judgment and positioning method coupled with data drive and image recognition according to claim 1 is characterized in that: The training process of the image recognition diagnosis model includes the following steps: The infrared image data is adjusted to a preset size and normalized to obtain an image data set; Based on the image dataset, with the goal of maximizing the prediction accuracy of photovoltaic module fault type labels, a ResNet50-CBAM model based on the self-attention mechanism was trained to obtain an image recognition and diagnosis model.

6. The photovoltaic module fault judgment and positioning method coupled with data drive and image recognition according to claim 1 is characterized in that: The power generation and environmental data include phase current, phase, irradiation intensity and photovoltaic panel temperature.

7. The photovoltaic module fault judgment and positioning method coupled with data drive and image recognition according to claim 1 is characterized in that: The array fault categories include normal operation, short circuit fault, aging fault, open circuit fault and shadowing.

8. The photovoltaic module fault judgment and positioning method coupled with data drive and image recognition according to claim 1 is characterized in that: The component failure categories include component hot spots, component cracking, diode failure, module offline failure, vegetation shading, and no failure.

9. An electronic device, characterized in that: include: One or more processors and a memory, wherein the memory stores one or more programs, wherein the one or more programs include instructions for executing the photovoltaic component fault judgment and positioning method coupled with data drive and image recognition as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: It includes one or more programs for execution by one or more processors of an electronic device, and the one or more programs include instructions for executing the photovoltaic component fault judgment and positioning method coupled with data drive and image recognition as described in any one of claims 1-8.

Images