Research On Detection Method Of Hot Spot Effect Of Photovoltaic Module Based On Transfer Learning

With the deepening of the world energy crisis,the disadvantages of fossil fuels are becoming more and more obvious,and countries around the world pay more and more attention to renewable energy.As the representative of clean and pollution-free energy,solar energy accounts for an increasing proportion in the field of new energy applications.With the wide application of photovoltaic power generation technology,the hot spot fault of photovoltaic modules often encountered in the process of power generation has become a pain point problem.The timely detection and treatment of photovoltaic hot spot faults plays a key role in maintaining the efficiency and safety of photovoltaic power generation system.Therefore,it is of great significance to design an efficient and accurate hot spot detection method.In this paper,taking the infrared image of photovoltaic module taken in the field as the research object,a hot spot detection method of photovoltaic module based on transfer learning is proposed,which mainly includes the following four points:The main contents are as follows:(1)Construct the infrared image data set of solar cells.Aiming at the problems existing in the infrared images of photovoltaic modules taken in the field,the preprocessing is carried out by using Canny edge detection,target region contour extraction,perspective transformation,segmentation according to specifications,and the traditional data enhancement method and the data enhancement method based on generating countermeasure network are used to expand the sample data,and the solar cell infrared image data set which can be used for model training is constructed.(2)A negative sample classification method based on H component difference in HSV color space is proposed.In order to solve the problem of sample imbalance of hot spot data set,the negative samples are reclassified according to the H-component characteristics of battery infrared images under normal operating conditions.The quantity difference between samples is balanced by adding sample categories.(3)A photovoltaic module hot spot detection method based on transfer learning is proposed.In order to solve the problem that the training of traditional neural network model needs a lot of data and the number of hot spot samples is relatively small and difficult to collect,according to the idea of transfer learning,the deep transfer learning method based on network is adopted.The Inception-v3 model is used as the pretraining model,part of its structure and parameters are transferred,and then combined with the full connection layer module which can adapt to the hot spot detection task to form the hot spot detection model.Finally,the model training is completed according to the appropriate hyperparameters and training strategies on the infrared image data set of solar cells.(4)Design and develop a hot spot detection platform based on BS architecture.From the point of view of engineering application value,in order to improve the practicability of the hot spot detection model,the hot spot detection model is encapsulated with web technology to realize the intellectualization of the detection process and the visualization of the test results.Finally,the simulation results show that the test set accuracy of the hot spot detection model is 95.83%,which verifies the accuracy and efficiency of the hot spot detection method proposed in this paper.Through the actual use of the detection platform,it shows the engineering application value of the hot spot detection platform developed in this paper.