Research On Hot Spot Detection Of Photovoltaic Panels Based On Generative Adversarial Network

With the rapid construction of the "carbon peak,carbon neutral" green economic system,solar power generation has developed rapidly as the main source of new energy.However,the hot spot failure of photovoltaic modules due to their own aging and outdoor environment during the power generation process has seriously affected the safe,stable and efficient operation of photovoltaic power plants.Therefore,the accurate detection of the type and location of hot spot faults in photovoltaic panels is of great significance for improving the power generation efficiency and safe operation of battery modules.In this paper,based on the deep learning model and image processing technology,a model for hot spot fault detection and hot spot area location based on infrared image data of photovoltaic panels is constructed.The realization of the model is completed from two aspects.First,a hot spot fault detection and location model are constructed based on the idea of image reconstruction to realize defect location based on the characteristics of clear image and strong classification ability generated by generative adversarial networks.The model is composed of generator,discriminator and encoder.Firstly,the generator and discriminator are trained by photovoltaic panel infrared image data set.After the training,the network parameters are fixed.Then the encoder and the trained generator are constructed into a coding-decoding network model,and the encoder is trained by the healthy samples in the data set.Finally,the test samples are input into discriminator and codec model to classify hot spot faults and locate hot spot locations.Simulation results show that for three types of hot spot faults in the data set,the hot spot fault identification accuracy and hot spot region location accuracy of the model are better than some traditional hot spot classifier models and hot spot image segmentation methods.Second,fault detection model optimization based on memory network and self-attention mechanism.In this paper,memory network is used to improve the matching degree between the health image reconstructed by generator and the real data,and the location performance of hot spots is improved.In order to further explore the internal features of image space and improve the overall performance of the hot spot detection model,an SA-FDGAN hot spot fault detection optimization model was proposed based on self-attention mechanism,and a double time scale updating and spectral normalization method were added to the model training.The simulation results show that compared with the original model and the traditional classifier model,the SAFDGAN model can distinguish the hot spot fault types,and the hot spot location accuracy is higher.In this paper,the reliability of hot spot fault detection model is proved by simulation experiments,and the hot spot detection of photovoltaic power station has good practicability.According to the simulation experiments designed by a variety of open source data sets,it is proved that the SA-FDGAN fault detection model has universal applicability and has reference value for fault detection and location in other fields.