Research On Hot Spot Recognition Method Of Small Sample Photovoltaic Module Based On Convolutional Denoising Autoencoder Network

The traditional fossil energy is facing a series of problems such as environmental pollution,raw material depletion and so on.Solar energy as a new energy occupies an increasing proportion in the energy structure with its advantages of green,environmental protection and renewable.With the rapid development of photovoltaic industry,its safety is also faced with more severe challenges.The solar cells with hot spot effect will reduce the power generation efficiency and service life of the photovoltaic system,increase the power generation cost and even cause power hazard.Therefore,the realization of hot spot image detection in photovoltaic modules is of great significance to ensure the security of photovoltaic power generation system.In this paper,taking the photovoltaic infrared image as the research object,based on the small number of photovoltaic hot spot image samples and difficult to collect,a small sample photovoltaic module hot spot recognition method based on Convolution Denoising Autoencoder network is proposed to realize the detection and location of hot spot battery chips.First of all,the original photovoltaic infrared image is preprocessed and a small sample data set is constructed.Image preprocessing mainly includes three parts: image segmentation,perspective transformation and equidistant segmentation.Based on the inherent characteristics and basic principle of infrared image,an image segmentation method based on region growth algorithm is adopted and its segmentation effect is obviously better than that based on edge and threshold.According to the temperature and color distribution law of different battery chips,they are divided into five states,and according to the serious imbalance of the distribution of battery chips in different states,five types of battery chips are selected,and the small sample data set of this experiment is constructed and its location information is indicated.Secondly,in order to realize the detection of hot spot battery chip,a small sample photovoltaic module hot spot recognition method based on Convolution Denoising Autoencoder network is proposed.The traditional Convolution Neural Network can drive a large amount of data to capture more abstract and effective features.However,the small sample photovoltaic hot spot image data set leads to the over-fitting phenomenon of the traditional Convolution Neural Network and reduces the recognition accuracy of the hot spot image.In order to solve this problem,this paper constructs a Convolutional Denoising Autoencoder network based on traditional Convolutional Neural Network and Denoising Autoencoder,which has strong feature extraction and anti-jamming ability.It can overcome the over-fitting phenomenon caused by insufficient sample data and improve the recognition accuracy of hot spot images.The experimental results show that for the small sample photovoltaic hot spot image data set,the average recognition accuracy of the Convolutional Denoising Autoencoder network is 9.43% higher than that of the traditional Convolutional Neural Network.For the batteries with different kinds of noise,the average accuracy of the test set of the Convolutional Denoising Autoencoder network model is also significantly higher than that of the Convolutional Neural Network,that is,it has stronger anti-jamming ability.The trained Convolutional Denoising Autoencoder network is loaded to locate the hot spot of the randomly selected photovoltaic module,and the positioning effect is good,which proves the effectiveness of the algorithm.