APPLICATION OF DEEP RESIDUAL NETWORK IN FAULT DIAGNOSIS OF PHOTOVOLTAIC MODULES

Fault diagnosis techniques can accurately determine the health condition of photovoltaic modules, as well as the specific fault type and level. Conventional machine learning diagnosis algorithms rely on expert knowledge and manual feature extraction to build feature engineering, which usually leads to limited performance. In addition, the existing fault diagnosis methods of photovoltaic modules have a limitation. They only take the influence of the area size of fault array into account when determining the fault level, and ignore the effect caused by the degree of fault factor itself. In view of the above problems, this paper proposes a fault diagnosis method of photovoltaic modules based on deep residual network, which can automatically extract useful fault features. Raw current-voltage characteristic curves and ambient conditions are used as the input of the network. To verify the effectiveness of the proposed methodology, two datasets are obtained based on a laboratory experiment platform and the corresponding simulation model respectively. Furthermore, five popular machine learning methods, including support vector machine, decision tree, random forest and two other convolution neural networks, are used for comparison. According to the results, the proposed method achieves better comprehensive performance with the highest diagnostic accuracy and convergence speed.