Deep learning neural network for power system fault diagnosis

This paper deals with application of deep learning neural network for power system fault diagnosis. Deep learning is a more effective approach than traditional neural network to solve problems including availability of data, better local optimum, and diffusion of gradients. In the paper, data is extracted from power system dispatching department and preprocessed before training in the deep learning network. Then, processed data is put into auto-encoders and the hidden features are observed in different dimensions so that we can preliminarily judge about the fault. Afterwards, trained stacked auto-encoders (SAE) is used to initialize and train a deep learning neural network (DLNN). The hidden features are observed in different dimensions so that the fault is preliminarily judged. The classifier is the last part of the network to reflect the types and possibility of diagnosis. The method of data availability, preprocess, and modeling is proposed in the paper. The result of simulation proves the feasibility of the approach and the influence factors are shown in the paper.