Health evaluation of power transformer using deep learning neural network

The health evaluation is an effective method to detect the health condition of power tran.3sformer. However, the redundancy, complexity, and small sample of dataset influence the performance of health evaluation method. To solve this issue, this paper presents a deep learning neural network (DLNN) to detect the HE of the power transformer. First, the echo state network (ESN) generates the data associated with the original data for solving the small sample problem. And then, the significant features are extracted by DLNN which contains improved Deep Residual Shrinkage Networks (IDRSN) and the one-dimension convolution neural network (1DCNN). Finally, the health status of the power transformer is obtained by Concat layer and Softmax layer. The DLNN is verified by datasets of dissolved gas collected on a real power transformer. The experiment results demonstrate that the proposed method obtains a better performance than the latest neural networks and health assessment method of power transformer.