Automatic Classification of Power Quality Disturbances Via Weighted Convolutional Neural Network

The identification and classification of power quality disturbances (PQDs) is critical for controlling the pollution in power grid. In this paper, we introduce an end to end automatic PQDs classification framework based on weighted convolutional neural network (WCNN). We first propose a weight residual network structure considering the difference in separability and effectiveness of the features located in different layers. The correlation distance scheme is utilized to determine the weight parameters. Then, the features can be automatically extracted from the raw PQD signals. Next, the feature visualization analysis is used to analyze the validity of the extracted PQD features. Last, a large number of experiments are performed to validate the convergence and effectiveness of the proposed method using both single and mixed PQDs in noisy and noise-free cases. Classification accuracy is presented, and the factors that affects the model performance are discussed.