Fault Classification in Transmission Network with Semi-supervised Learning Method

Abstract

Accurate fault classification in transmission network has been a critical problem with the development of smart grid. Recently, deep learning based methods have shown efficiency to handle this problem. However, the dependency on labeled data is challenging as it is time-consuming and laborious to label all the training data. Moreover, in the real situation, a large number of fault events are recorded without labels. Therefore, this paper represents an approach of diagnosing fault in transmission network with extreme limited labels(under 1% label) using semi-supervised learning method. First, the fault data is simulated through PSS/E to produce four kinds of short circuit faults with random noise, distance and O-U load fluctuation. Then a semi-supervised model based on pseudo label is proposed to detect the fault type. The experimental results show that the proposed method is able to handle large amount of data with extreme limited labels and represents superior performance on fault classification task.