Deep Learning Method with Application to Neutral-to-Ground Voltage (NTGV) Classification

The excessive neutral-to-ground voltage (NTGV) occurring on the secondary distribution system (SDS) may result in unnecessary losses and a safety hazard issue. This disrupting voltage is triggered by a number of reasons that are difficult to identify. This paper proposes an approach based on deep learning (DL) using real-world data from the NTGV signals for the classification of three categories: normal (N), lightning (LS), and ground fault (GF). The method consists of a special type of Recurrent Neural Networks (RNN s) namely Long Short-Term Memory (LSTM) that excels at capturing long-term dependencies. A total of 1531 two-cycle NTGV time series data, measured from several locations in an SDS, has been used in our studies in order to classify their categories. Results have shown that the proposed method is able to classify the type of NTGV from learned features in LSTM, with 99.56% classification accuracy on the test data set.