Power quality disturbance signal segmentation and classification based on modified BI-LSTM with double attention mechanism

Abstract

This paper proposes a recurrent neural network (RNN) based model to segment and classify multiple combined multiple power quality disturbances (PQDs) from the PQD voltage signal. A modified bi-directional long short-term memory (BI-LSTM) model with two different types of attention mechanism is developed. Firstly, an attention gate is added to the basic LSTM cell to reduce the training time and focus the memory on important PQD signal part. Secondly, attention layer is added to the BI-LSTM to obtain the more important part of the voltage signal by assigning weightage to the output of the BI-LSTM model. Finally a SoftMax classifier is applied to classify the combined PQD signal in 96 different combinations. The performance of proposed BI-LSTM model with attention gate and attention layer mechanism is compared with the performance of baseline models based on recurrent neural network (RNN) and convolution neural network (CNN). With this model, the PQD signal is easily segmented from the voltage signal which makes the process of PQD classification more accurate with less computation complexity and in less time.