Global attention-based LSTM for noisy power quality disturbance classification

An increased dependency of digital control systems in the modern electrical network demand for a better quality of power signal. The occurrence of power quality disturbances (PQDs) in the network reduces the lifespan of power semiconductors and solid states switching devices. Global attention-based long short-term memory (LSTM) network is proposed to perform automatic time-series PQD detection and classification. Attention-based LSTM helps improve the noise immunity to extract salient features from noisy signal for PQD classification. The aim of this article is to analyse the performance of proposed attention-based LSTM under different noise conditions. Addictive white Gaussian noise is added to synthetic PQDs in different signal-to-noise ratio. These random generated noises are used to train and test the performance of proposed method, as well compared towards generic LSTM model. This work also shows the sensitivity of proposed method towards unknown noises that is not seen by the model during training phase.