Deep-Learning based Multiple Class Events Detection and Classification using Micro-PMU Data

Real-time analysis of multiple class events is necessary for enhanced situational awareness and safe operation of the power distribution networks. The deep-learning-based data-driven techniques for multiple-class event analysis have not been explored in the literature. Thus, we proposed a deep learning-based data-driven method to detect events from large real-time data. The convolutional auto-encoder helps in the exact time of event detection and temporal localization using time-series voltage and current magnitude phasors. Then, we proposed a deep-learningbased data-driven event classifier for effective multiple-class event classification. Three-phase voltage and current magnitude features are selected as input into the multi-class LSTM (long short-term memory) classifier model. Based on the research, a unique time series voltage and current magnitude pattern is generated from the different classes of events. With the help of the proposed algorithm, five classes of events can be effectively detected and classified using the measured voltage and current magnitude phasors. In this perspective, the proposed multi-class LSTM classifier is trained and tested with IEEE-34 node test feeder, which are simulated in a testbed setup including real-time digital simulator (RTDS), hardware mu PMUs, hardware distribution phasor data concentrator (DPDC), and network switches. The outcomes verify that the proposed algorithm has high accuracy (100%) and requires only voltage and current magnitude phasors.