EV-integrated power system transient stability prediction based on imaging time series and deep neural network

Abstract

The market penetration of electric vehicles (EVs) has increased drastically. However, the high integration of EV fast-charging stations (EVFCS) into the power systems makes them more vulnerable to severe grid disturbances. In case of a disturbance driving the power system to instability, a fast prediction of stability status is vital for allowing sufficient time to take intelligent emergency control actions. Although various types of machine learning (ML) and deep learning (DL) algorithms have been developed for early detection of instability, the lack of reliable ML/DL models, trained with a realistic dataset, limits their practical application. This paper presents a reliable, accurate DL-based model for early detection of instability in power systems, and compares the results with/without coupling of the EVFCSs. For training our ML/DL models, a large set of realistic phasor measurement unit (PMU) data is generated through a new approach involving a hybrid-type simulation, as an alternative to conventional approaches in data generation. In our experiments, time-synchronized measurements of voltage signals obtained from PMUs are taken as raw input data. Through our proposed method, raw PMU data are encoded into images for developing a reliable and robust convolutional neural network (CNN) model, predicting the stability status of power systems.