Short-Term Forecasting of EV Charging Load Using Prophet-BiLSTM

Abstract

The fast-growing charging load of electric vehicles (EVs) has created significant impact on the secure and economic operation of electric power grid. To effectively quantify future operational risks and optimize control actions of the grid, this paper presents a novel method of short-term forecasting of EV charging load using artificial intelligence algorithms. First, a Prophet model is trained to select key features affecting EV forecasting performance; then, a Bidirectional Long Short-Term Memory (BiLSTM) model is trained to provide high-accuracy forecasting model of EV charging load. The proposed method is tested on actual charging load data obtained from a large EV station in Southern China, and compared with state-of-the-art machine learning algorithms including the traditional Prophet, LSTM, ANN, CNN-LSTM, transformer and N-BEATS. The proposed method of Prophet-BiLSTM model demonstrates higher prediction accuracy.