Machine learning‐based multivariate forecasting of electric vehicle charging station demand

Abstract The exponential rise of electric vehicles (EVs) is transforming the global automobile industry, driving a shift towards greater cleanliness and environmental sustainability. EV charging stations (EVCSs) play a pivotal role in this massive transition towards EVs, where accurate forecasting of EVCS demand is crucial for seamlessly integrating EVs into existing power grids. Most of the existing research mainly concentrates on univariate forecasting, neglecting the multiple factors influencing EVCS demand. Hence, this study offers a comparative analysis of different algorithms for univariate forecasting and multivariate forecasting, where the multivariate scheme incorporates metadata such as charging time, greenhouse gas savings, and gasoline savings. The experimental results indicate the superiority of the multivariate scheme over the univariate forecasting. For multivariate forecasting, the gated recurrent unit (GRU) has outperformed other models such as categorical boosting (Catboost), recurrent neural network (RNN), long short‐term memory (LSTM), extreme gradient boosting (XGBoost), random forest, convolutional neural network (CNN), CNN + LSTM, and LSTM + LSTM. The results of this study emphasize the significance of using the GRU model for multivariate forecasting with metadata during normal and noisy scenarios to yield more reliable and accurate predictions. This approach enhances decision‐making, policy development, and efficient grid integration in the growing EV sector.