Deep Learning Tackles Temporal Predictions on Charging Loads of Electric Vehicles

With the prediction of 145 million electric vehicles on the road by 2030, accommodation of charging needs for these electric vehicles will impose extra challenges to power grid strength. It is imperative to predict charging loads for future infrastructure improvement, including new charging stations' installation to meet the electric vehicles' charging needs and reduce the power grid overload. In this study, deep learning approaches including Artificial Neural Networks, Recursive Neural Networks, and Long-Short Term Memory models are used to predict the charging load with daily and weekly patterns using public datasets. The performances of the deep learning models were compared against the auto-regressive moving average model concerning convergence speed, MSE, RMSE, MAE, and R-squared. The long-short term memory model outperformed all other models concerning the evaluation metrics.