Comparative Analysis of Deep Learning Models for Electric Vehicle Charging Load Forecasting

Grid-connected plug-in electric vehicle charging stations having integrated renewable energy sources like photovoltaic (PV) systems with battery energy storage help manage the variability of electric vehicle charging load and reduce the stress on the grid. Such charging stations will benefit from short to medium term forecasting of load demand, as it allows them to optimize the operation through better utilization of the PV and battery stored energy rather than relying on the grid. The load forecasting accuracy has a direct correlation with the degree of optimization achievable. Deep learning network is a form of artificial neural network which can be effectively used for time series forecasting. In this paper, some of the widely researched deep learning models such as long short-term memory (LSTM), gated recurrent units (GRU), hybrid of convolution neural network (CNN) and LSTM, hybrid of CNN and GRU, multivariate LSTM and multivariate GRU are analyzed for fitment for the charging load forecasting problem. The datasets available from multiple charging stations in a region are used for training the models. The predictions made using these models and their performances are analyzed using standard metrics and are presented.