Travel Motif-Based Learning Scheme for Electric Vehicle Charging Demand Forecasting

To better plan, build, and operate charging infrastructure, understanding the spatial and temporal distribution of electric vehicle (EV) charging demand is of utmost importance. The motifs of daily travel are representations of human activities and journeys undertaken on a daily basis, and these motifs inevitably affect EV charging demand. In this paper, people's daily travel motifs based synthetic EV charging demand dataset is created and future demand is forecasted using a machine learning algorithm. For a given city, several features have been extracted and processed, such as the number of EVs, energy consumption per KM, and the locations of all the places where people usually travel. These locations have been utilized to calculate the daily travelling distance and time in real-time manner according to the highly followed daily travel motifs. Then, the daily travelling distance is used in Monte Carlo Simulation (MCS) to estimate the EV charging demand. Based on the created EV charging demand dataset, a hyperparameter fine-tuned long short term memory (LSTM-HT) model is designed to forecast the future charging demand. The performance of LSTM-HT model in terms of
$R^{{2}}$
and MSE is 0.97 and 0.0025. This result indicates that the synthesis data generation method integrated with LSTM-HT model could be an effective alternative to forecast EV charging demand.