Clustering and Previous Visit Dependency Technique for Electric Vehicle Station Visits

Abstract

Although electric vehicles (EV) are expected to increase in number, the EV ecosystem supporting this growth is still in the early stages. To manage the risks involved, ecosystem infrastructure investments such as battery charging stations need practical EV station visit predictions. In this research, a forecasting technique is proposed that employs an adapted K-means clustering approach and depends on previous visits. Using aggregated traffic, the practical cluster number is chosen based on a variance explained threshold. Representative probabilities from the clusters are then linked to individual travel behaviors. In contrast to conventional EV station forecasts, the proposed technique is dependent on previous visits creating a realistic case where the visit of EV owners will likely depend on their distance travelled and their previous station visit. The EV station visit forecasting technique has been recently performed in charging stations meant for city and inter-state use in Australia leveraging its potential for practical use in supporting the EV ecosystem.