Temporal data stream analysis for EV charging infrastructure in Jeju

Abstract

This paper first presents the main features of the dataset obtained from the charging station monitoring system currently in operation on Jeju island. Then, the time series analysis is conducted to find the behavioral trends according to districts and place types, based on periodic report archives from 49 fast chargers. Two most populated districts lead the occupancy rate of the whole island, while those chargers installed in administrative offices account for the most charging demand during office working hours. Combined with the dynamic time warping method, the hierarchical clustering process captures 3 main groups having the same day-by-day occupancy rate dynamics. Additionally, artificial neural network models are built to forecast the next day charging demand, and the prediction for working hours is acceptable as it is not affected by missing data. With the prediction model built upon open data and software, it is possible to develop a new smart grid application such as vehicle-to-grid and renewable energy generation.