Time series analysis and forecasting of electric vehicle charging energy data(focusing on the public fast charging infrastructure)

Abstract

Energy demand forecasting is very hot topic in contemporary energy industry and academic world. As the paradigm shift from internal combustion engine car to electric car(electric vehicle, EV), EV charging is one of the most important issue for EV eco-system. The necessity of EV charging demand forecasting is raising because we need to respond explosive increase of EV charging load proactively which is unprecedent and brand-new concept of power consuming, and also it helps us to select and allocate adequate charging station location in the view of supply side.

In this thesis we will handle Korean local EV charging energy demand data, especially focusing on public fast charging infrastructure(Two main public charging point operators : Ministry of Environment, Korea Electric Power Corporation). We divided the whole 3.5 years monthly data into two parts(3 years of fitting period and a half year of evaluation period) for suitable assessment of forecasting. Furthermore, we will check regional level of charging amount data and try to figure out the differences between national level data.

Not only just conducting conventional regression model, we expand the scope to forecasting models(Simple Forecasting Model, Exponential Smoothing, ARIMA and ARIMAX, GARCH, VAR). In addition to exploring the models, we check the validity of each model and compare the forecast performance by using evaluation criteria(MAE, RMSE, MAPE, MASE).

Regardless of certain limitations of this study, we can get some meaningful implications and identifications such as (1)national level data and forecasting analysis results do not respond to regional level (2)regardless of high correlation and significance from regression, EV registration volume does not always guarantee charging amount in specific regional level (3)COVID19 dummy variable and Fine dust concentration which are not covered as independent variable from preceding researches show significant result with charging amount from correlation and multiple regression analysis.