Hybrid Predictive Modeling for Charging Demand Prediction of Electric Vehicles

Sustainability

Jung-In Seo¹  Suk-Bok Kang²  Young-Eun Jeon² 

[1] Department of Information Statistics, Andong National University, Andong 36729, Korea;Department of Statistics, Yeungnam University, Gyeongsan 38541, Korea;

关键词: charging demands;    electric vehicles;    Fourier transform;    machine learning;    time series analysis;   

DOI  :  10.3390/su14095426

来源: DOAJ

【 摘 要 】

In recent years, the supply of electric vehicles, which are eco-friendly cars that use electric energy rather than fossil fuels, which cause air pollution, is increasing. Accordingly, it is emerging as an urgent task to predict the charging demand for the smooth supply of electric energy required to charge electric vehicle batteries. In this paper, to predict the charging demand, time series analysis is performed based on two types of frames: One is using traditional time series techniques such as dynamic harmonic regression, seasonal and trend decomposition using Loess, and Bayesian structural time series. The other is the most widely used machine learning techniques, including random forest and extreme gradient boosting. However, the tree-based machine learning approaches have the disadvantage of not being able to capture the trend, so a hybrid strategy is proposed to overcome this problem. In addition, the seasonal variation is reflected as the feature by using the Fourier transform which is useful in the case of describing the seasonality patterns of time series data with multiple seasonality. The considered time series models are compared and evaluated through various accuracy measures. The experimental results show that the machine learning approach based on the hybrid strategy generally achieves significant improvements in predicting the charging demand. Moreover, when compared with the original machine learning method, the prediction based on the proposed hybrid strategy is more accurate than that based on the original machine learning method. Based on these results, it can find out that the proposed hybrid strategy is useful for smoothly planning future power supply and demand and efficiently managing electricity grids.

【 授权许可】

Unknown