A hybrid method for power demand prediction of electric vehicles based on SARIMA and deep learning with integration of periodic features. (1st July 2022)

• Record Type: Journal Article
• Authors: Ren, Fei; Tian, Chenlu; Zhang, Guiqing; Li, Chengdong; Zhai, Yuan
• Is Part Of: Energy. Volume 250(2022)
• Publication Date: 2022-07-01
• Subjects: Electric vehicle -- Power demand prediction -- Periodic feature -- Deep learning -- SARIMA; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2022.123738
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: Accurate power demand prediction of electrical vehicles (EVs) is crucial to power grid operation. To fully utilize the existing knowledge of EVs' power demand and further improve the prediction accuracy, this paper proposes a hybrid method for power demand prediction of EVs based on Auto-Regressive Integrated Moving Average (SARIMA) and deep learning with the integration of periodic features. First, the general linear trend of power demand is extracted by SARIMA; then, the residual non-linear components are obtained by eliminating the linear trend from the original power demand. Meanwhile, the periodic features of the non-linear component are determined according to the periodic parameters of the SARIMA. Afterward, the non-linear components are approximated by Long-Short Term Memory (LSTM), and the periodic features of the non-linear components are taken as a part of the inputs of the LSTM. Finally, the extracted linear trend and the predicted non-linear components are combined to generate the final prediction results. To verify the effectiveness of the proposed method, three experiments are conducted on a real EV charging station. The experimental results indicate that the proposed method significantly improves the prediction accuracy compared with other popular data-driven models. Highlights: A novel hybrid method is proposed to predict the power demand of electric vehicles. The original power demand is considered to consist of a general linear trend and non-linear components. The general linear trend is extracted by SARIMA and the residual non-linear components are predicted by LSTM. The periodic features are extracted and utilized in the prediction process. The proposed hybrid method significantly improves the prediction accuracy.