Load prediction of multi‐type electric vehicle charging stations based on secondary decomposition and feature selection. (27th July 2022)

• Record Type: Journal Article
• Authors: Guo, Xingchen; Jia, Rong; Zhang, Gang; Zhang, Yu; Xie, Tuo
• Is Part Of: International journal of energy research. Volume 46:Number 13(2022)
• Page Start: 18342
• Page End: 18362
• Publication Date: 2022-07-27
• Subjects: charging station load prediction 1 -- ensemble learning 4 -- maximum information coefficient 3 -- secondary decomposition 2; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Power resources -- Research -- Periodicals; 621.042
• DOI: 10.1002/er.8449
• Languages: English
• ISSNs: 0363-907X

Abstract

Summary: This study proposes a load prediction method for multi‐type electric vehicle charging stations based on secondary decomposition and feature selection. First, the original load sequence of an electric vehicle charging station is decomposed into relatively simple components using variational mode decomposition (VMD). The residual term after the decomposition carries rich information; therefore, this study introduces the idea of quadratic decomposition, where the residual term is decomposed by empirical mode decomposition (EMD). Second, an in‐depth investigation of the characteristics of ordinary, energy storage, and integrated charging stations for optical storage and charging establishes the corresponding influencing factor sets, and uses the maximum information coefficient to obtain the optimal feature set. Third, for the subsequent different frequencies, this study uses the long short‐term memory (LSTM) and back‐propagation neural network (BPNN) as basic learners to conduct the prediction. Finally, this study adopts the stacking ensemble learning strategy, takes the prediction result of the base learner and the optimal feature set as the input, and then obtains the final prediction result. Taking the real load data of three different types of charging stations in a city in Northwest China as an example, the model proposed in this study is evaluated and compared with other models. Meanwhile, the model's performance under the 2 to 6 h forecast periods was compared and analyzed. The results show that the model has good stability and representation, and can be used for load prediction of electric vehicle charging stations.