Day-ahead prediction of electric vehicle charging demand based on quadratic decomposition and dual attention mechanisms

Electric vehicles (EVs) are poised to become a vital flexibility resource in the future power system. However, due to their inherent stochastic nature, accurately predicting EV charging demand in advance, which is essential for power dispatchers seeking to optimize decisions. This study introduces a novel day-ahead prediction model for EV charging demand leveraging quadratic decomposition and Dual Attention (DA) mechanism. Based on the principle that daily charging demand curve results from the superposition of EV charging sessions, this study using Affinity Propagation (AP) clustering algorithm aiming to extract typical EV charging sessions that represent the charging patterns of average EV users, thereby enabling primary decomposition of the charging demand curve. Subsequently, a quadratic decomposition based on Variational Modal Decomposition (VMD) separates the low-frequency trend component from the high-frequency perturbation component in the charging load curves. The DA mechanism is then integrated into a predictive framework to forecast the demand for EV charging for the next day. The performance of the proposed method is assessed through the analysis of nearly 6000 charging periods from charging sessions at a highway charging station and a public charging station in South China. The results indicate that both the primary decomposition, which is based on clustering EV charging sessions, and the secondary decomposition, which utilizes VMD have significantly enhanced prediction accuracy. Furthermore, the proposed model surpasses the control algorithm across various metrics and exhibits robust generalization performance when compared across different types of charging station data.