Interval Forecasting of Electric Vehicle Charging Load Based on FMD-LSTM-Bootstrap

With the rapid growth of electric vehicle (EV) ownership, the substantial charging demand has imposed operational impacts on power grids and created challenges for grid dispatch management. Consequently, forecasting EV charging loads has become a crucial research focus in the energy sector. This paper proposes an integrated FMD-LSTM-Bootstrap-based interval forecasting methodology for EV charging loads. The framework operates through three sequential phases: Initially, the Filtered Multivariate Decomposition algorithm decomposes historical EV charging data into multiple sub-modal components. Subsequently, Long Short-Term Memory networks perform deterministic predictions for each sub-modal component. Finally, the Bootstrap resampling technique is employed to construct prediction intervals for each decomposed component, which are then aggregated through signal reconstruction to generate comprehensive charging load interval forecasts. This hybrid approach effectively addresses the inherent limitations of deterministic forecasting in addressing prediction accuracy issues, while systematically integrating temporal pattern learning, uncertainty quantification, and multi-scale component analysis for enhanced grid operation decision-making.