Particle-LSTM: Advancing wind energy forecasting through a synergistic fusion of LSTM-VAR model and particle filter

Abstract

Wind energy forecasting is essential for the optimal functioning of power grids and energy markets. In this study, we introduce the Particle-LSTM, which revolutionizes wind power forecasting by merging a Long Short-Term Memory-Vector Autoregression (LSTM-VAR) model with a Particle Filter. The LSTM-VAR model adeptly executes an affine transformation of time-varying VAR parameters utilizing an LSTM network, while the Particle Filter dramatically enhances forecast accuracy through its integration with the LSTM-VAR model. A comparative analysis against established models clearly demonstrates that Particle-LSTM surpasses existing methods across a range of error metrics, underscoring its dominance in wind power forecasting.