FEDformer

Frequency Enhanced Decomposed Transformer for long-term time series forecasting. Combines Transformer with seasonal-trend decomposition and exploits sparse Fourier representations of time series via random frequency component selection. Achieves linear complexity and reduces prediction error by 14.8% (multivariate) and 22.6% (univariate) compared to prior state-of-the-art. Published at ICML 2022.