What is FEDformer and how does it integrate seasonal–trend decomposition and Fourier representations into the Transformer architecture?

FEDformer, introduced in Chapter 2, is the Frequency Enhanced Decomposed Transformer, a model designed for long-term series forecasting. It integrates seasonal-trend decomposition and Fourier representations directly into the Transformer architecture. The model works by selecting subsets of frequency components and performing computation in the spectral domain.

As described in the preface and chapter list, FEDformer (Fedformer: Frequency Enhanced Decomposed Transformer for Long-term Series Forecasting, by Tian Zhou et al.) aims to improve long-term forecasting by combining two ideas inside the Transformer framework: the classical decomposition of a time series into seasonal and trend parts, and the use of Fourier-based frequency representations. Instead of processing the entire sequence purely in the time domain, the model selects subsets of frequency components and operates in the spectral domain. This frequency-enhanced treatment is what gives FEDformer its name and what distinguishes it from standard Transformer architectures. The evidence does not provide further architectural details beyond this high-level description.