方法对比

	Autoformer：用于长期时间序列预测的分解Transformer	FEDformer：频率增强分解Transformer	状态空间模型（卡尔曼滤波器）
领域	深度学习	深度学习	计量经济学
方法族	Machine learning	Machine learning	Regression model
起源年份	2021	2022	1990
提出者	Haixu Wu et al. (Tsinghua)	Tian Zhou et al.	Harvey; Durbin & Koopman (state space treatment); Kalman filter
类型	Decomposition-based deep forecasting model	Frequency-domain decomposed Transformer for time-series forecasting	State space time series model
开创性文献	Wu, H., Xu, J., Wang, J., & Long, M. (2021). Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting. NeurIPS, 34.	Zhou, T., Ma, Z., Wen, Q., Wang, X., Sun, L., & Jin, R. (2022). FEDformer: Frequency enhanced decomposed transformer for long-term series forecasting. ICML.	Harvey, A. C. (1990). Forecasting, Structural Time Series Models and the Kalman Filter. Cambridge University Press.
别名	Auto-Correlation Transformer, Decomposition Transformer, Series Decomposition Forecaster, Oto-Korelasyon Ayrışım Transformer	Frequency Enhanced Decomposed Transformer, FED-Transformer, Frequency Domain Transformer, Frekans Tabanlı Ayrıştırılmış Dönüştürücü	state space, Kalman filter, unobserved components model, Durum Uzayı Modeli (State Space / Kalman Filter)
相关	4	3	4
摘要	Autoformer is a deep learning architecture for long-term time-series forecasting, introduced by Wu et al. from Tsinghua University at NeurIPS 2021. It replaces the standard self-attention mechanism with an Auto-Correlation mechanism that exploits periodic dependencies in the frequency domain, and embeds a progressive series decomposition block throughout the encoder and decoder to separately model trend and seasonal components.	FEDformer is a Transformer-based architecture for long-term multivariate time-series forecasting, introduced by Zhou et al. at ICML 2022. Its core innovation is the combination of seasonal-trend decomposition with frequency-domain attention: instead of computing full token-to-token attention in the time domain, FEDformer projects queries, keys, and values into the frequency domain via Fourier or wavelet transforms and operates on a randomly selected subset of frequency components, achieving linear complexity while preserving global temporal structure.	A state space model is a general time series framework that describes a series through unobserved (latent) state variables linked by a measurement equation and a transition equation, with the states estimated in real time by the Kalman filter. Developed in the state space tradition of Harvey (1990) and Durbin & Koopman (2012), it nests ARIMA and exponential smoothing as special cases.