方法对比

	Temporal Fusion Transformer	ARIMA（自回归积分滑动平均）模型	DeepAR
领域	深度学习	计量经济学	深度学习
方法族	Machine learning	Regression model	Machine learning
起源年份	2021	2015	2020
提出者	Lim, B., Arık, S. Ö., Loeff, N. & Pfister, T.	Box & Jenkins (Box-Jenkins methodology)	Salinas, D., Flunkert, V. & Gasthaus, J. (Amazon)
类型	Attention-based deep learning forecasting architecture	Univariate time-series model	Autoregressive recurrent neural network (probabilistic forecasting)
开创性文献	Lim, B., Arık, S. Ö., Loeff, N. & Pfister, T. (2021). Temporal Fusion Transformers for Interpretable Multi-Horizon Time Series Forecasting. International Journal of Forecasting, 37(4), 1748–1764.	Box, G. E. P., Jenkins, G. M., Reinsel, G. C. & Ljung, G. M. (2015). Time Series Analysis: Forecasting and Control (5th ed.). Wiley. ISBN: 978-1118675021	Salinas, D., Flunkert, V., Gasthaus, J. & Januschowski, T. (2020). DeepAR: Probabilistic Forecasting with Autoregressive Recurrent Networks. International Journal of Forecasting, 36(3), 1181–1191.
别名	Temporal Fusion Transformer (TFT), TFT, interpretable multi-horizon forecasting transformer	Box-Jenkins model, ARIMA(p,d,q), ARIMA Modeli	DeepAR — Olasılıksal RNN Tahmini, probabilistic autoregressive RNN forecasting, Amazon DeepAR
相关	6	5	5
摘要	The Temporal Fusion Transformer (TFT), introduced by Lim, Arık, Loeff and Pfister in 2021, is an interpretable deep learning architecture for multi-horizon time series forecasting. It combines variable selection, gating, multi-horizon attention and quantile outputs, processing static, past and known-future inputs together to produce multi-step forecasts.	ARIMA is a univariate time-series forecasting model that combines autoregressive, integrated (differencing), and moving-average components to predict a single continuous series from its own past. It is the centrepiece of the Box-Jenkins methodology set out in Box, Jenkins, Reinsel & Ljung's Time Series Analysis (5th ed., 2015).	DeepAR is Amazon's industrial forecasting model, introduced by Salinas, Flunkert and Gasthaus (2017; published 2020), that uses an autoregressive recurrent neural network to estimate the parameters of a probability distribution at each step, producing a confidence interval rather than a single point forecast. It can model many related time series jointly within one model.