时间序列预测：方法、进展与展望 

卢斌龙 

福建师范大学计算机与网络空间安全学院，福建福州，350117； 

摘要：时间序列数据因其内在的时间依赖性，在广泛领域具有关键价值。本文阐述了时间序列预测的理论框架， 追踪其方法论从传统统计学到机器学习、深度学习的演进路径。深度学习模型，如 CNN、RNN 和 Transformer，在 捕捉数据复杂模式和预测准确性方面表现卓越。文章系统性梳理了时间序列数据特性和预测任务类型，并展望了 异常检测、超参数优化等未来研究方向，以推动深度学习在时间序列分析中的创新应用。 

关键词：时间序列预测；自回归模型；深度学习 

参考文献

[1]WEN Q, ZHOU T, ZHANG C, 等. Transformers in Time Series: A Survey[A/OL]. arXiv, 2023[2024-11-22].

[2]LIU Y, WU H, WANG J, 等. Non-stationary Transformers: Exploring the Stationarity in Time Series Forecasting[A/OL]. arXiv, 2023[2024-11-22].

[3]OORD A van den, DIELEMAN S, ZEN H, 等. WaveNet: A Generative Model for Raw Audio[A/OL]. arXiv, 2016[2024-11-22].

[4]BAI S, KOLTER J Z, KOLTUN V. An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling[A/OL]. arXiv, 2018[2024-11-22].

[5]LIU M, ZENG A, CHEN M, 等. SCINet: Time Series Modeling and Forecasting with Sample Convolution and Interaction[A/OL]. arXiv, 2022[2024-11-22].

[6]WU H, HU T, LIU Y, 等. TimesNet: Temporal 2D-Variation Modeling for General Time Series Analysis[A/OL]. arXiv, 2023[2024-11-22].

[7]SALINAS D, FLUNKERT V, GASTHAUS J. DeepAR: Probabilistic Forecasting with Autoregressive Recurrent Networks[A/OL]. arXiv, 2019[2024-11-22].

[8]BERGSMA S, ZEYL T, ANARAKI J R, 等. C2FAR: Coarse-to-Fine Autoregressive Networks for Precise Probabilistic Forecasting[A/OL]. arXiv, 2023[2024-11-22].

[9]WITRAN: Water-wave Information Transmission and Recurrent Acceleration Network for Long-range Time Series Forecasting . [2024-11-22]

[10]ZHOU H, ZHANG S, PENG J, 等. Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting[A/OL]. arXiv, 2021[2024-11-22].

[11]WU H, XU J, WANG J, 等. Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting[A/OL]. arXiv, 2022[2024-11-22].

[12]NIE Y, NGUYEN N H, SINTHONG P, 等. A Time Series is Worth 64 Words: Long-term Forecasting with Transformers[A/OL]. arXiv, 2023[2024-11-22].

[13]WU Z, PAN S, LONG G, 等. Connecting the Dots: Multivariate Time Series Forecasting with Graph Neural Networks[A/OL]. arXiv, 2020[2024-11-22].

[14]YI K, ZHANG Q, FAN W, 等. FourierGNN: Rethinking Multivariate Time Series Forecasting from a Pure Graph Perspective[A/OL]. arXiv, 2023[2024-11-22].