Article Details

Abstract

Time series forecasting has been widely applied in fields such as energy scheduling, traffic flow prediction, and financial risk control. However, most existing methods perform unified modeling in the time domain, making it difficult to simultaneously capture periodic patterns and non-stationary perturbations. To address this issue, this paper proposes a Frequency-domain Expert Fusion (FdEF) model. The proposed method explicitly separates periodic and non-stationary components through the Discrete Fourier Transform (DFT), employs a frequency learning branch and a wavelet-based branch for feature extraction, and introduces a learnable gating mechanism for adaptive fusion. Experimental results on multiple long-term forecasting benchmark datasets demonstrate that FdEF outperforms mainstream models in terms of accuracy, validating its effectiveness and robustness.

Keywords

time series forecasting; frequency-domain decoupling; dual-branch modeling; adaptive fusion

References

1. [1] Liang Hongtao, Liu Shuo, Du Junwei, etc A Review of the Application of Deep Learning in Time Series Prediction Research [J]. Computer Science and Exploration, 2023, 17 (6): 1285-1300
2. [2] WANG H，PENG J，HUANG F，et al. MICN: Multi-Scale Local and Global Context Modeling for Long-Term Series Forecasting [C]//The Eleventh International Conference on Learning Representations.Kigali：ICLR，2023：1-22.
3. [3] JACOBS R A，JORDAN M I，NOWLAN S J，et al. Adaptive Mixtures of Local Experts [J].Neural Computation，1991，3（1）：79-87.
4. [4] WU H，HU T，LIU Y，et al. TimesNet: Temporal 2D-Variation Modeling for General Time Series Analysis [C]//International Conference on Learning Representations.Kigali：ICLR，2023：1-23.
5. [5] ZHOU H，ZHANG S，PENG J，et al. Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting [C]//Proceedings of the AAAI Conference on Artificial Intelligence.Virtual：AAAI，2021，35（12）：11106-11115.
6. [6] ZHOU T，MA Z，WEN Q，et al. FEDformer: Frequency Enhanced Decomposed Transformer for Long-Term Series Forecasting [C]//International Conference on Machine Learning.Baltimore：PMLR，2022：27268-27286.
7. [7] LIU S，YU H，LIAO C，et al. Pyraformer: Low-Complexity Pyramidal Attention for Long-Range Time Series Modeling and Forecasting [C]//International Conference on Learning Representations.Virtual：ICLR，2022：1-20.
8. [8] OSGOOD B. The Fourier Transform and Its Applications [R].Stanford：Stanford University，2009.
9. [9] BROWN I J. A Wavelet Tour of Signal Processing: The Sparse Way [J].Investigacion Operacional，2009，30（1）：85-87.
10. [10] CHEN S A，LI C L，YODER N，et al. TSMixer: An All-MLP Architecture for Time Series Forecasting [J/OL].arXiv:2303.06053 [cs.LG].（2023-08-11）.https://arxiv.org/abs/2303.06053.
11. [11] ZHANG Y，YAN J. Crossformer: Transformer Utilizing Cross-Dimension Dependency for Multivariate Time Series Forecasting [C]//The Eleventh International Conference on Learning Representations.Kigali：ICLR，2023：1-14.
12. [12] WU H，XU J，WANG J，et al. Autoformer: Decomposition Transformers With Auto-Correlation for Long-Term Series Forecasting [J].Advances in Neural Information Processing Systems，2021，34：22419-22430.
13. [13] ZENG A，CHEN M，ZHANG L，et al. Are Transformers Effective for Time Series Forecasting? [C]//Proceedings of the AAAI Conference on Artificial Intelligence. Washington：AAAI，2023，37（9）：11121-11128.

About the authors

Liu Zihan (2000-), female, Han ethnicity, from Wenzhou, Zhejiang, is currently pursuing a master's degree in time series research; Corresponding author: Yan Hua (1971-), male, Han ethnicity, from Dazhou, Sichuan, professor, PhD, research direction: intelligent information systems.