CTDFormer: controllable trend decomposition with dual-seasonal attention for multivariate time series forecasting

Abstract

Time series forecasting refers to constructing models based on historical observations to predict future values or trends. This method is widely applied in fields such as marine environmental monitoring, meteorological prediction, and power load forecasting. To address the issue of unstable trend prediction in multivariate time series and further enhance the capability to capture dependencies across variables, a multivariate time series forecasting model with controllable trend decomposition (CTDFormer) is proposed. This model comprises three modules: 1) the controllable trend decomposition module introduces control parameters along with similarity constraints to decompose the series optimally, yielding trend and seasonal information at an appropriate granularity; 2) the independent trend modeling module eliminates incidental interference between variables, independently predicts the decomposed trend information, and progressively constructs a global trend representation through fine-grained subsequence analysis for each variable; and 3) the dual-seasonal attention module integrates correlation analysis of seasonal information within and between variables, enhancing the model’s ability to capture complex seasonal dependencies. Experimental results demonstrate that CTDFormer achieves significant improvements in forecasting accuracy with only 6.73 million parameters, attaining an optimal balance between computational efficiency and forecasting performance.

Keywords

Time series forecasting / Controllable trend decomposition / Independent trend modeling / Dual-seasonal attention

References

[1] Aigrain S, Foreman-Mackey D (2023) Gaussian process regression for astronomical time series. Preprint at arXiv:2209.08940

[2] Das A, Kong WH, Leach A, Mathur S, Sen R, Yu R (2024) Long-term forecasting with TiDE: time-series dense encoder. Preprint at arXiv:2304.08424

[3] Hansun S (2013) A new approach of moving average method in time series analysis. In: 2013 Conference on New Media Studies (CoNMedia). IEEE, pp 1–4. https://doi.org/10.1109/CoNMedia.2013.6708545

[4] Hu YF, Liu PY, Zhu P, Cheng DW, Dai T (2025) Adaptive multi-scale decomposition framework for time series forecasting. In: Proceedings of the 39th AAAI Conference on Artificial Intelligence. AAAI, pp 17359–17367. https://doi.org/10.1609/aaai.v39i16.33908

[5] Jhin SY, Kim S, Park N (2024) Addressing prediction delays in time series forecasting: a continuous GRU approach with derivative regularization. In: Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. ACM, pp 1234–1245. https://doi.org/10.1145/3637528.3671969

[6] Knapp AB. Maritime narratives of prehistoric cyprus: seafaring as everyday practice. J Mari Arch, 2020, 15: 415-450

[7] Liao K, Xuan X, Ma KL. Deep learning for time series forecasting: a survey of recent advances. Front Comput Sci, 2026, 20(11 2011359

[8] Liu Y, Hu TG, Zhang HR, Wu HX, Wang SY, Ma LT et al (2024) iTransformer: inverted transformers are effective for time series forecasting. Preprint at arXiv:2310.06625

[9] Masserano L, Ansari AF, Han B, Zhang XY, Faloutsos C, Mahoney MW et al (2025) Enhancing foundation models for time series forecasting via wavelet-based tokenization. In: Proceedings of the 42nd International Conference on Machine Learning, vol 267. PMLR, pp 43248–43275. https://proceedings.mlr.press/v267/masserano25a.html

[10] Nie YQ, Nguyen NH, Sinthong P, Kalagnanam J (2023) A time series is worth 64 words: long-term forecasting with transformers. Preprint at arXiv:2211.14730

[11] Pinkus A. Approximation theory of the MLP model in neural networks. Acta Numer, 1999, 8: 143-195

[12] Salinas D, Flunkert V, Gasthaus J, Januschowski T. DeepAR: probabilistic forecasting with autoregressive recurrent networks. Int J Forecast, 2020, 36(31181-1191

[13] Sherstinsky A. Fundamentals of recurrent neural network (RNN) and long short-term memory (LSTM) network. Physica D, 2020, 404 132306

[14] Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez AN (2017) Attention is all you need. Preprint at arXiv:1706.03762

[15] Wu HX, Hu TG, Liu Y, Zhou H, Wang JM, Long MS (2023) TimesNet: temporal 2D-variation modeling for general time series analysis. Preprint at arXiv:2210.02186

[16] Wu HX, Xu JH, Wang JM, Long MS (2021) Autoformer: decomposition transformers with auto-correlation for long-term series forecasting. Preprint at arXiv:2106.13008

[17] Xu ZJ, Zeng AL, Xu Q (2024) FITS: modeling time series with 10k parameters. Preprint at arXiv:2307.03756

[18] Yang X, Zhang ZG, Lu SH (2025) MSCGrapher: learning multi-scale dynamic correlations for multivariate time series forecasting. In: Proceedings of the 41st Conference on Uncertainty in Artificial Intelligence, vol 286. PMLR, pp 4731–4751. https://proceedings.mlr.press/v286/yang25c.html

[19] Zeng AL, Chen MX, Zhang L, Xu Q (2023) Are transformers effective for time series forecasting? In: Proceedings of the 37th AAAI Conference on Artificial Intelligence. AAAI, pp 11121–11128. https://doi.org/10.1609/aaai.v37i9.26317

[20] Zhang YH, Yan JC (2023) Crossformer: transformer utilizing cross-dimension dependency for multivariate time series forecasting. In: International Conference on Learning Representations. ICLR, pp 1–21. https://mlanthology.org/iclr/2023/zhang2023iclr-crossformer/

[21] Zhou HY, Zhang SH, Peng JQ, Li JX, Xiong H, Zhang WC (2021) Informer: beyond efficient transformer for long sequence time-series forecasting. In: Proceedings of the AAAI Conference on Artificial Intelligence. AAAI, pp 11106–11115. https://doi.org/10.1609/aaai.v35i12.17325

Funding

the National Natural ScienceFoundation of China(U24A20328)

Rights & permissions

The Author(s)