EUVIMEDEUVIMED entdecken
Crossref

FMDformer: Frequency Mixed Decomposed Transformer for Time Series Forecasting

Xueli Liu, Shuaifei Lin, Nannan Xu

Concurrency and Computation: Practice and Experience · 25.12.2025 · 37 (27-28)

DOI: 10.1002/cpe.70434 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Concurrency and Computation: Practice and Experience, online erschienen am 17.11.2025. Verfasst von Xueli Liu und 2 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 42 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Time series forecasting has widespread applications in fields such as weather prediction and energy consumption. In recent years, Transformer‐based architectures have attracted significant attention due to their remarkable effectiveness in time series forecasting tasks. However, real‐world time series frequently display complex dynamic variations, and traditional methods, which mainly focus on extracting time‐domain features, struggle to effectively capture intricate periodic patterns. Furthermore, existing self‐attention mechanisms often prioritize low‐frequency features, potentially overlooking high‐frequency information, which may lead to model bias and information loss. To address this issue, we propose Frequency Mixed Decomposed Transformer (FMDformer), which directly trains in the frequency domain and incorporates sequence decomposition techniques to accurately capture and forecast complex periodic patterns within the data. Specifically, FMDformer alleviates the self‐attention mechanism's neglect of high‐frequency features by mixing the real and imaginary components of frequency elements in the frequency domain and progressively extracting low‐frequency components in both domains, enabling the model to effectively capture complex periodic variation patterns. We carried out comprehensive evaluations across six real‐world datasets to assess the effectiveness of FMDformer. The results show that FMDformer achieves superior consistency and robustness compared to the existing state‐of‐the‐art baseline models.

Zitieren

APA

Liu, X., Lin, S., & Xu, N. (2025). FMDformer: Frequency Mixed Decomposed Transformer for Time Series Forecasting. Concurrency and Computation: Practice and Experience, 37(27-28). https://doi.org/10.1002/cpe.70434

Vancouver

Liu X, Lin S, Xu N. FMDformer: Frequency Mixed Decomposed Transformer for Time Series Forecasting. Concurrency and Computation: Practice and Experience. 2025;37(27-28). doi:10.1002/cpe.70434

BibTeX
@article{Liu2025,
  author = {Liu, Xueli and Lin, Shuaifei and Xu, Nannan},
  title = {FMDformer: Frequency Mixed Decomposed Transformer for Time Series Forecasting},
  journal = {Concurrency and Computation: Practice and Experience},
  year = {2025},
  volume = {37},
  number = {27-28},
  doi = {10.1002/cpe.70434}
}
Kostenlos starten →