TimeEmb: A Lightweight Static-Dynamic Disentanglement Framework for Time Series Forecasting

TL;DR

TimeEmb addresses non-stationarity in time-series forecasting by explicitly disentangling a globally learned time-invariant component from a dynamic, frequency-filtered component. It relies on a learnable embedding bank to capture static patterns and a learnable spectral filter to model dynamic changes in the frequency domain, followed by a lightweight MLP predictor. The approach achieves state-of-the-art accuracy with substantially lower computational cost and can serve as a plug-in to enhance existing forecasting models. This work provides interpretable insights into temporal patterns via spectral disentanglement and demonstrates strong performance across diverse real-world datasets with robust scalability.

Abstract

Temporal non-stationarity, the phenomenon that time series distributions change over time, poses fundamental challenges to reliable time series forecasting. Intuitively, the complex time series can be decomposed into two factors, \ie time-invariant and time-varying components, which indicate static and dynamic patterns, respectively. Nonetheless, existing methods often conflate the time-varying and time-invariant components, and jointly learn the combined long-term patterns and short-term fluctuations, leading to suboptimal performance facing distribution shifts. To address this issue, we initiatively propose a lightweight static-dynamic decomposition framework, TimeEmb, for time series forecasting. TimeEmb innovatively separates time series into two complementary components: (1) time-invariant component, captured by a novel global embedding module that learns persistent representations across time series, and (2) time-varying component, processed by an efficient frequency-domain filtering mechanism inspired by full-spectrum analysis in signal processing. Experiments on real-world datasets demonstrate that TimeEmb outperforms state-of-the-art baselines and requires fewer computational resources. We conduct comprehensive quantitative and qualitative analyses to verify the efficacy of static-dynamic disentanglement. This lightweight framework can also improve existing time-series forecasting methods with simple integration. To ease reproducibility, the code is available at https://github.com/showmeon/TimeEmb.

Figures (8)

• Figure 1: Efficiency and performance comparison on the Electricity dataset.
• Figure 2: TimeEmb framework overview.
• Figure 3: Disentangled features visualization in frequency spectrum. Frequency components from 20 to 49 are zoomed in at the top right corner.
• Figure 4: T-SNE visualization results.
• Figure 5: Key components contribution analysis.