Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SparseTSF: Modeling Long-term Time Series Forecasting with 1k Parameters

Shengsheng Lin, Weiwei Lin, Wentai Wu, Haojun Chen, Junjie Yang

TL;DR

SparseTSF introduces Cross-Period Sparse Forecasting, a principle that decouples periodicity from trend by downsampling time series with a known period and forecasting across sparse subsequences with shared parameters. The model achieves competitive long-term forecasting performance using fewer than 1k parameters, supported by a theoretical analysis of parameter efficiency and an empirical demonstration across standard LTSF benchmarks. It also shows strong generalization under channel independence and simple normalization, making it suitable for resource-constrained settings. The approach offers substantial efficiency gains over state-of-the-art methods while maintaining robust accuracy, and it highlights future work for ultra-long and multi-period data scenarios.

Abstract

This paper introduces SparseTSF, a novel, extremely lightweight model for Long-term Time Series Forecasting (LTSF), designed to address the challenges of modeling complex temporal dependencies over extended horizons with minimal computational resources. At the heart of SparseTSF lies the Cross-Period Sparse Forecasting technique, which simplifies the forecasting task by decoupling the periodicity and trend in time series data. This technique involves downsampling the original sequences to focus on cross-period trend prediction, effectively extracting periodic features while minimizing the model's complexity and parameter count. Based on this technique, the SparseTSF model uses fewer than *1k* parameters to achieve competitive or superior performance compared to state-of-the-art models. Furthermore, SparseTSF showcases remarkable generalization capabilities, making it well-suited for scenarios with limited computational resources, small samples, or low-quality data. The code is publicly available at this repository: https://github.com/lss-1138/SparseTSF.

SparseTSF: Modeling Long-term Time Series Forecasting with 1k Parameters

TL;DR

SparseTSF introduces Cross-Period Sparse Forecasting, a principle that decouples periodicity from trend by downsampling time series with a known period and forecasting across sparse subsequences with shared parameters. The model achieves competitive long-term forecasting performance using fewer than 1k parameters, supported by a theoretical analysis of parameter efficiency and an empirical demonstration across standard LTSF benchmarks. It also shows strong generalization under channel independence and simple normalization, making it suitable for resource-constrained settings. The approach offers substantial efficiency gains over state-of-the-art methods while maintaining robust accuracy, and it highlights future work for ultra-long and multi-period data scenarios.

Abstract

This paper introduces SparseTSF, a novel, extremely lightweight model for Long-term Time Series Forecasting (LTSF), designed to address the challenges of modeling complex temporal dependencies over extended horizons with minimal computational resources. At the heart of SparseTSF lies the Cross-Period Sparse Forecasting technique, which simplifies the forecasting task by decoupling the periodicity and trend in time series data. This technique involves downsampling the original sequences to focus on cross-period trend prediction, effectively extracting periodic features while minimizing the model's complexity and parameter count. Based on this technique, the SparseTSF model uses fewer than *1k* parameters to achieve competitive or superior performance compared to state-of-the-art models. Furthermore, SparseTSF showcases remarkable generalization capabilities, making it well-suited for scenarios with limited computational resources, small samples, or low-quality data. The code is publicly available at this repository: https://github.com/lss-1138/SparseTSF.
Paper Structure (48 sections, 3 theorems, 13 equations, 7 figures, 14 tables, 1 algorithm)

This paper contains 48 sections, 3 theorems, 13 equations, 7 figures, 14 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Development of Long-term Time Series Forecasting
  4. Progress in Lightweight Forecasting Models
  5. Methodology
  6. Preliminaries
  7. Long-term Time Series Forecasting
  8. Channel Independent Strategy
  9. SparseTSF
  10. Cross-Period Sparse Forecasting
  11. Instance Normalization
  12. Loss Function
  13. Theoretical Analysis
  14. Parameter Efficiency of SparseTSF
  15. Effectiveness of SparseTSF
  16. ...and 33 more sections

Key Result

Theorem 3.1

Given a historical look-back window length $L$, a forecast horizon $H$, and a constant periodicity $w$, the total number of parameters required for the SparseTSF model is $\left \lfloor \frac{L}{w} \right \rfloor \times \left \lfloor \frac{H}{w} \right \rfloor + 2 \times \left \lfloor \frac{w}{2} \

Figures (7)

  • Figure 1: Comparison of MSE and parameters between SparseTSF and other mainstream models on the Electricity dataset with a forecast horizon of 720.
  • Figure 2: SparseTSF architecture.
  • Figure 3: Comparison of autocorrelation in original and downsampled subsequences for the first channel in the ETTh1 dataset.
  • Figure 4: Visualization of normalized weights of the model trained on the ETTh1 dataset with both look-back length (X-axis) and forecast horizon (Y-axis) of 96.
  • Figure 5: Schematic illustration of SparseTSF.
  • ...and 2 more figures

Theorems & Definitions (8)

  • Theorem 3.1
  • Definition 3.2
  • Lemma 3.3
  • Theorem 3.4
  • Definition 3.5: AutoCorrelation Function (ACF) acf
  • proof
  • proof
  • proof