Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CycleNet: Enhancing Time Series Forecasting through Modeling Periodic Patterns

Shengsheng Lin, Weiwei Lin, Xinyi Hu, Wentai Wu, Ruichao Mo, Haocheng Zhong

TL;DR

The Residual Cycle Forecasting (RCF) technique is introduced, which utilizes learnable recurrent cycles to model the inherent periodic patterns within sequences, and then performs predictions on the residual components of the modeled cycles.

Abstract

The stable periodic patterns present in time series data serve as the foundation for conducting long-horizon forecasts. In this paper, we pioneer the exploration of explicitly modeling this periodicity to enhance the performance of models in long-term time series forecasting (LTSF) tasks. Specifically, we introduce the Residual Cycle Forecasting (RCF) technique, which utilizes learnable recurrent cycles to model the inherent periodic patterns within sequences, and then performs predictions on the residual components of the modeled cycles. Combining RCF with a Linear layer or a shallow MLP forms the simple yet powerful method proposed in this paper, called CycleNet. CycleNet achieves state-of-the-art prediction accuracy in multiple domains including electricity, weather, and energy, while offering significant efficiency advantages by reducing over 90% of the required parameter quantity. Furthermore, as a novel plug-and-play technique, the RCF can also significantly improve the prediction accuracy of existing models, including PatchTST and iTransformer. The source code is available at: https://github.com/ACAT-SCUT/CycleNet.

CycleNet: Enhancing Time Series Forecasting through Modeling Periodic Patterns

TL;DR

The Residual Cycle Forecasting (RCF) technique is introduced, which utilizes learnable recurrent cycles to model the inherent periodic patterns within sequences, and then performs predictions on the residual components of the modeled cycles.

Abstract

The stable periodic patterns present in time series data serve as the foundation for conducting long-horizon forecasts. In this paper, we pioneer the exploration of explicitly modeling this periodicity to enhance the performance of models in long-term time series forecasting (LTSF) tasks. Specifically, we introduce the Residual Cycle Forecasting (RCF) technique, which utilizes learnable recurrent cycles to model the inherent periodic patterns within sequences, and then performs predictions on the residual components of the modeled cycles. Combining RCF with a Linear layer or a shallow MLP forms the simple yet powerful method proposed in this paper, called CycleNet. CycleNet achieves state-of-the-art prediction accuracy in multiple domains including electricity, weather, and energy, while offering significant efficiency advantages by reducing over 90% of the required parameter quantity. Furthermore, as a novel plug-and-play technique, the RCF can also significantly improve the prediction accuracy of existing models, including PatchTST and iTransformer. The source code is available at: https://github.com/ACAT-SCUT/CycleNet.
Paper Structure (37 sections, 4 equations, 7 figures, 12 tables, 2 algorithms)

This paper contains 37 sections, 4 equations, 7 figures, 12 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related work
  3. CycleNet
  4. Residual cycle forecasting
  5. Periodic patterns modeling
  6. Residual forecasting
  7. Backbone
  8. Instance normalization
  9. Loss function
  10. Experiments
  11. Setup
  12. Datasets
  13. Baselines
  14. Environments
  15. Main results
  16. ...and 22 more sections

Figures (7)

  • Figure 1: Shared daily periodic patterns present in the Electricity dataset.
  • Figure 2: CycleNet architecture. CycleNet/Linear and CycleNet/MLP represent using a single-layer Linear model and a dual-layer MLP model, respectively, as the backbone of CycleNet. Here, $D=3$.
  • Figure 3: Alignments and repetitions of the recurrent cycles $Q$. Here, $D=1$.
  • Figure 4: Visualization of the periodic patterns learned by CycleNet/Linear. Panels (a-d) display different periodic patterns learned from different datasets, and panels (e-h) show different periodic patterns learned from different channels within the same dataset. The $i$ th indicates the index of the channel within the dataset.
  • Figure 5: Performance of CycleNet and comparative models with different look-back lengths. The forecast horizon is set as 96.
  • ...and 2 more figures