FPN-fusion: Enhanced Linear Complexity Time Series Forecasting Model

Chu Li,Pingjia Xiao,Qiping Yuan

2024-06-06

Abstract:This study presents a novel time series prediction model, FPN-fusion, designed with linear computational complexity, demonstrating superior predictive performance compared to DLiner without increasing parameter count or computational demands. Our model introduces two key innovations: first, a Feature Pyramid Network (FPN) is employed to effectively capture time series data characteristics, bypassing the traditional decomposition into trend and seasonal components. Second, a multi-level fusion structure is developed to integrate deep and shallow features seamlessly. Empirically, FPN-fusion outperforms DLiner in 31 out of 32 test cases on eight open-source datasets, with an average reduction of 16.8% in mean squared error (MSE) and 11.8% in mean absolute error (MAE). Additionally, compared to the transformer-based PatchTST, FPN-fusion achieves 10 best MSE and 15 best MAE results, using only 8% of PatchTST's total computational load in the 32 test projects.

Machine Learning,Artificial Intelligence

What problem does this paper attempt to address?

The paper attempts to address several key challenges in time series forecasting:

1. **Limitations of Data Decomposition Methods**: Traditional time series forecasting methods typically decompose data into trend and seasonal components, then predict each separately and finally combine the results. While effective, this approach has two main issues:
- It ignores the intrinsic relationship between trend and seasonal features, leading to information loss.
- The methods of separately extracting and merging trend and seasonal features fail to fully capture the complex interactions between these features.

2. **Balance of Computational Complexity and Number of Parameters**: Existing deep learning models, while improving prediction accuracy, often come with high computational complexity and a large number of parameters, limiting their scalability and efficiency in practical applications.

To address these issues, the paper proposes a new time series forecasting model—FPN-fusion. The main innovations of this model include:

- **Feature Pyramid Network (FPN)**: Describes time series data through multi-layer operations (such as pooling functions), replacing traditional raw data methods and binary decomposition, and more efficiently extracting shallow (seasonal and trend) and deep features.
- **Multi-level Fusion Mechanism**: Combines deep trend information and low-level features extracted by FPN, gradually fusing these features to improve the model's prediction performance.

Experimental results show that FPN-fusion performs excellently on multiple public datasets, significantly improving prediction accuracy and computational efficiency compared to existing linear complexity models (such as DLiner) and Transformer-based models (such as PatchTST).