FPN-fusion: Enhanced Linear Complexity Time Series Forecasting Model

TL;DR

FPN-fusion is a new time series forecasting model that uses a feature pyramid network and multi-level fusion to achieve high accuracy with linear complexity, outperforming existing models on multiple datasets.

Contribution

Introduces FPN-fusion, a novel linear-complexity model employing FPN and multi-level fusion for improved time series prediction without increasing parameters.

Findings

Outperforms DLiner in 31 of 32 test cases with 16.8% MSE reduction

Achieves 10 best MSE and 15 best MAE results compared to PatchTST

Uses only 8% of PatchTST's computational load

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.