FPN-fusion: Enhanced Linear Complexity Time Series Forecasting Model
This work addresses efficient time series forecasting for applications requiring low computational cost, though it is incremental as it builds on existing methods like DLiner and PatchTST.
The study tackled time series forecasting by introducing FPN-fusion, a model with linear computational complexity that outperformed DLiner in 31 out of 32 test cases, reducing MSE by 16.8% and MAE by 11.8% on average, and used only 8% of PatchTST's computational load while achieving competitive results.
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.