MS-DFTVNet:A Long-Term Time Series Prediction Method Based on Multi-Scale Deformable Convolution
Chenghan Li, Mingchen Li, Yipu Liao, Ruisheng Diao
TL;DR
This work tackles long-term time series forecasting by introducing MS-DFTVNet, a multi-scale 3D deformable CNN that jointly models cross-variable dependencies and multi-scale temporal patches. It combines a Fourier-based Multi-Patch module to identify dominant periods with a context-aware dynamic deformable convolution that learns both amplitude modulation and kernel sampling offsets, enabling adaptive receptive fields. Across six public datasets, the method achieves state-of-the-art accuracy with an average improvement of about $7.5\%$ and maintains robustness at extended horizons, while exhibiting favorable computational efficiency compared with Transformer-based models. The contributions advance CNN-based time-series forecasting by effectively integrating multi-scale patch modeling and cross-variable correlations in a lightweight, scalable framework.
Abstract
Research on long-term time series prediction has primarily relied on Transformer and MLP models, while the potential of convolutional networks in this domain remains underexplored. To address this, we propose a novel multi-scale time series reshape module that effectively captures cross-period patch interactions and variable dependencies. Building on this, we develop MS-DFTVNet, the multi-scale 3D deformable convolutional framework tailored for long-term forecasting. Moreover, to handle the inherently uneven distribution of temporal features, we introduce a context-aware dynamic deformable convolution mechanism, which further enhances the model's ability to capture complex temporal patterns. Extensive experiments demonstrate that MS-DFTVNet not only significantly outperforms strong baselines but also achieves an average improvement of about 7.5% across six public datasets, setting new state-of-the-art results.