SDGF: Fusing Static and Multi-Scale Dynamic Correlations for Multivariate Time Series Forecasting
Shaoxun Wang, Xingjun Zhang, Qianyang Li, Jiawei Cao, Zhendong Tan
TL;DR
The paper tackles the challenge of forecasting multivariate time series by capturing inter-series dependencies that evolve across multiple temporal scales. It proposes SDGF, a dual-path graph fusion network that uses a static prior graph and per-scale dynamic graphs derived from multi-level wavelet decomposition, fused by an attention gate and processed with multi-kernel dilated convolutions. The approach achieves state-of-the-art or competitive performance on seven real-world datasets, particularly excelling on data with strong multi-scale patterns. The work provides a scalable framework that blends prior structural knowledge with data-driven scale-aware relationships, with public code for reproduction.
Abstract
Accurate multivariate time series forecasting hinges on inter-series correlations, which often evolve in complex ways across different temporal scales. Existing methods are limited in modeling these multi-scale dependencies and struggle to capture their intricate and evolving nature. To address this challenge, this paper proposes a novel Static-Dynamic Graph Fusion network (SDGF), whose core lies in capturing multi-scale inter-series correlations through a dual-path graph structure learning approach. Specifically, the model utilizes a static graph based on prior knowledge to anchor long-term, stable dependencies, while concurrently employing Multi-level Wavelet Decomposition to extract multi-scale features for constructing an adaptively learned dynamic graph to capture associations at different scales. We design an attention-gated module to fuse these two complementary sources of information intelligently, and a multi-kernel dilated convolutional network is then used to deepen the understanding of temporal patterns. Comprehensive experiments on multiple widely used real-world benchmark datasets demonstrate the effectiveness of our proposed model. Code is available at https://github.com/shaoxun6033/SDGFNet.