Learning Dynamic Temporal Relations with Continuous Graph for Multivariate Time Series Forecasting (Student Abstract)

The recent advance in graph neural networks (GNNs) has inspired a few studies to leverage the dependencies of variables for time series prediction. Despite promising results, existing GNN-based models cannot capture global dynamic relations between owing inherent limitation their learning module. Besides, multi-scale temporal information is usually ignored or simply concatenated prior methods, resulting inaccurate predictions. To overcome these limitations, we present CGMF, Continuous Graph method Multivariate Forecasting (CGMF). Our CGMF consists continuous module incorporating differential equations long-range intra- and inter-relations embedding sequence. We also introduce controlled equation-based fusion mechanism that efficiently exploits representations form evolutional dynamics learn rich patterns shared across different scales. Comprehensive experiments demonstrate effectiveness our variety datasets.