AFSTGCN: Prediction for multivariate time series using an adaptive fused spatial-temporal graph convolutional network

• DOI: 10.1016/j.dcan.2022.06.019
• Publication Date: 2022-07-02T16:07:21Z
• Authors: Yuteng Xiao; Kaijian Xia; Hongsheng Yin; Yu-Dong Zhang; Zhenjiang Qian; Zhaoyang Liu; Yuehan Liang; Xiaodan Li
• Keywords: Spatial correlation; Adjacency matrix

The prediction for Multivariate Time Series (MTS) explores the interrelationships among variables at historical moments, extracts their relevant characteristics, and is widely used in finance, weather, complex industries other fields. Furthermore, it important to construct a digital twin system. However, existing methods do not take full advantage of potential properties variables, which results poor predicted accuracy. In this paper, we propose Adaptive Fused Spatial-Temporal Graph Convolutional Network (AFSTGCN). First, address problem unknown spatial-temporal structure, (AFSTG) layer. Specifically, fuse graph based on interrelationship spatial graphs. Simultaneously, adaptive adjacency matrix using node embedding methods. Subsequently, overcome insufficient extraction disordered correlation features, (AFSTGC) module. module forces reordering temporal, dependencies into rule-like data. AFSTGCN dynamically synchronously acquires correlations, thereby fully extracting rich hierarchical feature information enhance Experiments different types MTS datasets demonstrate that model achieves state-of-the-art single-step multi-step performance compared with eight deep learning models.