Phase Space Graph Convolutional Network for Chaotic Time Series Learning
Weikai Ren, Ningde Jin, Lei OuYang
Abstract
Complex network has been a powerful tool for time series analysis by encoding dynamical temporal information in network topology. In this article, we introduce a framework to build a bridge between complex network and artificial intelligence for chaotic time series analysis. First, the chaotic time series are transformed to graph signals by phase space embedding. Then, the node information has been aggregated along the links through a cutting-edge technology termed graph convolutional network. We tested this method in the typical chaos system, and the phase space graph convolutional network (PSGCN) achieves better performance in the system control parameter prediction. To validate it in practical application, PSGCN is utilized in the flow-parameter prediction of gas-liquid two phase flow. The result indicates that complex network combined with graph convolutional network provide a potential perspective for exploring chaotic time series in practice.