Network-based characterization of time series and its application to signal classification

Time series analysis in complex systems can help us to peep into the inner structure and operation law of the system so as to make relevant decisions. In this paper, we propose a binary symbolic pattern state transfer network for measuring the complexity of series. First, we capture the spatio-temporal characteristics of series through a weighted change pattern matrix, and then we define a new binary coding mode that accomplishes the conversion of complex series to symbolic series. In addition, we fully consider the temporal evolution of the series, construct a horizontal viewable view of the state transfer series and generate a complex network, and extract the relevant metrics. Simulation experiments verify the validity of the model and its robustness to parameters. Finally, the model is applied to physiological signal analysis. For two EEG datasets, we depicted the brain region activities of the subjects in different states and successfully categorized the subjects. In summary, our approach captures the intrinsic patterns and features of series from a new perspective and provides an effective way to measure the complexity of series, it also provides an effective way to recognize and classify complex signals.