基于有限穿越可视图的时间序列网络模型

提出了一种改进的时间序列有限穿越可视图建网方法,并对三种可视图(可视图、水平可视图、有限穿越可视图)网络度分布进行了评价.结果表明:水平可视图网络均无法有效识别各类时间序列信号(周期、分形、混沌);对分形信号,可视图及有限穿越可视图网络均具有无标度幂律形式,但抗噪能力较差;对周期信号及混沌信号,有限穿越可视图网络比可视图具有更强的抗噪性.在此基础上,采用有限穿越可视图网络从油气水三相流电导波动信号中提取了度分布特征参数,通过其特征参数组合实现了对三种典型三相流流型(水包油泡状流、水包油泡状-段塞过渡流型及水包油段塞流)较好的辨识效果.

Limited penetrable visibility graph for establishing complex network from time series

We propose an improved visibility graph method, i.e., limited penetrable visibility graph, for establishing complex network from time series. Through evaluating the degree distributions of three visibility algorithms(visibility graph, horizontal visibility graph, limited penetrable visibility graph), we find that the horizontal visibility graph cannot distinguish signals from periodic, fractal, and chaotic systems; for fractal signal, the degree distributions obtained from visibility graph and limited penetrable visibility both can be well fitted to a power-law(scale-free distribution), but the anti-noise ability is not good; for periodic and chaotic signals, the limited penetrable visibility graph shows better anti-noise ability than visibility graph. In this regard, we use the limited penetrable visibility graph to extract the network degree distribution parameters from conductance fluctuating signals measured from oil-gas-water three-phase flow test. The results indicate that combination parameters of network degree distribution can be used to classify typical three phase flow patterns, e.g., oil-in-water bubble flow, bubble-slug transitional flow and slug flow.