Visibility Graph Based Network Model For Time Series

With the arrival of the era of big data, the theory and technology carried out on thetime series data mining has important guiding significance。with the couple of timesequence and complex networks, through the study of topology structure of network,wecan effectively dig out some new sequence structure characteristics of time series. In theareas of using the theory of complex network to analyse and forecast time series,theresearch is still in its infancy. there still exist many typical problems such as the noiseproblem, innovation of methods, there is no sure be innovated path for the more hiddendata characteristics, etc.In current academic circles, the existing three methods to analyse time series (VG,HVG, LPVG) through complex network theory can be able to inherit the basiccharacteristics of time series, But the corresponding performance such as noiseperformance and algorithm complexity all have many limitations.so in this paper, twoimprove methods (LPHVG and DLPVG) is put forward on the basis of the existing threemethods. LPHVG was conducted on the basis of the original algorithm, and DLPVGbreak the undirection of original algorithm and has deep improvement The improvedalgorithm has the following highlights, New algorithm consider the reversibility of timeseries, firstly propose the orientation of converting time series into a complex network,and compared the in degree distribution and out degree distribution of the directed network.The algorithm that have been improved can effectively extract the characteristics ofcomplex networks for noise time series, fractal time seriesand cycle time sequence, andits antinoise ability is more prominent than existing three network algorithm,the mostimportant is it highlight the reversible characteristics of time series.In addition, In themodel test,not as the previous literature, use the white noise time series model forinspection. Because white noise time series is the common data type in financial sector.Based on the research conclusion, this paper puts forward that we can use theirreversibility of time series to other algorithms such as VG, not only limited to LPVG andfor directed visibility graph algorithm we can analyse the correlation of in and out degree and so on deep research.