Performance Analysis Of Complex-Network-Based Time Series Analysis Methods And Application Research

Time series analysis plays a very important role in the study in areas such as physics,biology,finance,however,nonlinear nonstationary time series analysis has always been a challenging problem for researchers.Emerging in recent years,complex-network-based time series analysis methods have been used by domestic and foreign researchers to study the important features contained in nonlinear time series.However,there still are some shortcomings in existing research: currently,the researchers put forward a variety of complex network analysis methods,and have proved the validity of the method on some specific objects,but haven't been studied for universal purposes;Complex-network-based time series analysis methods have been widely applied,but there is still something to improve in network-constructing process such as model representation,similarity measure and anti-noise performance;There is little application in complex nonlinear wind field analysis reported on complex network analysis method.In view of the above problems,this thesis review complex network method to test and compare the performance of these two kinds of complex network analysis method.In the aspect of application,this thesis puts forward a new method of wind field of time series classification,main work is as follows:firstly,we compared the performance of recurrence complex networks and the visibility graph method in characterizing typical nonlinear system.By comparing the two methods in different Hurst index,fractal Brownian motion sequence,and in the classification of the Logistic Map system of different chaotic states,the results show that the recurrence network algorithm is superior to the visibility graph in characterizing fractal Brownian motion sequence and Logistic Map system;When the Hurst index of fractal Brownian motion increases their corresponding recurrence complex network shows a trend community structure merging;both Recurrence networks and visibility graph need improve.Secondly,this thesis proposed two improved methods of transform time series to complex.By improving the similarity measure of construct recurrence network,we put forward a improvement recurrence network methods based on Dynamic Time Warping(DTW)distance.We put forward a method based on instantaneous energy spectrum and the visibility graph,the method using the Hilbert-huang transform the timing signal into instantaneous energy spectrum,based on the instantaneous energy spectrum we construct visibility graph.Test results show that the new method's ability of extracting nonlinear features is better than the original method.Thirdly,this thesis proposes a fusion of the EMD-based noise reduction and DTW recurrence network to classify wind field signal.By using the EMD and recurrence quantitative analysis method we put forward a signal de-noising method.Together with DTW recurrence complex network method,they can effectively classify the indoor and outdoor wind speed signal.