Research On Similarity Of Scene Time Series Based On Visibility Graph

With the wide application of time series data mining technology,it's an important research direction to study the similarity measure of time series.Traditional direct use of the time series correlation coefficient(Pearson Correlation Coefficient,Spearman Rank Correlation)and distance(Euclidean distance,Dynamic Time Warping)similarity measure method to deal with scene class time series ineffective.In order to deal with the problem,a scene time series similarity measure method based on the Visibility Graph network feature is proposed in this paper,through the in-depth analysis of the temporal features of the scene time series and combined with the tidal characteristics of the scene time series and the microscopic structural differences between different scene time series.The method is to convert the time series into a visibility graph complex network,extract the structural feature sequence of the network,and then calculate the distance between the feature sequences or the correlation coefficient as a measure of the similarity of the original time series.The scene time series has the characteristics of similarity in macro structure and difference in microstructure,and the complex network constructed by Visibility Graph can reflect the convex and concave properties of time series.In order to reflect the local structure differences of time series better,this paper proposes a limited-range Visibility Graph.The limited-range Visibility Graph maps the time series local structure pattern to a fewer visibility graph complex network motif.In the case of normalization of the distance space,the distance between the visibility graph network motifs is larger than the distance between the time series patterns.Because the classical visibility graph method has the phenomenon that multiple time series patterns corresponding to the same visibility graph network motif.In order to describe the difference of the same visibility graph network motif better,this paper proposes a weighted view model by using the visual angle,which can reflect the difference of the variation range of the time series,as the weight,so that this method can distinguish the different time series patterns in the same visibility graph network motif.This paper deeply analyzes the relationship between the local values of the time series and the visibility graph network motifs.We found local pattern differences in time series and can cause that small changes in the time series of local values may cause figure network characteristics such as node degree sequences in change under the condition of complex network structure alter.Hence the change of network structure changes the minute sequence of the time series in the large probability,after converting the time series into a complex network by the visibility graph method.And it is also inspected by analog data and empirical data.The experimental results show that the time series similarity measure method based on the characteristics of visibility graph,in the treatment of the macro structure similar to the microstructure of dissimilar scene time series similarity measure problem,compared with the method based on the original time series similarity,our method has more excellent performance.