Research On Nonlinear Time Series Classification Based On Recurrence Plots Image And State Transition Network Image Analysis Methods

Nonlinear time series classification is one of the most important problems in data analysis,pattern recognition and application.Many important tasks of pattern recognition and classification involve time series analysis,such as biomedical signals,financial data,industrial equipment,biometrics,video processing,music mining,weather prediction and so on.In view of the fact that the traditional time series classification research based on statistical ideas has developed more perfect,this paper attempts to carry out some new research perspectives proposed by researchers in recent years for the research of nonlinear time series classification,and the relevant methods have become a useful supplement to the traditional time series classification.First of all,the time series to be analyzed are transformed into recurrence plot,and the recursive features on the graph are quantitatively analyzed.The recursive quantitative features extracted by calculation are used as explanatory vectors for the classification of time series.This algorithm achieves ideal classification results on some data sets,which provides a new research perspective and algorithm flow reference for the traditional classification of time series.Secondly,based on the newly proposed Gramian Angular Field,Markov Transition Field and the improved transfer matrix method,this paper transforms the time series to be analyzed,mapping the "nonlinear time series set" to the "image data set" to be analyzed.On this basis,this paper attempts to transform the quasi classified time series into image objects for classification.That is,after transforming "time series data set" into "image data set",based on convolutional neural network analysis method,we carried out the classification and Application Research of typical nonlinear time series based on convolutional neural network classifier(CNN),and achieved good classification results;furthermore,this paper also discussed the inverse process of this analysis process and its application.Finally,this paper also carries out the correlation prediction research based on the image feature analysis of time series.In this paper,recurrence plot analysis,graph network analysis and convolution neural network are combined to propose a series of algorithm flow and technical implementation for nonlinear time series classification.The improvement of related algorithms and the design of auxiliary software provide a new methodology and technical basis for data classification in specific scenarios,which has a certain theoretical value and application prospect.