Trend recurrence analysis and time series classification via trend fuzzy granular recurrence plot method

Abstract

The recurrence plot (RP) method proposed by Eckmann et al. is a powerful tool for visualizing the recurrent states of a system and is successfully applied in time series classification. However, the traditional RP method is easily affected by noise, and cannot effectively express trend differences between time series especially opposite trends. To solve these problems, this paper proposes a method of trend fuzzy granular recurrence plot (TFGRP). The TFGRP method is implemented at the level of granular time series, which integrates linear fuzzy information granulation into time series recursive analysis. It can not only reduce the effect of noise on the signal but also address the trend confusion problem. Based on the TFGRP method, we combined the support vector machine (SVM) to propose a time series classification model named as TFGRP-SVM. The experimental results demonstrate that the TFGRP method can reduce the influence of noise on data and the TFGRP-SVM model has superior classification performance.