A multivariate descriptor method for change-point detection in nonlinear time series

The purpose of this paper is to present a novel method that is applied to detect dynamic changes in nonlinear time series. The method combines a multivariate control chart that monitors the variation of three normalized descriptors -- Hjorth's descriptors of activity, mobility and complexity -- and is applied to the change-point detection problem of nonlinear time series. The approach is estimated using six simulated nonlinear time series. In addition, a case study of six time series of short-term electricity load consumption was used to illustrate the power of the method.