In this paper we propose some dissimilarity measure functions for trends of nonstationary time series.If time series are stationary,the cross correlation function can be applied as a similarity measure.However,the val...In this paper we propose some dissimilarity measure functions for trends of nonstationary time series.If time series are stationary,the cross correlation function can be applied as a similarity measure.However,the validity of the cross correlation function is lost for nonstationary time series.Moreover,the cross correlation function cannot be calculated if one of trends is constant.The proposed functions can be applied even if trends are constant and their values are determined through the minimization.The clustering is considered as an application of the dissimilarity measure.Furthermore the problem of the common trend within multiple time series is considered and the estimation algorithm is proposed.Usability of the proposed method is demonstrated by applying to series of COVID-19 cases in Japan.展开更多
文摘In this paper we propose some dissimilarity measure functions for trends of nonstationary time series.If time series are stationary,the cross correlation function can be applied as a similarity measure.However,the validity of the cross correlation function is lost for nonstationary time series.Moreover,the cross correlation function cannot be calculated if one of trends is constant.The proposed functions can be applied even if trends are constant and their values are determined through the minimization.The clustering is considered as an application of the dissimilarity measure.Furthermore the problem of the common trend within multiple time series is considered and the estimation algorithm is proposed.Usability of the proposed method is demonstrated by applying to series of COVID-19 cases in Japan.
