L'individuazione di variazioni in sistemi dinamici: una proposta per la scelta di parametri per l'algoritmo "generalized Likelihood ratio"

Abstract

In statistics and engineering the problem of change point detection in dynamical systems can be addressed using the class of sequential detection rules called "window limited generalized likelihood ratio (GLR) schemes". In this paper, some open problems concerning the statistical theory and practical implementation of these procedures are discussed. A method to select the window size and the threshold value, based on the joint use of Monte Carlo simulation and theoretical results, is proposed and illustrated through an example.