Detection and Diagnosis of Model Parameter and Noise Variance Changes Using Sliding Windows Approach

Abstract

The problem of change detection and data segmentation has received considerable attention in a research context and appears to be the central issue in various application areas. The change detection and segmentation model used in this paper is the simplest extension of the linear regression models to data with abruptly changing properties. Usually, a change detection algorithm consists of two stages: residual generation and decision making. The residuals represent the difference between the observed and expected system behavior. In the stage of decision making, the residuals are processed and analyzed under certain decision rules to determine the system change status. In the first part of the paper we give a general view on the main techniques used in change detection and segmentation: filtering techniques with a whiteness test and techniques based on sliding windows and distance measures. This patent review a new algorithm based on a likelihood technique when sliding windows are used, for diagnosis of model parameter and variance changes. The results of some Monte-Carlo simulation for detection and diagnosis of model parameter and variance changes are presented. PACS numbers: 02.50.-r, 02.70.-c, 95.75.Pq, 95.75.Wx Keywords: Change detection, diagnosis, regression models, likelihood techniques, Monte Carlo simulation, Noise Variance Changes, Change evaluation, fuzzy logic, fuzzy models, probabilistic simulation, adaptive filtering, real-time tracking, Learning patterns, mapping forest mortality, detection techniques