Information-theoretic system identification
Abstract
The essential problem of system identification is to generate a useful representation of a system from measurements of the environment upon which the system acts and is acted upon. If the environmental influences are unknown a priori, they must ultimately be determined through direct or indirect empirical observation. The purpose of this investigation is to develop and demonstrate a principled information-theoretic method for the efficient identification of a dynamical system given only the assumption that the system is causal in respect to some subset of observed measurements. For convenience, all measurements are synchronous in time. Notable secondary contributions of the investigation include: a method of estimation of an information-theoretically optimal uniform interval of probabilistic structure of dependent data sets, a modified bootstrap method for the estimate of confidence intervals of mutual information estimates of spatially dependent data sets, and a computationally efficient branch-and-bound algorithm for supervised nonlinear input set selection.
Degree
Ph.D.
Advisors
Meckl, Purdue University.
Subject Area
Electrical engineering|Mechanical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.