Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.485082
Title: System Identification for Cellular Automata with Applications to Excitable Media
Author: Zhao, Yifan
Awarding Body: The University of Sheffield
Current Institution: University of Sheffield
Date of Award: 2006
Availability of Full Text:
Access from EThOS:
Abstract:
As an important class of spatia-temporal systems, cellular automata (CA) have attracted more and more attention from researchers, but only a few have studied how to extract the CA rules directly from observed data, especially from real data. This thesis therefore centres around an investigation into the identification of CA and the application to real data with the aim of building a link connecting the behaviour of real spatia-temporal systems with CA model realizations. Some basic concepts associated with the developments of CA and the corresponding identification are reviewed initially in a literature survey to provide the motivation and background for this study. The applications of four modified orthogonalleast squares (OL8) methods in the identification of CA are then investigated to select one as the core algorithm for use in this thesis. After a discussion on the improvement of all methods, a group of tests is conducted to compare the feasibility, speed and accuracy of each algorithm when dealing with CA data. The main contribution in theoretical research of this thesis has been the intraduction of a new neighbourhood detection method using mutual information. This has important advantageous over existing methods because it can detect an exact or slightly large neighbourhood without any a prior information. With excellent approximation properties and easy implementation, the new method is well-suited for deterministic CA, probabilistic CA, and even highly noise corrupted systems. By combining the new neighbourhood detection method with the OL8 estimator, a coarse-to-fine approach is then proposed which provides a generic routine to extract polynomial models directly from observed data for binary CA., The identification of a CA model of an excitable media system directly from observed data is demonstrated for the first time. A multi-model realization is introduced to represent the excitable media system and the identification of the system model is achieved by transforming the identification problem from excitable media to binary CA. To further evaluate the effectiveness of the methods on real data, the new algorithms are then applied to data imaged from a Belousov-Zhabotinsky (BZ) chemical reaction. To extend the application of the proposed methods, the identification of hybrid CA which can represent complex behaviours and which has never been studied before, is investigated. Based on the difference in the evolution characteristics in different regions, segmentation methods are employed before estimating the CA models. The final representation of hybrid CA can then be obtained by combining the mathematical models of each region. Numerous examples using simulation and real data are used to demonstrate the effectiveness and applicability of the new methods.
Supervisor: Not available Sponsor: Not available
Qualification Name: The University of Sheffield, 2006 Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.485082  DOI: Not available
Share: