Use this URL to cite or link to this record in EThOS:
Title: On the development of knowledge driven optimisation methods : application to complex reactor network synthesis
Author: Ashley, Victoria M.
ISNI:       0000 0001 3430 4318
Awarding Body: University of Surrey
Current Institution: University of Surrey
Date of Award: 2004
Availability of Full Text:
Access from EThOS:
Access from Institution:
Existing methods for reactor network synthesis vary from graphical approaches such as Attainable Region to superstructure-based optimisation using both deterministic and stochastic methods. Complex reactor network applications, consisting of many components and highly non-linear reaction kinetics, tend to face problems using current methods. Dimensionality limitations, initialisation problems, convergence difficulties and excessive computational times are some of the shortcomings, In an effort to overcome these difficulties, this project introduces the concept of Knowledge Driven Optimisation, utilising systems knowledge in order to develop rules that help to focus the optimisation search on high performance regions. The method uses knowledge derived from kinetic infoimation to gain an understanding of the system and devise a set of rules to guide the optimisation search. Data mining techniques are engaged to analyse serial and parallel pathways, relating concentration and temperature variables to regions of high performance. Extracted trends are translated into optimal design rules, and applied to a customised Tabu Search. Rule violations identify directions for improvement and aide move selection, guiding the superstructure optimisation search towards well performing structures, achieving more effective knowledge-based decision making than can be realised by a random stochastic search alone. Results show optimal solutions obtained for a number of examples agree with published literature whilst achieving faster convergence and reduced computational times compared to standard Tabu Search. To increase rule performance automatically, dynamic rule updates are implemented to tune the rule limits as the optimal search progresses. A hybrid optimisation approach, combining the stochastic rule-based search with deterministic techniques, is developed to promote efficient fine-tuning of the final structure. Application of the methodology to complex systems is demonstrated through a biocatalytic case study, metabolism by Saccharomyces cerevisiae, and the knowledge-driven rule-based approach significantly outperforms random Tabu Search. Preliminary studies into nonisothermal applications trial the use of temperature profiles. Finally, parallel processing and Grid technology is briefly investigated to assess the potential for achieving results in reduced times.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available