Accession Number : ADA296310
Title : Fault Tolerant Design Using Single and Multicriteria Genetic Algorithm Optimization.
Descriptive Note : Master's thesis,
Corporate Author : AIR FORCE INST OF TECH WRIGHT-PATTERSON AFB OH
Personal Author(s) : Schott, Jason R.
PDF Url : ADA296310
Report Date : MAY 1995
Pagination or Media Count : 203
Abstract : This thesis incorporates a mixed discrete/continuous parameter genetic algorithm optimization capability into the Design Optimization/Markov Evaluation (DOME) program developed by the Charles Stark Draper Laboratory of Cambridge, Massachusetts. DOME combines the merits of Markov modeling and the Optimal Design Process to generate a systematic framework for system design with realistic reliability and cost analyses. The addition of genetic algorithms expands the design problem domain to include discrete parameter problems, which current optimization methods continue to struggle with. A new variant of the genetic algorithm called the steady-state genetic algorithm is introduced to eliminate the idea of distinct generations. Functional constraints are dealt with by ingenious use of the function information contained in the genetic algorithm population. The optimal genetic algorithm parameter settings are investigated, and the genetic algorithm is compared to the Monte Carlo method and the Branch and Bound method to show its relative utility in optimization. This research shows that a single criterion genetic algorithm can be expected to outperform other methods in efficiency, accuracy, and speed on problems of moderate to high complexity. The work then extends to multicriteria optimization, as applied to fault tolerant system design. A multicriteria genetic algorithm is created as a competitive means of generating the efficient (Pareto) set. Method parameters such as cloning, sharing, domination pressure, and population variability are investigated. The method is compared to the epsilon-constraint multi criteria method with a steady-state genetic algorithm performing the underlying single-criterion optimization.
Descriptors : *ALGORITHMS, *OPTIMIZATION, *SYSTEMS ENGINEERING, *GENETIC ENGINEERING, *FAULT TOLERANT COMPUTING, VELOCITY, STEADY STATE, METHODOLOGY, COST ANALYSIS, PARAMETERS, ACCURACY, EFFICIENCY, THESES, MONTE CARLO METHOD, POPULATION, VARIATIONS, TOLERANCE, CLONES, GENETICS, MASSACHUSETTS, FAULTS.
Subject Categories : Genetic Engineering and Molecular Biology
Distribution Statement : APPROVED FOR PUBLIC RELEASE