Genetic Algorithms in Decomposition Based Multicriteria Optimization

Abstract

The paper presents strategies for implementing genetic algorithms in decomposition based multicriteria design optimization. The decomposition approach requires that the system design problem be partitioned into smaller sized subsystems, and the system solution obtained as a combination of the solutions from the subsystems. The absence of gradient information in a genetic algorithm based search strategy requires alternative methods for coordinating the solutions in different subsystems. Two newly developed methods referred to as experiential inheritance and interspecies migration were used to coordinate the solutions of sub systems in the decomposition based approach. Both the weighted sum and weighted minimax methods were explored in the solution of the multicriteria design problem. The proposed strategies were validated through implementation in representative algebraic and structural design problems.