Improving Pareto Optimal Designs Using Genetic Algorithms
John S. Gero, Sushil J. Louis · Computer-Aided Civil and Infrastructure Engineering · 1995
Abstract: Pareto optimal designs are the best designs that can be produced for a given problem formulation for a given set of criteria when the criteria are not combined in any way. If the goal is to improve the performance in those criteria, then it is possible to manipulate the problem formulation to achieve an improvement. The approach adopted is to encode the formulation in a genetic algorithm and to allow the formulation to evolve in the direction of improving Pareto optimal designs. A set of rules (in the form of a shape grammar), the execution of which produces a design, is encoded as the genes in a genetic algorithm. However, the rule set is allowed to evolve, not just the order of execution of rules. We present an example demonstrating both the approach and its utility in improving Pareto optimal designs.