|
Abstract:
|
Given the high computational cost of function evaluation in most real world problems, the development of efficient optimization algorithms has raised a lot of interest. In this paper, we describe the design of both an efficient multi-objective optimization approach and an enhancement technique to reduce computational cost. First, we propose a multi-objective optimizer based on the particle swarm strategy, that provides very competitive results. Then, we provide the first proposal to incorporate the concept of fitness inheritance into a multi-objective particle swarm optimizer. After a study of several different techniques to incorporate fitness inheritance into our approach, we conclude that this enhancement technique is able to reduce computational cost without dramatically deteriorating the quality of the results. Also, we show that when fitness inheritance is applied dynamically throughout the evolutionary process, savings of about 32% of the total number of evaluations can be obtained without affecting the quality of the solutions. Furthermore, even reducing the computational cost by 78%, the proposed approaches are able to obtain very good approximations of the true Pareto front.
|
