Towards Automatic Selection of Evolutionary Algorithms for an Engineering Problem: A Portfolio Approach
DescriptionEvolutionary algorithms (EA) are powerful meta-heuristics for solving engineering optimization problems. Many EA have been proposed in recent years. Unfortunately, there is little study or guideline on how to automatically select the best algorithm given a problem. As a result, much time and energy is expended by general users on selecting suitable EA.In this research, we wish to study the solution of this problem by a novel approach: We select a subset of powerful EA and put them into a portfolio. A novel predictive measure of performance is proposed. A new solution is generated by the algorithm with the best predicted performance.We propose to study the many interesting properties of this approach. This includes the convergence properties as a function of the computational budget; influence of the choice of an unsuitable algorithm, and comparison with existing methods. We also propose to study better performance measures.
|Effective start/end date||1/05/12 → 13/11/14|