Selecting Simulation Algorithm Portfolios by Genetic Algorithms

Abstract

An algorithm portfolio is a set of algorithms that are bundled together for increased overall performance. While being mostly applied to computationally hard problems so far, we investigate portfolio selection for simulation algorithms and focus on their application to adaptive simulation replication. Since the portfolio selection problem is itself hard to solve, we introduce a genetic algorithm to select the most promising portfolios from large sets of simulation algorithms. The effectiveness of this mechanism is evaluated by data from both a realistic performance study and a dedicated test environment.