An Efficient Experimental Methodology for Configuring Search-Based Design Algorithms

Many problems in high assurance systems design are only tractable using computationally expensive search algorithms. For these algorithms to be useful, designers must be provided with guidance as to how to configure the algorithms appropriately. This paper presents an experimental methodology for deriving such guidance that remains efficient when the algorithm requires substantial computing resources or takes a long time to find solutions. The methodology is shown to be effective on a highly-constrained task allocation algorithm that provides design solutions for high integrity systems. Using the methodology, an algorithm configuration is derived in a matter of days that significantly outperforms one resulting from months of ´trial-and-error´ optimisation.