Performance measure and tool for benchmarking metaheuristic optimization algorithms

Abstract

In the last decade, many new algorithms have been proposed to solve optimization problems. Most of them are meta-heuristic algorithms. The issue of accurate performance measure of algorithms is still under discussion in the scientific community. Therefore, a new scoring strategy via a new benchmark is proposed. The idea of this new tool is to determine a score, a measure of efficiency taking into account both the end value of the optimization and the convergence speed. This measure is based on an aggregate of statistical results of different optimization problems. These problems are judiciously chosen to cover as broad a spectrum of resolution configurations as possible. They are defined by combinations of several parameters: dimensions, objective functions and evaluation limit on dimension ratios. Aggregation methods are chosen and set in order to make the problem weight relevant according to the computed score. This scoring strategy is compared to the CEC one thanks to the results of different algorithms: PSO, CMAES, Genetic Algorithm, Cuttlefish and simulated annealing.