Modified DIRECT Algorithm for Scaled Global Optimization Problems

DIRECT is a popular deterministic algorithm for global optimization problems. It can find the basins of attraction for global or local optima efficiently, especially when dimension is small. Recently, we have proposed a class of modified DIRECT algorithms to eliminate the sensitivities of the original DIRECT to linear scaling of the objective function. In this paper, we devote to find a specific algorithm with best performance among this class. We compare the performance of the modified DIRECT algorithms on the GKLS test set. Numerical results show that DIRECT-median performs outstanding among this class. What is more, numerical results also show that DIRECT-median can find solutions with high accuracy much more efficiently than the original DIRECT.