Global Optimization with Derivative-free, Derivative-based and Evolutionary Algorithms

This paper investigates global optimization methodsfrom the perspective of population-based and restarted pointbasedheuristics. We examine the performance of a standardevolutionary computation (EC) methodology, a derivative-basedsequential quadratic programming (SQP) algorithm and a novelderivative-free stochastic coordinate ascent (SCA) algorithm. Allmethods are analyzed by random sampling of the feasible searchspace. A comparison was made to evaluate the three algorithms,in the light of newly updated IEEE CEC2013 benchmarks, on aset of multimodal and composite test cases. Results revealed thatwhile the standard EC algorithm is generally more robust, onthe basis of convergence efficiency both the restarted SCA andSQP algorithms have shown remarkable performance on someof these benchmarks. The results further suggest that dependingon the nature of the problem landscape and dimensionality thethree algorithms, chosen from different optimization frameworks,perform complementary to each other.