Harmony search algorithm with ensemble of surrogate models

Recently, Harmony Search Algorithm (HSA) is gaining prominence in solving real-world optimization problems. Like most of the evolutionary algorithms, finding optimal solution to a given numerical problem using HSA involves several evaluations of the original function and is prohibitively expensive. This problem can be resolved by amalgamating HSA with surrogate models that approximate the output behavior of complex systems based on a limited set of computational expensive simulations. Though, the use of surrogate models can reduce the original functional evaluations, the optimization based on the surrogate model can lead to erroneous results. In addition, the computational effort needed to build a surrogate model to better approximate the actual function can be an overhead. In this paper, we present a novel method in which HSA is integrated with an ensemble of low quality surrogate models. The proposed algorithm is referred to as HSAES and is tested on a set of 10 bound-constrained problems and is compared with conventional HSA.