Emperical analysis of hyper-heuristic search algorithms in expensive numerical optimzation

Expensive optimization problems refer to real-world problems that will require a large amount of resources to run and solve. This has attracted significant recent interest from researchers to investigate simple yet highly efficient search methodologies for solving this problem domain. The main goal of this problem domain is to be able to locate desirable solutions within a short number of search iterations. In this paper, the implementation of a hyper-heuristic framework for solving expensive optimization problems is presented. Hyper-heuristics utilize a set of low-level heuristics that work together to search for optimum solutions. Although hyper-heuristics have been shown to outperform many other search methodologies in discrete optimization, to the best of our knowledge, hyper-heuristics have yet to be investigated for expensive optimization problems. Two variants of hyper-heuristics are used in this paper, Simple-Random All Moves Acceptance (SRAMA) and Tabu Search All Moves Acceptance (TSAMA). The Congress on Evolutionary Computation 2015 (CEC2015) expensive optimization benchmark problems and the top performing algorithm from that competition, which is the Mean Variance Mapping Optimization (MVMO), were used to benchmark and compare the suggested hyper-heuristics. The results obtained were very encouraging when compared against the top performing expensive optimization algorithm MVMO using this comprehensive benchmark test set.