A preference-based indicator selection hyper-heuristic for optimization problems

Heuristics have been effective in solving computationally difficult optimization issues, but because they are often created for certain problem domains, they perform poorly when the challenges are significantly altered. The currently available techniques are either designed to address single- or multi-objective optimization issues solely, or they perform poorly with the same parameters. The multi-domain approach known as hyper-heuristics (HHs) can be used to solve optimization issues with minor variations. Motivated by the notion of utilizing the benefits of low-level heuristics (LLHs) in order to obtain well-distributed and convergent optimum solutions along with taking into account the shortcomings of the work completed in many-objective HHs. For many-objective optimization problems, this paper develops a high-level selection approach that employs indicators by preference and offers a unique selection hyper-heuristic called Preference-based Indicator Selection Hyper-heuristic (PBI-HH). In order to establish fairness between exploration and exploitation, the method makes use of a randomization mechanism and a greedy strategy to address a significant problem faced by HHs. Three well-known many-objective evolutionary algorithms are combined in the unique technique that is being proposed. The efficacy of the proposed strategy is assessed by contrasting it with cutting-edge HHs. PBI-HH performs better or equal to the state-of-the-art HHs on 155 out of 160 cases employing the HV indicator and has the optimal μ norm mean values across all datasets.