An open-source parallel algorithm of Bayesian-based global search with Hooke–Jeeves local refinement for multi-objective optimization problems

Abstract

Contemporary engineering and scientific problems often involve computationally intensive optimization tasks. This paper proposes a parallel version of the hybrid algorithm of the previously proposed Bayesian-based global search with Hooke–Jeeves local refinement for multi-objective optimization problems. The Bayesian-based hybrid algorithm has been complemented with multi-process data exchange using Open MPI to obtain a scalable parallel application. Each parallel process executes the Bayesian-based hybrid algorithm, and at the end, the Pareto optimal solutions of each process are merged into an aggregated set of Pareto optimal solutions. The master-slave pattern was used to parallelize the computations, where slave processes execute optimization algorithm and then send the obtained Pareto optimal solutions to the master process, which, in turn, also executes optimization algorithm and merges the Pareto optimal solutions of all processes. The developed parallel algorithm was tested under the same conditions previously used for testing other Bayesian algorithms to enable comparison of performance. Finally, the proposed parallel algorithm was published on the GitHub developer platform for code sharing.