MOCSA: multiobjective optimization by conformational space annealing

TL;DR

MOCSA is a new multiobjective optimization algorithm that combines conformational space annealing with dominance and distance measures to produce diverse solutions closer to the Pareto front, outperforming NSGA2.

Contribution

It introduces a novel multiobjective optimization method integrating conformational space annealing with new fitness and update rules.

Findings

MOCSA produces solutions closer to the Pareto front.

It covers a wider range of the objective space.

Outperforms NSGA2 on benchmark problems.

Abstract

We introduce a novel multiobjective optimization algorithm based on the conformational space annealing (CSA) algorithm, MOCSA. It has three characteristic features: (a) Dominance relationship and distance between solutions in the objective space are used as the fitness measure, (b) update rules are based on the fitness as well as the distance between solutions in the decision space and (c) it uses a constrained local minimizer. We have tested MOCSA on 12 test problems, consisting of ZDT and DTLZ test suites. Benchmark results show that solutions obtained by MOCSA are closer to the Pareto front and covers a wider range of the objective space than those by the elitist non-dominated sorting genetic system (NSGA2).