Handling Constrained Multi-Objective optimization with Objective Space Mapping to Decision Space Based on Extreme Learning Machine

Constrained multi-objective optimization is frequently encountered from the point of view of practical problem solving. The difficulty of constrained multi-objective optimization is how to offer guarantee of finding feasible optimal solutions within a specified number of iterations. To address the issue, this paper proposes an innovative optimization framework with objective space mapping to decision space for constrained multiobjective optimization and a novel multi-objective optimization algorithms are proposed based on this framework. Extreme learning machine implements prediction of decision variables from modified objective values with distance measure and adaptive penalty. This algorithm employs the framework of artificial bee colony to divide this optimization process into two phases: the employed bees and the onlooker bees. In the phase of employed bees, multi-objective strategy employs fast non-dominant sort and crowded distance to push the population toward Pareto front. In the phase of onlooker bees, multi-objective strategy employs Tchebycheff approach to enhance the population diversity. The experimental results on a series of benchmark problems suggest that our proposed algorithm is quite effective, in comparison to other state-of-the-art constrained multi-objective optimizers.