Robust Multi-objective Optimization with Less Computational Effort

Abstract

Aiming at efficiently and effectively solving robust multi-objective optimization problems and enhancing the uncertainty handling ability of evolutionary algorithms, we design a robust multi-objective optimization framework by fully exploiting the population-based nature of evolutionary algorithms. The goal of our research includes 1) obtaining a group of robust optimal solutions which can balance the optimality and robustness in the maximum degree, as well as 2) reducing the computational resource consumed in the evolutionary process. First, in multi-objective optimization, besides the obtained global Pareto optimal solutions, all visited solutions in the evolutionary process are organized in both decision space and objective space. Then, based on the exploration of multi-objective optimization, the robust optimization effectively searches for the robust region. Afterward, a part of regions are picked up among all robust regions so that these selected regions together construct a robust optimal front which achieves the best convergence and diversity performance. Under this framework, we provide two approaches, one involves sampling process while the other does not. Experimental results on a group of benchmark functions demonstrate the superiority of both design in terms of both solutions' quality under the disturbance and computational efficiency in solving robust multi-objective optimization problems.