Robust Multi-Objective Particle Swarm Optimization Algorithm Based on Hybrid Metric

Abstract

In this paper, a robust multi-objective particle swarm optimization algorithm, based on hybrid metric (RMOPSO-HM), is proposed to depress the adverse effect of uncertainties on optimization performance. First, an adaptive sampling strategy is developed to dynamically adjust the sampling quantity of decision variables by evaluating their robust relevance. Unnecessary computation is reduced and sampling efficiency is improved by this strategy. Second, a sampling-based robustness evaluation metric is designed by incorporating both the preference robustness (PR) and the dominance robustness (DR). The hybrid metric facilitates a more comprehensive assessment of the robustness and avoids missing robust optimal solutions. Third, the robustness is embedded into the evolutionary mechanism by a weight preference operator and a new objective function is defined, which can tradeoff the optimality and the robustness. Finally, the hybrid metric and the improved evolutionary mechanism are applied to a parametric adaptive particle swarm optimization algorithm to guide the population evolution. Experimental results illustrate that the designed robust optimization algorithm maintains a superior optimization performance despite the impact of uncertainties.