A weighted uncertainty measure-based surrogate-assisted evolutionary algorithm for expensive many-objective optimization and application to flue gas desulfurization processes

Typically a small number of function evaluations are accessible in many real-world many-objective optimization problems (MaOPs), and these are costly to compute and execute. In this article, we propose a weighted uncertainty measure-based surrogate-assisted evolutionary algorithm (WUM-SAEA) for expensive many-objective optimization. The fundamental idea is that the suggested algorithm provides a weighted uncertainty measure technique (WUM) to estimate each objective function by using the uncertainty information provided by several Kriging models. Additionally, an adaptive reference points-based infill criterion is proposed to choose the most suitable sampling strategy for re-evaluation using expensive objective functions by identifying the convergence and diversity demands. The dataset for model training is then enhanced using the chosen data. The effectiveness of WUM-SAEA is assessed by contrasting it with a number of cutting-edge surrogate-assisted evolutionary algorithms on two sets of many-objective test problems and an optimization problem for industrial processes. Its superiority over the compared algorithms has been shown by the experimental results. Furthermore, the WUM-SAEA algorithm’s application in the flue gas desulfurization process shows its usefulness in solving real-world issues.