An Efficient Two‐Stage Robust Multiobjective Optimization Algorithm With Application to the Carbonization Process

ABSTRACT Robust multiobjective optimization seeks solutions that achieve optimal or near‐optimal convergence across multiple objectives while remaining reliable under perturbations. However, evaluating robustness often requires multiple evaluations under small perturbations, which increases computational cost. Many heuristic methods consider robustness from the early search phase. This often leads to substantial resource consumption before the algorithm reaches global or local optima. To improve efficiency, this paper introduces an efficient two‐stage robust multiobjective evolutionary algorithm (TRMEA). The first stage—parallel exploration—performs a broad search of the decision space without perturbation evaluations, using a feature‐based coevolution mechanism to maintain exploration in suboptimal regions and accelerate convergence. The second stage—robust exploitation—focuses on refining promising regions identified earlier, where two archives with different robustness levels are employed to generate diverse, robust solutions. An adaptive switching strategy enables automatic transition between stages. Comparative experiments on multimodal benchmark problems with varying robustness levels demonstrate that TRMEA effectively finds robust optimal solutions while minimizing unnecessary computation, consistently outperforming peers across different problem dimensionalities. Furthermore, its application to optimizing carbonization parameters in carbon fiber production demonstrates TRMEA's effectiveness in balancing convergence and robustness under real‐world conditions.