RDEx-CMOP: Feasibility-Aware Indicator-Guided Differential Evolution for Fixed-Budget Constrained Multiobjective Optimization

A research team led by Sichen Tao has developed RDEx-CMOP, a new AI-powered optimization algorithm that won the constrained multiobjective track at the prestigious IEEE CEC 2025 numerical optimization competition. The algorithm is specifically designed for "fixed-budget" scenarios where computational resources are strictly limited, combining three key innovations: an ε-level feasibility schedule that gradually enforces constraints, a SPEA2-style indicator-driven fitness assignment method for balancing objectives, and a fitness-oriented mutation operator called current-to-pbest/1.

When tested on the official CEC 2025 benchmark problems using the median-target U-score framework, RDEx-CMOP outperformed all other released comparison algorithms. It achieved the highest total score and best overall average rank, with particularly strong performance in target-attainment behavior and near-zero final constraint violations across most problems. This represents a significant advancement in solving complex real-world optimization challenges where solutions must satisfy multiple competing objectives (like cost vs. performance) while adhering to strict limitations, all within limited computational time.

The algorithm's practical impact lies in its ability to handle "constrained multiobjective optimization" problems commonly found in engineering design, logistics planning, and resource allocation. Unlike traditional methods that might violate constraints or fail to converge within budget, RDEx-CMOP's feasibility-aware approach ensures solutions are both practical and high-performing. The researchers evaluated their method using the competition's released trace data, providing transparent and reproducible results that validate its effectiveness for industrial applications where evaluation budgets are tight but solution quality cannot be compromised.