Model Management Agent for Expensive Constrained Multi-Objective Optimization

Surrogate-assisted evolutionary algorithms (SAEAs) have been extensively studied for their effectiveness in solving expensive optimization problems in various domains. A critical component of the SAEA framework is model management, which is used to query new solutions and update the surro-gate model. This paper introduces a novel model management agent based on reinforcement learning. The developed SAEA, MMA-SAEA, addresses expensive constrained multi-objective optimization problems (ECMOPs). MMA-SAEA conceptualizes the SAEA as an agent that utilizes two distinct model man-agement strategies to achieve global optimization. The first strategy emphasizes objective optimization and constraint sat-isfaction, while the second strategy balances exploration and exploitation. The performance of MMA-SAEA was evaluated using the benchmark test suites and compared with other state- of-the-art methods. Results indicate that MMA-SAEA delivers superior performance for ECMOPs. In addition, MMA-SAEA offers improved performance on real-world operational indices in beneficiation processes.