- Conference Article
- 10.1145/3712255.3726732
A Large Language Model-assisted Evolutionary Algorithm for Expensive Constrained Multi-objective Optimization
- Jul 14, 2025
- Siyu Chen + 3 more +3
Surrogate-assisted constrained multi-objective evolutionary algorithms are popular for solving expensive constrained multi-objective optimization problems. They build computationally cheap surrogate models to replace time-consuming function evaluations. In this way, algorithms can obtain optimal solutions within a limited number of function evaluations. However, most surrogate models are built for approximating each expensive objective function and constraint. These models may accumulate approximation errors into a nonnegligible level, and their frequent updates bring tremendous computational overhead. In this paper, a large language model (LLM)-assisted constrained multi-objective evolutionary algorithm (LECMO) is proposed. LECMO adopts the two-stage evolutionary search strategy. In the first stage, an existing surrogate-assisted evolutionary algorithm ignoring constraints is applied to drive solutions across infeasible regions. In the second stage, an LLM-assisted constrained multi-objective evolutionary algorithm is proposed to locate the constrained Pareto front. Experimental results on well-known test suites demonstrate the superiority of LECMO over compared algorithms.
Read more