•  
  •  
 

Journal of System Simulation

Abstract

Abstract: The increase in the dimension of objective functions in optimization problems makes it challenging to balance convergence and diversity in population-based algorithms, which is a key factor leading to performance degradation of multi-objective evolutionary algorithms. To address this issue, a dynamic switching mechanism assisted by a large language model is proposed. By continuously evaluating the changes in hypervolume and convergence metrics of the solution set, the search process is adaptively divided into three phases: convergence-oriented, diversity-oriented, and large language model-suggested phases, achieving precise allocation of computational resources. Furthermore, to mitigate the high computational cost of large language models in high-dimensional scenarios, a low-cost invocation strategy is designed: the model is invoked only when performance declines for several consecutive generations, thereby guiding the population toward more efficient search. Comparative experiments with six state-of-the-art algorithms are conducted on 5 to 15-objective test sets from the DTLZ and WFG benchmark problems. The results demonstrate that the proposed method exhibits superior performance in both convergence and diversity metrics, confirming the effectiveness of large language models in assisting evolutionary algorithms to handle complex high-dimensional optimization problems.

First Page

2574

Last Page

2589

CLC

TP391.9

Recommended Citation

Ma Yichao, Wei Jingxuan. High-dimensional Multi-objective Optimization Algorithm Based on Large Language Model Assistance[J]. Journal of System Simulation, 2026, 38(9): 2574-2589.

Corresponding Author

Wei Jingxuan

DOI

10.16182/j.issn1004731x.joss.25-1017

Download

Share

COinS