| 张晶*,裴东兴** ***,马瑾*,沈大伟** ***.基于动态分布计算资源的昂贵多目标优化算法[J].高技术通讯(中文),2025,35(8):861~867 |
| 基于动态分布计算资源的昂贵多目标优化算法 |
| An expensive multi-objective optimization algorithm based on dynamically distributed computing resources |
| |
| DOI：10. 3772 / j. issn. 1002-0470. 2025. 08. 005 |
| 中文关键词: 进化算法； 昂贵多目标优化问题； 代理模型； 填充准则； 不确定度 |
| 英文关键词: evolutionary algorithm, expensive many objective optimization problem, surrogate model, infill criterion, uncertainty |
| 基金项目: |
| 作者 | 单位 | | 张晶* | (*山西财贸职业技术学院物联网技术系太原 030031)
(**中北大学电子测试技术国家重点实验室太原 030051)
(***中北大学仪器科学与动态测试教育部重点实验室太原 030051) | | 裴东兴** *** | | | 马瑾* | | | 沈大伟** *** | |
|
| 摘要点击次数: 85 |
| 全文下载次数: 71 |
| 中文摘要: |
| 代理模型辅助的多目标优化算法广泛用于求解评价昂贵的多目标优化问题，其中，采用样本更新模型是提高算法性能的必要过程。然而，传统方法未对模型的状态进行评估而同时更新所有模型，浪费了大量的计算资源。针对该问题，本文提出基于动态分布计算资源的昂贵多目标优化算法，该算法提出了自适应选择模型更新策略。具体地，依据模型对当前种群估值的不确定度来判断模型的性能，当种群中解不确定度的中值大于均值时，该目标函数模型被选择进行更新；当种群中的解不确定度的中值小于均值时，该模型不被更新。为了验证该策略的有效性，将该策略用于代理模型辅助的自适应贝叶斯优化算法(an adaptive Bayesian approach to surrogate-assisted evolutionary algorithm，ABSAEA)和代理模型辅助的参考向量引导的进化算法(surrogate-assisted reference vector guided evolutionary algorithm，KRVEA)中，并且在DTLZ函数上进行实验。实验结果表明，该算法可以显著降低昂贵多目标优化算法的计算复杂度。 |
| 英文摘要: |
| Surrogate assisted multi-objective evolutionary algorithms are widely used to solve expensive multi-objective optimization problems, where updating models is necessary to improve the performance. However, the traditional algorithms update all models simultaneously without evaluating the state of the models, so, a large amount of computational resources will be wasted. This paper proposes an expensive multi-objective optimization algorithm based on dynamically distributed computing resources for this challenge, in which an adaptive selection of model updating strategies is proposed. Specifically, the performance of surrogate is evaluated based on the uncertainty of the model’s evaluation of the current population. When the median uncertainty of individuals in the population is greater than the mean, the model is selected for updating. Conversely, when the median uncertainty is less than the mean, the model is not updated. To validate the effectiveness of this strategy, it is applied in the an adaptive Bayesian approach to surrogate-assisted evolutionary algorithm (ABSAEA) and surrogate-assisted reference vector guided evolutionary algorithm (KRVEA), and experiments are conducted on the DTLZ function. The experimental results demonstrate that the proposed algorithm can significantly reduce the computational complexity of expensive multi-objective optimization algorithms. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
