基于有限资源组合分配的贝叶斯优化研究  

Bayesian optimization research based on limited resource portfolio allocation

作　　者：欧阳林寒 李彤彤 陶宝平 车玉帅 OUYANG Linhan;LI Tongtong;TAO Baoping;CHE Yushuai(College of Economics and Management,Nanjing University of Aeronautics and Astronautics,Nanjing 211100,China;College of Economics and Management,Nanjing University of Science and Technology,Nanjing 210094,China)

机构地区：[1]南京航空航天大学经济与管理学院,南京211100 [2]南京理工大学经济管理学院,南京210094

出　　处：《系统工程理论与实践》2025年第1期236-247,共12页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(72072089,71931006,71872088);中央高校基本科研业务费(NE2023004)。

摘　　要：针对资源限制下随机仿真优化的预算分配问题,本文综合对冲思想与Portfolio策略,引入考虑偏差与方差的资源分配准则,提出了一种基于有限资源组合分配的两阶段贝叶斯优化方法,以期更好地解决仿真预算和优化精度的权衡问题.首先在初始样本集合上综合考虑内外部噪声对输出响应的影响,构建异方差随机Kriging模型;其次在搜索阶段基于Portfolio策略,优化采集函数得到序贯优化点集合,并依据其性能表现计算自适应选择概率;然后在分配阶段运用考虑偏差与方差的资源分配准则为采样点分配预算;最后,通过Hartmann-3维测试函数和随机需求下的库存问题,将本文所提方法与现有组合优化算法进行比较,实验结果表明,本文所提方法兼具较好的全局寻优能力与稳健性.Aiming at the budget allocation problem of stochastic simulation optimization under resource constraints,this paper combines hedging idea and portfolio strategy,introduces resource allocation criteria considering deviation and variance,and proposes a two-stage Bayesian optimization method based on finite resource portfolio allocation,in order to better solve the trade-off problem between simulation budget and optimization accuracy.First of all,a heteroscedasticity stochastic Kriging model is constructed by considering the influence of internal and external noise on the output response on the initial sample set.Secondly,in the search phase,based on the portfolio strategy,the acquisition function is optimized to obtain the sequential optimization point set,and the adaptive selection probability is calculated according to its performance.Then,in the allocation stage,the resource allocation criterion considering deviation and variance is used to allocate the budget for the sampling points.In the end,the Hartmann-3 dimensional test function and the inventory problem under random demand are used to compare the proposed method with the existing portfolio optimization algorithm.The experimental results show that the proposed method has good global optimization ability and robustness.

关 键 词：随机Kriging Portfolio策略 资源分配准则 贝叶斯优化 

分 类 号：TB114.3[理学—概率论与数理统计]