采用自适应最小化置信下限和SMOTE算法的动态代理模型  

Dynamic surrogate model based on adaptive minimal lower confidence bound and SMOTE algorithm

作　　者：戚林辉 潘伟锋 罗伟林[1] QI Linhui;PAN Weifeng;LUO Weilin(School of Mechanical Engineering and Automation,Fuzhou University,Fuzhou,Fujian 350108,China)

机构地区：[1]福州大学机械工程及自动化学院,福建福州350108

出　　处：《福州大学学报（自然科学版）》2023年第6期811-818,共8页Journal of Fuzhou University(Natural Science Edition)

基　　金：福州海洋研究院科技资助项目(2022F13)。

摘　　要：利用最优拉丁超立方试验进行初始采样,建立基于自适应最小化置信下限和SMOTE算法的动态径向基函数代理模型.将自适应平衡常数引入到最小化置信下限准则中,通过多岛遗传算法对置信下限进行寻优.根据代理模型精度,在最优解处运用SMOTE算法动态地新增样本点,进而更新代理模型,直至收敛.经过数学算例测试后,将该优化策略应用于深潜器耐压舱的优化中,与其他动态代理模型相比,该策略的优化效率和精度显著提高.Optimal Latin supercubic design of experiment is employed for initial sampling.The radial basis function is used to construct a dynamic surrogate model based on adaptive minimal lower confi⁃dence bound(LCB)and synthetic minority oversampling technique(SMOTE).In the LCB algorithm,an adaptive balance parameter is designed.The optimal lower confidence bound is obtained by multi⁃island genetic algorithm.SMOTE algorithm is applied to improve the accuracy of the surrogate model by adding new samples around obtained optimal solutions to update until it converges.A pressure shell design of underwater vehicle is taken as the verification model after the proposed optimization strategy is tested by several mathematical testing examples.Results show the better accuracy and efficiency of optimization when compared with other optimization strategies based on dynamic surrogate model.

关 键 词：动态代理模型 径向基函数 最小化置信下限 SMOTE算法 深潜器 耐压舱 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]