基于Lasso-Bayesian改进的Kriging代理模型优化方法及其应用

Improved Kriging Surrogate Model Optimization Method Based on Lasso-Bayesian and Its Application

为提高Kriging模型的性能并构建高精度代理模型,基于最小绝对收缩和选择算子(Lasso)与Bayesian算法对Kriging方法进行改进,实现了对Kriging模型的超参数调优,提出Lasso-Bayesian-Kriging代理模型的构建方法。采用Lasso正则化对模型输入进行特征选择,以降低模型复杂度,提高模型的泛化能力。使用Bayesian算法对Kriging中的相关参数、相关函数以及回归函数进行调优,得到高精度的Kriging代理模型。针对某车间加工矿用钻杆过程中的搬运桁架的实际工程问题,采用4种不同方法对桁架静力学分析进行代理建模,以桁架质量和变形量为代理对象,通过k折交叉验证,结果表明,Lasso-Bayesian-Kriging方法构建的代理模型精度最高,其交叉验证的平均决定系数R2分别为0.999、0.962。将优化算法与Lasso-Bayesian-Kriging模型相结合对桁架进行迭代优化,结果表明优化后的桁架在满足刚度的前提下实现了轻量化。

In order to improve the performance of Kriging model and construct a high-precision surrogate model,the Kriging method was improved based on least absolute shrinkage and selection operator(Lasso)and Bayesian algorithm,and the hyper-parameter tuning of Kriging model was realized.The construction method of Lasso-Bayesian-Kriging surrogate model was proposed.Lasso regularization was used for feature selection of model input to reduce the complexity of the model and improve the generalization ability of the model.The Bayesian algorithm was used to optimize the relevant parameters,correlation functions and regression functions in Kriging to obtain a high-precision Kriging surrogate model.Aiming at the practical engineering problem of a mining drill pipe handling truss in a workshop,four different methods were used to model the static analysis of the truss.Taking the truss mass and deformation as the surrogate objects,the k-fold cross-validation was carried out.The results show that the Lasso-Bayesian-Kriging method has the highest accuracy of the surrogate model,and the average determination coefficient R2 of the cross-validation is 0.999 and 0.962,respectively.The optimization algorithm was combined with the Lasso-Bayesian-Kriging model to iteratively optimize the truss.The results show that the optimized truss achieves lightweight under the premise of satisfying the stiffness.