基于近似贝叶斯计算的包装件模型选择和参数估计

Packaging model selection and parametric estimation based on approximate Bayesian calculation

准确构建包装件模型是进行运输包装安全评价的基础,构建包装件模型包括模型类型选择和模型参数估计。考虑到包装件模型的不确定性和误差,需在贝叶斯推断框架下构建包装件的参数不确定模型,由于在计算似然函数中存在着一系列困难,利用传统贝叶斯推断估计模型参数存在着计算效率低,计算误差大的缺点;结合序贯蒙特卡洛和重要性采样,采用近似贝叶斯计算替代传统的贝叶斯推断,该方法可避免似然函数的计算,随着阈值的减小,在多种备选模型中优选出最佳模型,同时模型参数收敛至参数真值附近。对包装件进行随机振动试验,对试验数据进行分析,结果表明,Bouc-Wen模型(n=2)是最佳包装件模型,该模型可准确预测包装件的振动响应。

Correctly constructing a packaging model is a foundation for evaluating transport packaging safety.Building a packaging model includes model type selection and model parametric estimation.Considering the uncertainty and error of packaging model,it is necessary to build parametric uncertainty model for packages under Bayesian inference framework.Because there are a series of difficulties in calculating Likelihood function,using traditional Bayesian inference to estimate model parameters has disadvantages of low calculation efficiency and large calculation error.Here,combining sequential Monte Carlo and importance sampling,approximate Bayesian calculation was proposed to replace traditional Bayesian inference.This method could avoid calculation of Likelihood function.With decrease in threshold,this method could select the optimal model from a variety of alternative models,and the optimal model’s parameters could converge to near true values of parameters.Random vibration experiments were conducted for packages,and the experimental data were analyzed.The results showed that Bouc-Wen model(n=2)is the optimal packaging model,it can correctly predict vibration responses of packages.