The Efficient Global Optimization(EGO)algorithm has been widely used in the numerical design optimization of engineering systems.However,the need for an uncertainty estimator limits the selection of a surrogate model....The Efficient Global Optimization(EGO)algorithm has been widely used in the numerical design optimization of engineering systems.However,the need for an uncertainty estimator limits the selection of a surrogate model.In this paper,a Sequential Ensemble Optimization(SEO)algorithm based on the ensemble model is proposed.In the proposed algorithm,there is no limitation on the selection of an individual surrogate model.Specifically,the SEO is built based on the EGO by extending the EGO algorithm so that it can be used in combination with the ensemble model.Also,a new uncertainty estimator for any surrogate model named the General Uncertainty Estimator(GUE)is proposed.The performance of the proposed SEO algorithm is verified by the simulations using ten well-known mathematical functions with varying dimensions.The results show that the proposed SEO algorithm performs better than the traditional EGO algorithm in terms of both the final optimization results and the convergence rate.Further,the proposed algorithm is applied to the global optimization control for turbo-fan engine acceleration schedule design.展开更多
非刚性点集配准是计算机视觉和模式识别领域的基础研究问题,现今的非刚性点集配准算法在存在大量离群点、噪声、点集对应关系缺失、旋转和形变情况下,不能非常准确地评估出两个点集间的对应关系.本文通过交替执行点集对应关系评估和空...非刚性点集配准是计算机视觉和模式识别领域的基础研究问题,现今的非刚性点集配准算法在存在大量离群点、噪声、点集对应关系缺失、旋转和形变情况下,不能非常准确地评估出两个点集间的对应关系.本文通过交替执行点集对应关系评估和空间转换更新两个步骤来逐步恢复点集间一一对应关系.在对应关系评估步骤,首先本文基于有限重尾学生t分布隐变量混合模型(student-t distribution Latent Mixture Model,简称TLMM)构造变分贝叶斯层次概率模型(Variational Bayes Hierarchical Probability Model,简称VBHPM)并将其分为对应关系评估组件和离群点聚合组件,分别用来评估点集间对应关系和聚合离群点,同时使用贝叶斯线性回归方法来抵抗噪声的干扰.其次本文加入Dirichlet先验分布来动态调节模型的混合比例,为对应关系缺失的点分配较小的混合比例以保持点集结构的稳定性.在空间转换更新步骤,本文基于变分贝叶斯(Variational Bayes,简称VB)框架来迭代更新模型参数,并提出树状平均场因式分解方法来维持模型参数间的依赖关系,以获得更紧致的变分下界.此外,本文提出自适应全局-局部约束策略来维持点集间结构的稳定性,抵抗形变和旋转影响的同时实现从局部到全局的约束过程.最后,本文采用了双阶段先验退火方案,在退火过程中使用Gamma先验分布来动态调节精度,实现由粗到精的配准过程.在实验部分,本文不仅测试了VBHPM的性能,而且展示了点集和图像配准的结果,并与当前流行的13种算法进行了比较,VBHPM皆能展现较准确的配准结果和较高的精度.展开更多
基金the financial support of the National Natural Science Foundation of China(Nos.52076180,51876176 and 51906204)National Science and Technology Major Project,China(No.2017-I0001-0001)。
文摘The Efficient Global Optimization(EGO)algorithm has been widely used in the numerical design optimization of engineering systems.However,the need for an uncertainty estimator limits the selection of a surrogate model.In this paper,a Sequential Ensemble Optimization(SEO)algorithm based on the ensemble model is proposed.In the proposed algorithm,there is no limitation on the selection of an individual surrogate model.Specifically,the SEO is built based on the EGO by extending the EGO algorithm so that it can be used in combination with the ensemble model.Also,a new uncertainty estimator for any surrogate model named the General Uncertainty Estimator(GUE)is proposed.The performance of the proposed SEO algorithm is verified by the simulations using ten well-known mathematical functions with varying dimensions.The results show that the proposed SEO algorithm performs better than the traditional EGO algorithm in terms of both the final optimization results and the convergence rate.Further,the proposed algorithm is applied to the global optimization control for turbo-fan engine acceleration schedule design.
文摘非刚性点集配准是计算机视觉和模式识别领域的基础研究问题,现今的非刚性点集配准算法在存在大量离群点、噪声、点集对应关系缺失、旋转和形变情况下,不能非常准确地评估出两个点集间的对应关系.本文通过交替执行点集对应关系评估和空间转换更新两个步骤来逐步恢复点集间一一对应关系.在对应关系评估步骤,首先本文基于有限重尾学生t分布隐变量混合模型(student-t distribution Latent Mixture Model,简称TLMM)构造变分贝叶斯层次概率模型(Variational Bayes Hierarchical Probability Model,简称VBHPM)并将其分为对应关系评估组件和离群点聚合组件,分别用来评估点集间对应关系和聚合离群点,同时使用贝叶斯线性回归方法来抵抗噪声的干扰.其次本文加入Dirichlet先验分布来动态调节模型的混合比例,为对应关系缺失的点分配较小的混合比例以保持点集结构的稳定性.在空间转换更新步骤,本文基于变分贝叶斯(Variational Bayes,简称VB)框架来迭代更新模型参数,并提出树状平均场因式分解方法来维持模型参数间的依赖关系,以获得更紧致的变分下界.此外,本文提出自适应全局-局部约束策略来维持点集间结构的稳定性,抵抗形变和旋转影响的同时实现从局部到全局的约束过程.最后,本文采用了双阶段先验退火方案,在退火过程中使用Gamma先验分布来动态调节精度,实现由粗到精的配准过程.在实验部分,本文不仅测试了VBHPM的性能,而且展示了点集和图像配准的结果,并与当前流行的13种算法进行了比较,VBHPM皆能展现较准确的配准结果和较高的精度.
