Abstract: To improve the predicted accuracy and calculating efficiency of Radial Basis Function (RBF) metamodel,an Augmented Radial Basis Function (ARBF) metamodel method based on multi-strategy was proposed.Local intensive adding-points strategy,global uniform selecting-points strategy and minimum distance filtering strategy were applied to construct RBF metamodel,and RBF model was established by obtaining initial samples with Latin hypercube sampling.Then the optimal solution was obtained with Seven-spot Ladybird Optimization(SLO)algorithm.To balance the exploration and exploitation of the proposed method,the training samples were obtained by combining the local with the global strategies based on known samples.Afterwards,the minimum distance filtering strategy was used to filter the current samples so as to guide the model to predict precisely.Simulation experiments were carried out using numerical and engineering optimization examples,the results showed that ARBF was more accurate and efficient.Especially for the engineering problem,the result relatived to the theoretical optimal solution was only 0.01%,the calling number of metamodel with ARBF was decreased by 33.10% 、66.19% and 72.78% compared to other three methods.

Key words: radial basis function metamodel, local intensive adding-points strategy, global uniform sampling strategy, minimum distance filtering strategy, reducer design optimization

摘要： 为提高径向基代理模型的预测精度和计算效率,提出一种基于多策略的改进径向基代理模型方法。该方法将局部密集加点策略、全局均匀选点策略和最小距离筛选策略应用于径向基代理模型构建过程中,通过拉丁超立方抽样方法获取初始样本,并建立径向基模型,利用七星瓢虫优化算法求得最优解信息;借助已知样本信息,采用局部密集加点策略与全局均匀选点策略相结合的方式获取训练样本,以平衡所提方法的探索与开发能力;进而利用最小距离筛选策略对训练样本进行筛选,引导模型进行有效预测。利用数值和工程算例进行仿真测试,结果表明该方法不仅能满足精度要求,还能明显提高计算效率;特别是对于工程设计问题,该方法的优化结果相对理论解的误差仅为0.01%,调用模型次数相比其他3种方法减少了33.10%,66.19%,72.78%。

关键词: 径向基代理模型, 局部密集加点策略, 全局均匀选点策略, 最小距离筛选策略, 减速器设计优化