现有的稳健参数设计（RPD）广泛采用单个模型（如多项式、非线性模型等）来构建响应曲面，由于单个模型的选择会随样本改变而改变，使得单个模型缺乏稳健性，进而影响最优因子搭配水平的获得。本项目拟通过组合建模技术来增强模型的稳健性，并提高RPD结果的可信性。首先，从多项式回归模型中因子筛选的stepwise方法中获得启发，针对备选模型提出筛选策略。其次，由于模型仅靠单次组合往往难以达到理想回归效果，本项目将递归思想与基于预测方差的组合建模方法相结合，不断优化成员模型的权值。再次，将组合建模技术分别针对质量特性的均值/方差数据建立起均值/方差响应模型，然后将所得均值/方差响应模型代入RPD优化模型中，得到最优因子搭配水平，再进行验证性试验。最后，结合具体的工业配方设计案例来验证和应用。本项目的研究有着重要的理论意义和应用价值，研究成果可为RPD的实践提供科学指导，丰富质量设计理论。

The single models, such as polynomial and nonlinear model, are widely used in the existing robust parameter design (RPD) to construct the response surface. The choice of a single model changing with the samples leads to the lack of robustness of the model, and furthermore affect the obtaining of the recommended settings. This project use ensemble of models to control the uncertainty in modeling, and furthermore enhance the reliability of the results in RPD. Firstly, enlightened from the stepwise method in the variable selection of polynomial regression model, the model selection strategy for the candidate ones is proposed, and the corresponding algorithem is designed to increase the speed and accuracy of modeling. Secondly, considering that it is difficult for combining the candidate models just once to obtain desirable result, the recursive process and the ensemble method based on the reciprocal of prediction variance are combined to enhance the fitting/prediction accuracy. During the recursive process, the values of the weights for these single models are optimized continually. Thirdly, the ensemble technique is used to fit the mean/variance data of the quality character, then these fitting models are employed into the robust parameter optimization model, and then the recommended settings are obtained, after which, a confimatory experiment is taken around the neighborhood of the settings recommended by the ensemble. Finally, this approach of RPD based on ensemble of models is applied to industry formula design. Findings of this project have theoratical significances and practical values, whose achievements can provide scientific guide for the practice of RPD and enrich the theory of design for quality.