中文摘要：

针对非线性相关的信号—响应系统的稳健性参数设计,提出一种基于支持向量回归机和遗传算法的实现方法。首先,利用过程方差描述信号—响应之间非线性关系的波动,建立可控因子与过程方差之间的支持向量回归模型,利用遗传算法全局性寻优得到使非线性关系最稳定的可控因子水平;其次,将噪声因子水平的变化看作重复,建立可控因子、信号因子和响应变量之间的支持向量回归模型,进而预测出最优可控因子水平下信号—响应的样本集;最后,根据此样本集拟合出信号因子与响应变量之间的具体作用关系模型。理论与实证研究表明,与现有信噪比分析和响应建模方法相比,所提方法能够较为真实地反映各类因子与响应变量之间的复杂作用关系,得到波动性更小的稳健性优化解。

英文摘要：

For the Robust Parameter Design （RPD） of nonlinear correlation signal-response system, an approach based on Support Vector Regression （SVR） and Genetic Algorithm （GA） was proposed. The fluctuation of nonlin- ear correlation in signal-response system was represented by process variation, and a SVR model for controllable fac- tors and process variation was set up. The global optimal level of controllable factors with the steadiest nonlinear correlation of the system were obtained by using GA. The SVR model between controllable factors, signal factor and response variable was constructed by regarding the noised factors as repeated runs, Subsequently, the sample set of signal-response under the optimal controllable factors was predicted by using SVR model. The empirical model of the signal factor and response variable was simulated based on the proposed sample set. The theoretical analysis and empirical research showed that the proposed approach could objectively reflect the complex influential relation- ship among all kinds of factors and response variable comparing with signal-to-noise ratio analysis approach and re- sponse modeling approach, therefore the robust optimization result with smaller process variation was obtained.