基于KCCA虚假邻点判别的非线性变量选择

特征变量选择技术是非线性系统建模过程中降低信息冗余和提高精度的有效方法。提出一种结合核典型相关法(kernel canonical correlation analysis,KCCA)与虚假最近邻法的变量选择法。首先引入核方法,将非线性原始数据映射到线性空间,再采用典型相关法有效合理地消除因子之间的多重共线性,受混沌相空间虚假最近邻点法的启示,通过计算原始数据在KCCA子空间中投影的距离,判断其对主导变量的解释能力,由此进行变量的选择。该方法用氢氰酸生产工艺工程中的非线性模型验证,并与全参数模型进行比较,结果显示该方法有良好的变量选择能力。因此,该研究为非线性系统建模的变量选择方法提供了一种新方法。

Variable selection for nonlinear modeling based on false nearest neighbors in KCCA subspace

Selection of secondary variables is an effective way to reduce redundant information and improve efficiency in nonlinear system modeling.A novel method based on kernel canonical correlation analysis(KCCA) and false nearest neighbor(FNN) is proposed for selecting the most suitable secondary process variables used as nonlinear modeling inputs.In the proposed approach,the KCCA can be employed to reasonably eliminate the existing multi-collinearity among the factors.In the new KCCA feature subspace,inspired by FNN,the interpretation capability for primary variable is estimated through calculating the variable mapping distance in the KCCA space and then the secondary variables are selected.The proposed method was used in the nonlinear model verification of hydrogen cyanide production process and compared with the full parametric model.Results show that the method is effective and suitable for variable selection;this study provides a new method for the variable selection of nonlinear system modeling.