优化岭参数的非线性岭回归及4-CBA含量软测量

针对二甲苯氧化反应过程中影响主要副产物对羧基苯甲醛含量的因素众多且呈高度非线性的问题，提出基于优化岭参数的非线性岭回归MNRR算法，并应用于建立4-CBA含量软测量模型，获得满意的结果．MNRR采用非线性变换对原始模式特征空间进行扩张，以预测性能为指标，采用进化算法确定最佳岭参数，最终建立具有强非线性表达能力以及预测性能良好的模型，与非线性最小二乘叫归和基于广义交叉有效性逐步估计岭参数的非线性岭回归相比，MNRR模型具有更高的预测精度且克服了传统岭回归算法最佳岭参数难以确定的缺点。

Modified nonlinear ridge regression with optimal ridge parameter and its application to 4-CBA soft sensor

Considering that there exist many factors having high-nonlinear and complex effect on the concentration of the 4-carboxybenzaldehyde (4-CBA) in product, which was the most important intermediate product of p-xylene oxidation reaction, a modified nonlinear ridge regression (MNRR) based on optimal ridge parameter, was proposed to develop the 4-CBA concentration soft sensor. Satisfactory results were obtained. Firstly, MNRR applied the nonlinear transformation for initial pattern independent variables to expand pattern space. Secondly, considering that there exists correlation or multicollinearity among the variables in the expanding pattern space, the ridge regression was employed, in which evolution algorithm was used to obtain the global optimal ridge parameter according to the predicting ability of the model. Thus, the model was obtained that can describe complex nonlinear system and has good predict accuracy. The comparison results show that the MNRR model has better predict accuracy than nonlinear least square regression and nonlinear ridge regression based on generalized cross-validation of selecting the ridge parameter. In addition, MNRR overcomes the main flaw in traditional ridge regression that is difficult to obtain the global optimal ridge parameter, and thus has the robust character.