Radial Basis Function-weighted Partial Least Square Regression and Its Application to Develop Dry Point Soft Sensor
基于径基函数-加权偏最小二乘回归的干点软测量

A novel approach of integrating the radical basis function (RBF) with weighted partial least squares regression (WPLSR) has proposed to develop the dry point sensor in petroleum distillation products. Many operation factors have effect on the dry point products and correlation among them. Firstly, this approach uses RBF to carry out the nonlinear transformation for the sample data. Secondly, the space distribution a of a nonlinear transformation sample data set is analyzed, and each nonlinear transformation sample is self-adaptively weighted according to its different ratios of predicting contribution for the predicting sample. Thirdly, PLSR is applied to weighted nonlinear transformation sample data set to remove the correlation and develop a model with high predicting precision. Finally, PLSR, WPLSR, RBF-PLSR and RBF-WPLSR are utilized to develop the naphtha dry point soft sensor. The comparison results show that the prediction by RBF-WPLSR is the most precise.