Soft Sensor Modeling Based on KPCA and Least Square SVM
XU Ye, DU Wen-li, QIAN Feng (State-Key Laboratory of Chemical Engineering, East China University of Science and Technology, Shanghai 200237, China)
Soft sensor is necessary for industrial process control and analysis, and the core problem is how to construct appropriate model having fast convergence speed and good generalization performance. A kind of soft sensor method was proposed based on kernel principle component analysis (KPCA) and least square support vector machine (LSSVM). KPCA was applied to choose the nonlinear principal component of the model input data space, and LSSVM was applied to proceed regression modelling, which could not only reduce the complexity of calculation but could improve the generalization ability. The proposed KPCA-LSSVM was applied to predict the granularity of PTA. Simulation indicates that this method features high learning speed, good approximation and good generalization ability compared with SVM and PCA-SVM, and is proved to be an efficient modeling method.