A Soft Sensor Based on Orthogonal Nonlinear Principal Component Analysis

摘要

An accurate on-line measurement of quality variables is essential for the successful monitoring and control tasks in chemical process operations. A soft sensor is developed based on orthogonal nonlinear principal component analysis, due to its ability to capture the linear and nonlinear features of the data. An orthogonal nonlinear principal component analysis network is utilized to retain a compact representation of the data optimally. Then the linear relationship between the scores and estimated variable is gained by robust linear regression based on M-estimation. The proposed method is applied on an industrial crude oil atmospheric distillation tower, and illustrated by comparison with other familiar methods. The results have shown that the proposed method gives a better performance over the conventional PCA method and neural networks method.