針對某丙酮精制過程，提出采用FA 與SVR 相結合的方法建立丙酮產品質量的軟測量模型。采用因子分析（FA）方法提取輔助變量的特征信息，并消除各變量之間的相關性，然后利用支持向量回歸（SVR）建立丙酮產品質量指標的軟測量模型。在實際生產過程數據上進行了仿真實驗，并與傳統的穩健回歸分析及神經網絡等方法進行了比較，結果表明本方法具有良好的預測效果。
關鍵詞：軟測量；支持向量回歸；因子分析；丙酮精制

FA-SVR based Soft-Sensor Model of Product Quality in Acetone Refining Duan Ying, Fan Quanyi, Xiong Zhihua (Department of Automation, Tsinghua University, Beijing 100084) Abstract: A soft-sensor model of product quality in acetone refining process is proposed by combining factor analysis (FA) and support vector regression (SVR). During the data processing, FA is applied to extract the characteristics of secondary variables and eliminate the collinearity among these variables. The extracted principal factors are then used as input parameters of soft sensor model, and the model is built by using SVR to predict the acetone quality. The experiment results on the real industrial data show that the method of FA-SVR is more effective than robust regression and neural network methods.
Keywords: Soft sensor; Support vector regression (SVR); Factor Analysis (FA) ; Acetone refining