Dynamic soft sensor modeling based on multiple least Squares Support Vector Machines

In order to estimate the quality parameters in industrial processes, taking actual dynamic transition characteristics into account, a dynamic soft sensor modeling approach based on multiple least squares support vector machines (LSSVM) is presented. The input variables acquired in transition period are segmented according to their acquisition time point. Then LSSVMs are employed to map the relations of different time series of input variables to output variables. Moreover, a synthetical LSSVM is delivered to embody the dynamic characteristic of all the sub-networks to the model. A simulation example is put forward at last. The result shows that proposed method has simple dynamic structure and clear physical meanings, which features in better estimation precision and robustness than static soft sensors.