Soft sensor modeling based on multiple support vector machines

Soft sensor which solved the difficult problem of measuring the un-measurable output variables has been widely used in industrial process control. The core problem of soft sensor is to construct an appropriate mathematic model. Support vector machine (SVM) which has high generalization is adopted to establish soft sensor models. Based on the idea that the accuracy of model could be significantly improved by combining several sub-models, a multiple support vector machine (MSVM) modeling approach was proposed to build the soft sensor model. In this method, subtractive clustering was adopted to divide the input space into several sub-spaces, and sub-models were built by Least Square SVM (LS SVM) in every sub-space. In order to minimize the severe correlation among sub-models, to improve the accuracy and robustness of the model, the sub-models were combined by principal components regression (PCR). The software sensor model accuracy is perfectly improved. The simulation results demonstrate the efficiency of the method.