A soft sensor modeling approach using support vector machines

Artificial neural networks (ANNs) such as radial basis function networks (RBF NNs) have been successfully used in soft sensor modeling. However, the generalization ability of conventional ANNs is not very well. For this reason, we present a novel soft sensor modeling approach based on support vector machines (SVMs). Since standard SVMs have the limitation of speed and size in training large data set, we hereby propose least squares support vector machines (LS_SVMs) and apply it to soft sensor modeling. Systematic analysis is performed and indicates that the proposed method provides satisfactory performance with excellent approximation and generalization property. Monte Carlo simulations show that our soft sensor modeling approach achieves superior performance to the conventional method based on RBF NNs.