Soft sensing modeling based on stacked least square-support vector machine and its application

The robust soft sensing model based on stacked least square - support vector machine is presented. Data for building nonlinear models is re-sampled using principle components analysis techniques to form a number of sets of training and test data. For each data set, a support vector machine model is developed and the developed models are then combined with different weigh factors. Model robustness is shown to be significantly improved as a direct consequence of using simple support vector machine representations. An improved algorithm for data sorting is also presented which deals with the problem of the stage division for the process. Based on the different contribution of the principal components, a new data set is formed of principal components by weighted them using contribution factors. The proposed soft sensing method has been applied to a penicillin fermentation process to estimate the product concentration, and the results of testing are listed.