A novel update algorithm of least squares support vector machine for industrial process modeling

An update algorithm of least squares support vector machine (LSSVM) is proposed to tackle the time-varying characteristics of the real industrial process. The process variations are concluded to two categories, and accordingly the samples adding and samples replacement are proposed to update the initial LSSVM model incrementally. Then the LSSVM model with proposed updating measures is applied in the prediction of SO₂ concentration in the sulfur recovery unit (SRU) process. The results reveal that the prediction accuracy of the model with update maintains high in spite of the process characteristics varying.