A new online modeling method based on adaptive local learning and least squares support vector regression was proposed for nonlinear multi-input multi-output processes.Both distance measure and angle measure were used to evaluate the similarity between data

thus a more comprehensive and relevant set was constructed.In the present method

the model parameters were optimized online by using the fast leave-one-out criterion.In addition

an online adaptive model selection strategy for modeling of fermentation processes was developed.Simulations on a fed-batch streptokinase fermentation process showed that the active biomass and streptokinase concentrations could both be predicted simultaneously just from the second batch.The results also showed that the proposed method was more accurate and adaptive compared with the alternative modeling methods.