Multiway kernel partial least squares(MKPLS) can be used for modeling and optimal control of batch processes. But the calculated optimal control policy based on the model may no longer be optimal

when applied to the actual process due to model-plant mismatches and unknown disturbances. Based on the similar idea between repetitive batch runs and iterations during numerical optimization

a batch-to-batch optimization correction strategy coupled with MKPLS model and sequential quadratic programming(SQP) was proposed. During gradient calculation

the plant data

in place of the predictions of MKPLS model

were used to correct the iterative searching direction. The proposed strategy was illustrated by simulations of bulk polymerization of styrene and fed-batch ethanol fermentation.