Aiming at the problem that a matrix feed rate controller is designed based on least square（LS） regression algorithm and improved just-in-time learning（JITL）strategy to solve the problems of time-varying and non-linear in microbial fermentation.first

the data are collected by fermentation instruments to form a historical database

and then the weighted fuzzy C mean（WFCM）clustering algorithm is used to classify the data

so that the local prediction model for bacterial concentration of marine alkaline protease based on JITL-LS-SVM can be established quickly when the query value arrived.At the same time

in order to avoid solving nonlinear problems in predictive control

Taylor linearization method is used to predict and control the bacterial concentration of marine alkaline protease by generalized prediction control（GPC） algorithm.Experimental simulation shows that the JITL-LS-SVM model based on improved real-time learning can achieve better control effect

and the glucose feed rate response is rapid

which is suitable for the control of matrix flow and feeding during the fermentation of marine alkaline protease.