Abstract: Based on fuzzy T-S model,a fuzzy method with self-learning capability for batch process modeling and optimal control is presented.By introducing the dynamic error transfer factor associated with mean squared error and using the improved gradient descent approach,the proposed method can identify the fuzzy T-S prediction model.For constrained nonlinear optimal control of batch process,with the help of Pontryagin's minimum principle(PMP) and parallel distributed compensation algorithm(PDC),this method can transform the optimal control problem of a complex nonlinear system into the optimal problem of several local linear systems based on fuzzy T-S prediction model.Thus an effective and simple fuzzy optimal control strategy is offered.The presented method has been applied to the modeling and optimal control of a semi-continuous reactor,and simulation results show that this new method is effective and accurate.