Optimized ILC For Polymerization Reaction Process

Due to nonlinear, complexity, variability and uncertainty that batch polymerization processes involve, the control for polymerization processes offers attractive and challenging problems for control community. The application of traditional control methods can't meet the demand of perfect control performance, so the new advanced control method should be adopted.Iterative learning control (ILC) is a novel control arithmetic, which doesn't depend on the precise model. It can generate input signal and reduce error through repeating learning so that the output of the system can approximate to the expectation. The research of ILC is more significant for solving control precision problem with highly nonlinear, complexity and difficulty in modeling.The integrated control method based on analyzing the characteristics of polymerization reaction process is proposed in this paper, which combines ILC with feedback control, and the parameters of iterative controller are optimized by. neural network (NN), so the iterative control law can be optimized accordingly. Furthermore, feedback control is introduced to regulate the system tracking errors and feed forward compensation is implemented by ILC in this method. Fixed learning gain will make the learning speed slower and the iterative times more. Neural network possesses the computational ability and the satisfactory capability of approximating any nonlinear mapping that can reduce the huge computing burden and fasten the convergent rate of right value largely compared with least square method. So the algorithm that the parameters of controller are optimized by NN is adopted in this paper. This method not only improves disturbance and robustness of the controlled system, but also fully exerts the intellectualized virtue of ILC without precise model.The optimized control method is applied to the temperature control of ABS resin polymerization reactor process. The simulation result indicates, compared ILC process with the fix learning gain, the method proposed in this paper is better with less iterative, double-quicker convergence and more high tracking precision, and it assures that the temperature trajectory can track the anticipant trajectory precisely with less iterative times.