Dynamic Modeling And Control Based On The Temperature Distribution Of The Tubular Reactor

How to build a dynamic model of the temperature distribution in tubular polymerization reaction and try to control it are the main targets in this paper . The temperature distribution, which can be measured on line, is the most equal function of the molecular weight distribution. So in order to control the molecular weight distribution, it is very important to build this model and choose proper control strategy.Firstly, a dynamic model of temperature distribution for tubular cationic polymerization reaction is built up via hybrid neural networks, which are composed of a B-spline neural network and a recurrent neural network. So-called distributed parameter system, which must be described by one or more dimension functions, is different from the popular functions.Secondly, According to this distribution parameter system, an improved PI control algorithm is used, which connects outputs and the dynamic weights through the B-spline function. However, a control problem is reduced to a tracking problem of nonlinear dynamic weights, which separates the time and the space effectively. Then, a referred tuning method is given. Based on the model and a random situation, simulations are given to demonstrate the efficiency and stability of the proposed approach.Thirdly, prove the system of the Cationic polymerization reactor is controllable, and choose the control strategy of LQR. Therefore, a control problem of distributed parameter systems is simplified to a control problem of a linear system. Meanwhile, a new control method of the distributed parameter systems has been brought forward, which can change a nonlinear problem to a linear problem. In this paper, extend to other classical linear control method, which has universal meaning for this kind of problem.In this paper, a dynamic model of temperature distribution is built up via hybrid neural networks, which supplies a new method and example for the modeling of distributed parameter systems. Simulations are demonstrated that the precision of the model is satisfied. Then, using the improved PI strategy, which separates the time and the space effectively, the results of the control reach a desired effect. At last, using LQR to control the distribution of the temperature, which simplifies the control process, and can extend to other classical control strategy.