Control and modeling of temperature distribution in a tubular polymerization process

Temperature distribution along tubular reactors reveals the process of a polymerization reaction, so that it can be used as an indicator to monitor polymer molecular weight distribution. In this study, model of temperature distribution in a tubular polymerization reaction was developed using a B-spline neural network in conjunction with a linear recurrent neural network for the control purpose. This provides a new method for modeling distributed parameter system. Both dynamic and static neural network were applied to resolve the modeling of distribution function from a high dimensional data set. The dynamic neural network describes the time relationship of distribution function and the manipulated variables, whereas the static neural network describes the algebraic relationship of temperature distribution and position in tubular reactor. Using the error set of the expected and the measured temperature distribution as control indexes, optimal control sequence based on the distribution model can be derived. An adaptive control strategy was investigated under conditions with un-measurable noises and disturbances. An extended integral square error (EISE) control index was proposed, which introduces the real-time model error into the control strategy. This provides a feedback channel for the control, and therefore largely enhances the robustness and anti-disturbance performance of the control method. Simulation results demonstrate the effectiveness of the proposed method.