Research On Modeling And Control Methods Of A Sort Of Reactor-Heat Exchanger Networks

Continuously Stirred Tank Reactor (CSTR) is a reactor with complex and nonlinear characteristics which is widely used in chemical plants. Most of reactions produce a lot of heat and if the heat is not removed in time there will be dangerous accidents, and too high or too low temperature will influence the reaction rate. In order to make sure the security of exothermic reactions and the reaction processing at a right temperature, research on modeling and control of reaction-heat exchanger networks is significant.The paper focuses further research on modeling with the theory of singular perturbation and output tracking controller with robust performance, and gets some valuable results.Firstly, the complex and multi-time scale characteristics of the CSTR reactor-heat exchanger networks are analyzed in detail. Singular perturbation variable is introduced to decompose the mechanism model into fast and slow time scales, which reduces the size and complexity of the model and becomes adapted to the analysis, control and optimization of the process network.Secondly, the theory of RGA is introduced to deal with the correlation treatment among the input and output variables. The best variable-pairing is obtained, which provides guidance for controllers design based on fast and slow time scales in the following chapters.Finally, according to the feedback linearization approach of differential geometric theory, the robust output tracking controller design for the CSTR reactor-heat exchanger networks is investigated. The paper discusses feedback linearization approach controller, PI controller and increment type predictive controller based on the network model, fast-time scale and slow-time scale models respectively, and does some simulations to analyze the reference tracking and robust characteristics.The simulation results of the paper show that the method of introducing singular perturbation variable to do model reduction not only simplifies the complexity of the primary model but also favors to design controllers respectively according to different characteristics of controlled variables which are on different scales. The PI controller applied to fast-time scale model and the increment type predictive controller applied to slow-time scale model can get the practical output to track the output reference rapidly and accurately, and guarantee the tracking error to get zero rapidly. The results are obtained when the network runs without and with the parameter uncertainties.