Research On Modeling And Predictive Control Approach Of Simulated Moving Bed Separation Process

Recently, the Simulated Moving Bed (SMB) technology is acquiring greater and greater interest, in particular for the area of oil industry, fine chemistry and pharmaceutical industry. In the last decade, the research focuses on modeling and optimization of SMB, but how to implement advanced process control on this process is receiving more and more attention.Simulated moving bed is a complex process with a lot of effecting factors, so it's difficult to get high-performance separation using classical controllers. The predictive control has lots of advantages in dealing with multivariable processes, so it is expected to have good performance for the SMB process. After building the model of SMB, a predictive model is obtained through subspace identification, and is used for the internal model of the predictive controller. Considering the problem of time-varying parameter caused by the aging of the absorbent, an adaptive predictive controller based on the recursive subspace identification is designed. The main works and contributions of this thesis are as followings:(1) The background, concept and industrial applications of SMB are introduced. A survey of modeling and control methods of SMB is given.(2) The Distributed Plug-Flow (DPF) model is selected as the process model from the existing models, and applied in two separation processes: binaphthol enantiomer separation and p-xylene separation. The relations between the adsorbent parameters and the separation performance in PX adsorption process are analyzed, and the effects of parameters on the separation performance are studied.(3) Considering the mass transfer resistance is important, a predictive controller based on subspace identification is proposed. According to the separation process of binaphthol enantiomers, the control performances for several cases are tested.(4) An adaptive predictive controller is designed for the time varying feature of SMB. In this control strategy, the model is updated with the new input/output data of the process, and a matching function is used as switching principle. It is shown that the proposed controller demonstrates satisfactory control performance, even though the parameters vary with time.