Neural Model-Based Advanced Control of Chylla-Haase Reactor

摘要: The objective of this thesis is to develop advanced control method and design system for the polymerization reactor (Chylla-Haase) maintain high accurate temperature. first stage research start with development mathematical model process. sub-models monomer concentration, rate, temperature jacket outlet/inlet are developed implemented in Matlab/Simulink. Four conventional methods were applied reactor: a Proportional –Integral-Derivative (PID), Cascade (CCs), Linear-Quadratic-Regulator (LQR), Linear predictive (LMPC). simulation results show that PID controller unable perform satisfactorily due change physical properties unless constant re-tuning takes place. Also, Control most common used such processes cannot guarantee robust performance under varying disturbance uncertainty. In addition, LQR linear MPC lead better compared previous two methods. But it still an assumption linearized plant. Three neural network based schemes also proposed thesis: radial basis function RBF inverse feedforward-feedback scheme, multi-layer perception (MLP) control. major these within its tolerance range disturbances Satisfactory terms effective regulation robustness have been achieved. In predict Then, estimate valve position reactor, manipulated variable. This can identify the controlled identifier. A feedback regulate actual by compensating output. Simulation has strong adaptability, satisfactory performance. These achieved much improved schemes. The main contribution lies following aspects. theory realised Chylla-Haase reactor. Two adaptive models including MLP multiple-step-ahead values Their modelling ability fixed parameters proven be better. trained recursive Least Squares (RLS) algorithm parameter uncertainty nonlinear dynamics strategy on achieve set-point tracking output against disturbances. result shows gives reliable presence some keeps tight around specified reaction temperature. Moreover, reduce batch time order shorten period. considered as prediction purpose which minimize cost determine optimal sequence moves. variation keeping without harming quality