Research On Model Predictive Control Algorithms And Applications In Chemical Processes

With the increasing competiton in global market, decrease of non-renewable resources and high pressure in environment, process control is required to be more efficient and profitable. Model preditive control technique, due to its advantages of control performance, robustness and constraints handling, has become an important issue in the control engineering fields. Some problems of predictive control are researched in this dissertation, and the main research works are as follows:(1) A modified DMC algorithm that uses an adaptive disturbances model is proposed. The dynamics of unmeasured distubances are estimated by an ARMA process. An on-line recursive method is adopted to estimate the coefficients of ARMA process considering the time-varing feature of disturbances. With the adaptive disturbance model, the accuracy of model prediction in DMC is improved, and consequently, a better performance in disturbance rejection is abtained.(2) A novel multi-iteration pseudo-linear regression (MIPLR) method is proposed. In estimating ARMA(X) processes, the cost function of recursive identification algorithms is not a quadratic form of the model coefficients vector, thus no analytical solution is available and the performance of traditional methods is not satisfactory. This can be improved by introducing the concept of multi-iteration into recursive methods. MIPLR uses each data sample in multiple iterations, improving the model accuracy and convergence rate.(3) A robust model predictive control (RAMPC) technique with an adaptive disturbance model is developed. The dynamics of unmeasured disturbances are modeled by ARMA processes. In addition, the optimization in MPC is formulated as a Min-Max problem which takes into account data uncertainties. For lower computational burden, the Min-Max problem is reduced to a nonlinear Min one, and is solved by multi-step linearization method. Numerical simulations demonstrate the effectiveness of the proposed methods.(4) A closed-loop particle swarm optimization (CLPSO) algorithm is proposed by mapping PSO elements into the closed-loop system based on control theories. At each time step, a proportional-integral-derivative (PID) controller is used to calculate an updated inertia weight for each particle in swarms from its last fitness. With this modification, limitations caused by a uniform inertia weight for the whole population are avoided, and the particles have enough diversity.(5) Tuning of general multivariable process controllers is formulated as a Min-Max optimization problem, considering model uncertainty. A novel engineering oriented performance index is proposed and CLPSO is employed to solve the problem. Simulation results demonstrate the superiority of CLPSO over other methods, and the applications to PTA equipment control loops prove the effectiveness of the proposed method.At the end of this dissertation, the author also gives some suggestions to the further research in these fields.