Nonlinear Model Predictive Control Based On Differential-Algebraic-Equation Systems

With the pressure from energy saving and increasing production costs, chemical processes are more than ever expected to operate under variable conditions. Process control need to consider nonlinear dynamics of the process under significant changes. Therefore research on nonlinear controller is important in theory and applications. As a kind of computer control algorithm developed for industrial processes, predictive control method has been attracting considerable attentions for its good performance, strong robustness and ability to deal with constraints. Nonlinear predictive control based on first principle models has clear physical meaning and adapts to wide range of operating points. Differential-algebraic equations (DAEs) arise naturally as dynamic models of chemical applications which is difficult to solve and to guarantee the stability of optimization based on this kind of model. This dissertation studies the theory and algorithms of nonlinear predictive control for varying load processes, which can be described as Hessenberg DAEs. Numerical tests on a high temperature gas cooled reactor pebble bed module (HTR-PM) of a nuclear power plant validate the research on varying load control.The main research work and contributions of the thesis are listed as follows:1. Characteristics of DAEs are analyzed firstly. Discretization strategies with different orthogonal collocation points are studied in the frame of simultaneous approach, and criteria are put forward to ensure the degrees of freedom of the resulting NLP. At last, continuous and discontinuous cases are presented to compare these strategies and conclusions are drawn from the results.2. The uniqueness of solutions to DAEs and the stability of the simultaneous approach based on Radau collocation points are studied. A two-layer optimization is presented for open-loop dynamic optimization. Here the same process model is used for both layers and the optimization of control performance and economic benefit can be implemented synchronously. The upper layer is a steady state optimization whose objective is the economic maximization in the terminal time. This layer is used to compute the set points for the controlled variables and the manipulated variables in the lower layer. The lower layer is a dynamic optimization whose objective includes both economic objective and terminal conditions.3. A two-layer closed-loop control structure combining feedforward compensation with feedback correction is constructed to deal with the nonlinear production process under varying load. Existence and uniqueness of the solution in rolling optimization are investigated from the practical point of view and asymptotic stability is proved under certain assumptions. The proposed algorithm has anti-interference ability in the presence of uncertain disturbances. An universal NMPC system based on DAEs model is completed.4. A simultaneous approach based on Radau collocation points is presented to solve the DAEs model of the reactor of a nuclear plant. This approach has good accuracy in discretization as well as has very good stability in solving stiff problems. Simulation results of the varying load processes demonstrate high efficiency of the approach, which takes much less time than the nested approach.5. Characteristics of HTR nuclear mechanism model are analyzed, and the model is verified. A controller combining nonlinear model predictive control with PID control is proposed to deal with the large-scale model, and the block technique is applied to achieve the purpose of online real-time applications. Different control schemes of varying load are designed and simulation results verify the effectiveness of the controller.At the end of the dissertation, contributions of this research are concluded and possible extensions in future research are discussed.