Research On Nonlinear Model Predictive Control And Its Application In Power Generation Control

Due to the frequent changes of the operating point right across the whole operation range, industrial processes are generally nonlinear. In real-time industrial processes, there exists a large amount of physics constraints. The control problems of the constrained nonlinear industrial processes present great challenge. Model predictive control(MPC) is just an advanced optimal control scheme based on a system model, which can directly address the receding horizon and constraint handling. Traditional MPCs are most effective for linear system. In general, the nonlinear model predictive control (NMPC) also online solves an optimization problem, using the sequential quadratic programming(SQP) method, the penality function method, and the Karush-Kuhn-Tucker Conditions method, ect. The resulting nonlinear programming problems are usually non-convex, and the online computational burden is generally large for most complex systems. This dissertation is concerned with the nonlinear MPC optimazitation method and its application in complex systems.Considering a class of affine nonlinear state-space system, the paper proposed a nonlinear model predictive control strategy based on input/output feedback linearization. An iterative quadratic program (IQP) routine is utilized, which can guarantee its convergence. The proposed algorithm can reach a feasible solution over the entire prediction horizon. This strategy is proved effective in both a continuous-time numerical example and the continuous stirred tank reactors (CSTR) example.This developed input/output feedback linearization based MPC is extended to the fast doubly fed induction generators (DFIGs) in wind power plants, and also the permanent magnet synchronous motors(PMSMs). These two fast process are quite nonlinear and contain many uncertainties, since the DFIG’s dynamic is dependent on the wind speed, and the DFIG’s electromagnetic torque is nonlinear function of the stator and rotor current. The PMSMs are with interaxis nonlinear couplings, unavoidable and unmeasured disturbances, as well as parameter variations and load torque. The DFIG is input/output feedback linearized, since the total degree is smaller than the number of the states; while the PMSM is exactly state feedback linearized since the total degree is equals to the number of the states. Compared to the existing methods, the computational burden is reduced while the convergence is guaranteed.The ultra-supercritical (USC) unit is an advanced power generation technique with high plant efficiency, high coal utilization and low emission. It is more difficult to realize a coordinate control to achieve fast and stable dynamic response during load tracking and grid frequency disturbances, due to its complexity, nonlinearity and large-scale. This dissertation presents a decentralized hierarchical nonlinear model predictive control (DHMPC) to incorporate both the plant-wide economic process optimization and regulatory process control into a hierarchical control structure, in which the model predictive control technology is utilized to solve the multi-layer optimization problem. In the DHMPC, the upper layer optimizes the plant-wide operation based on the nonlinear USC model, subject to the performance index containing economic and environmental criteria. The aim of this level is to reduce the overall operational costs, and realize the disturbance rejection from power generation process and power grid. The lower level concentrates on designing the high accuracy USC tracking control set by the upper layer.While the nonlinear DHMPC optimization problems can be non-convex, neuro-fuzzy network(NFN)modeling on USC is incorporated to facilitate the convex quadratic program(QP) routine. Detailed analysis on load tracking and grid frequency disturbances via simulations has been addressed to demonstrate the effectiveness of the proposed nonlinear DHMPC.