Study On The Theories And Applications Of Nonlinear Model Predictive Control

Nonlinear Model based predictive control (NMPC) not only is a valuable approach for solving practical control problems, but also is the frontier of Nonlinear control theory. The perceptible successes of MPC strategies can be attributed to several factors including its inherent ability to handle input and output constraints, time delay and incorporation of an explicit model of the plant into the optimization problem. This dissertation discusses two kinds of nonlinearity (or nonlinear system). One is a special nonlinear system with constraints, a linear system with time delay, and the affine-type nonlinear system, and the other one is a general nonlinear system with constrains.In the first part, not only a general overview of the linear and nonlinear model based predictive control algorithms is provided, also a generalized predictive control (GPC) algorithm is proposed, which takes full advantage of predictive information. Meanwhile, an adaptive Smith-generalized predictive controller is used to improve the dynamic behavior of the time-delay system. The Smith predictor structure instead of an optimal predictor in GPC is used to compute the predictions of the output of the plant and to compute the sequence of future control signal. Based on the model matching on the frequency of zero, the system parameters and delay time are identified by RFFM(Recursive Forgetting Factor Method) or RHT(Recursive Householder Transform). The algorithm can be applied to the control of the long time delay system with variable parameters and with unknown delay time.In the second part, the predictive control algorithms with constraints of variables are derived. The stability of the closed loop predictive control system and optimization feasibility are investigated. Some technical solutions to the optimization unfeasibility are proposed. Genetic algorithm (GA) is used for optimizing predictive controllers. GAs allow the use of Non-quadratic index so that it improves control performance.In the third part, since the plant is a general nonlinear model, the neural generalized predictive control algorithms, recursive least squares (RLS) learning algorithm, based on multilayer feedforward neural network, and orthogonal least square (OLS) learning algorithm, based on radial basic neural network, are presented respectively. Meanwhile, the model predictive neural control is investigated, in which the neural network controller is optimized using a calculus of variations approach to minimize the MPC cost function and is differs from work above.In the fourth part, the development of a nonlinear model predictive control algorithm based on feedback linearization method is studied. Two feedback linearization methods are presented. One is differential geometry method for affine-type nonlinear system, and the other one is based on the inversion of nonlinear system, which is capable of managing general nonlinear system from the theoretical point of view. Unfortunately, the inversion of nonlinear system sometimes is hard to be obtained.In the fifth part, since the time of computation can confine the applied area of NMPC, a hierarchical predictive control algorithm is developed. This algorithm is based on the decompositon-coordinations method.In the last part, NMPC studied above are applied to AUV's station keeping in a shallow waterwave environment.