Identification And Control For Non-Square And Nonlinear Systems With Constraints

With the development of industry technology, multivariable systems widely exist in the process of industrial production. On the implementation of automatic control of nonlinear multivariable systems and non-square systems, it is often accompanied by some difficult problems because of their special structures. And it is also difficult to control nonlinear multivariable systems and non-square systems directly if the constraints of the systems are considered in the actual process. This thesis is aimed at control problems of the typical nonlinear multivariable Hammerstein system and the non-square systems under the condition of the existence of constraints. The theoretical foundation of the method researched in this thesis is model predictive control (MPC). MPC is widely used in industrial production which is one of advanced control algorithms. Because of the congenital advantage for handling linear system with constraints and interference treatment, this thesis extends these advantages to more complex nonlinear, non-square system constraints.Firstly, the research necessity and practical value of this thesis are discussed, the problems which automatic control is met with are explained. Research on the control of constrained non-square and nonlinear multivariable system is very necessary. And the basic theory of MPC is illustrated.Secondly, the basic principle and the basic algorithms of MPC are detailed, including the most commonly used dynamic matrix control (DMC), generalized predictive control (GPC), etc. The problem description of MPC for constrained multivariable systems is detailed, ready for the next expansion of the theory.Aiming at the control problem of constrained nonlinear multivariable system, this thesis takes the multivariable Hammerstein model as the research object and a modified generalized predictive control algorithm (MGPC) is proposed. MPC is a model based control algorithm, so effective identification method to get a model is needed. An identification method for Hammerstein model based on steady-state responses is also proposed. The proposed MGPC considers the nonlinear characteristics and the constraint conditions of the system simultaneously, which deals with the nonlinearity directly in the process of getting optimal control law, and translates the optimization problem into a standard quadratic programming (SQP). The optimal control law is solved by one step during the iteration. Modelling and control simulations of a polymerization reaction is detailed to show the superiority of MGPC algorithm compared with the traditional QDMC and GPC.To solve the control problem of non-square systems with constraints, this thesis designs a QP transformation based predictive control algorithm which has an important practical value. The core of this method is also transforming the control problem into a quadratic programming problem. Simulations of the classic Kalman’s system and Shell control problem show the advantages of the proposed algorithm.In order to reflect the practical value of the research, the proposed method is applied to the automatic control of flushing process of fuel oil system. Through a comprehensive comparison with manual control, the proposed mothed has a better performance in both set point tracking and disturbance rejection.