On Lexicographic Multi-objective Nonlinear Model Predictive Control

As a class of computer control algorithms, which are relatively well-studied, model predictive control (MPC) based on linear models has be applied to many fields successfully. In recent years, because of the increasing requirement of industrial control, the theoretical value and practical usefulness of nonlinear model predictive control (NMPC) become more and more evident, which makes NMPC be a focus of the research on control theory.At present, there are two theoretical obstacles in the application of NMPC. Firstly, the computational load of NMPC algorithms is generally too heavy to satisfy the demand on computing speed of online optimization in industrial engineering. Secondly, in recent research on NMPC, the multi-objective control problem is often transformed into a single-objective control problem for the convenience in solving, which, however, lacks consideration of the constrained, multi-objective control problem resulted from actual industrial control. To handle with above problem, this dissertation mainly worked on the structure and efficient algorithms of the lexicographic multi-objective nonlinear model predictive controller. The achievements are:(1) As the essential work, research on multi-objective model predictive controller for linear systems has been done at first. For a multi-zone furnace for crystal growth, by introducing the stair-like control strategy, the stair-like multivariable model predictive controller based on genetic algorithm has been proposed and the simulations were carried out. Based on lexicographic method, the characters of modular multivariable controller was investigated, including the structure, selection of the primary control input and the management of control constraints and objectives in different forms. Then the modular multivariable model algorithmic controller was proposed and demonstrated by simulations of the Shell standard control problem.(2) For modular multivariable controller, some efficient algorithms of NMPC have been improved. The cause of error in existing one-step NMPC was studied, and the expression of the error was then presented analytically, so the compensation could be raised. Using several one-step predictions instead of a multi-step prediction, a stair-like efficient NMPC algorithm has been proposed, and it could solve the multi-step NMPC with light computational load. Simulations and experiments on the water-tank control system have been done to validate the efficiency of the above algorithms, especially the efficiency with model mismatch.(3) The structure of modular multivariable NMPC was discussed, the management of objectives and the selection of control input was also studied. Based on one-step NMPC, modular multivariable NMPC has been realized, simulations on the water-tank control system validated its efficiency in solving nonlinear lexicographic multi-objective control problem.(4) To overcome the limitations of modular multivariable NMPC, the lexicographic multi-objective genetic algorithm was proposed. Then, the lexicographic multi-objective NMPC, which is more universal than the modular multivariable NMPC, has been established based on this genetic algorithm with stair-like control strategy. In this structure, a new strategy of selection of the primary control input has been proposed. Simulations on the water-tank control system verified its efficiency and equivalency to modular multivariable NMPC in solving nonlinear lexicographic multi-objective control problem.At last, a summarization of the dissertation was given and the remaining problems were pointed out for future research.