A Multimodel Approach Based On Nonlinearity Measures And MLD-MPC

A multimodel approach based on nonlinearity measures and MLD-MPC is proposed for processes with strong nonlinearities as a result of large setpoint changes and wide operating ranges. In order to deal with the basic problems of the multimodel approach, i.e. the "division" and the "combination", two nonlinearity-measure-based dividing methods are developed, and the mixed logical dynamical (MLD) modeling method is adopted to "combine" the linear model bank which approximates the nonlinear system. For a better control performance, the MPC technique is employed based on the MLD model to design a global controller. The main contributions of this dissertation are as follows:(1) An included angle nonlinearity measure is proposed for SISO Hammerstein-like systems (i.e. nonlinear systems with Hammerstein or Wiener model structures). This method assesses the nonlinear degree of a system within an operating space by computing the variation of the slope of its static I/O curve, and then provides information for selection of a control strategy and decomposition of the operating space.(2) The included angle method is only applicable to SISO Hammerstein-like systems. To break through this limitation, a gap metric based nonlinearity measure is proposed for general analytic nonlinear systems. The gap metric based method takes full advantage of the gap metric as a measurement tool. It analyzes the dynamical behavior of the considered nonlinear system before a controller is designed, measures its open-loop nonlinearity, and finally helps to select a control strategy.(3) Based on the two proposed nonlinearity measures, two methods for dividing nonlinear systems are proposed. The idea is as follows:Under a certain measurement criterion, compare the distances between the linear models with the threshold value one by one. If the distances between the linear models within a range are all smaller than the threshold value, then this range is defined as a subspace and a linear submodel is built for it. Then repeat the whole procedure from the next linear model until the entire operating space is decomposed. This method is systematic and effective; it addresses the "division" problem satisfactorily.(4) A mixed logical dynamical model is employed to "combine" the linear models determined by the proposed dividing methods. Based on the MLD model, a global MPC controller (MLD-MPC based multimodel controller) is designed under a unified and integrated performance index. The MLD-MPC based multimodel controller can schedule the linear submodels properly, which avoids the chattering during model switching and improves the robust stability of the system; and can make full use of the dynamics of all the linear submodels in every sample period to optimize the nonlinear system dynamically, which improves the control performance.