On-line dynamic data system modeling and generalized forecasting compensatory control with applications.

Abstract

On-line Dynamic Data System (on-line DDS) modeling and Generalized Forecasting Compensatory Control (GFCC) with applications are studied in this thesis. On-line DDS is a real time modeling method for a time varying process, which identifies the model order and parameters simultaneously. Based on a defined Model Information Matrix (MIM) and its UDU$\sp{\rm T}$ factorization, a recursive algorithm estimates parameters for a candidate model set and provides the order determination information. The realization on a digital signal processing (DSP) board indicates that a 400Hz sampling frequency can be achieved for the on-line modeling of an ARMA(n,n-1)(n = 1, ... 8) model set. GFCC for high dimensional nonlinear systems is also studied. An on-line linearization method, a hierarchical forecasting and control strategy and an iterative algorithm are proposed. The convergence condition for the iteration algorithm is given. An adaptive cautious GFCC is developed for stochastic nonlinear systems where the separation theorem does not hold. The adaptive cautious GFCC takes both the state estimates and their confidence levels into account. A stability analysis for the MIMO linear system has been conducted, and the main result can be used to optimize the key parameters in the GFCC controller to get the maximum robustness boundary. The on-line DDS and GFCC have been successfully implemented in a CMM vibration control application. Two controller design philosophies, robust control and supervisory adaptive control, are used. Experiments indicate that both control methods can be used to suppress the vibration and achieve more than a 70% reduction in vibration magnitude. The developed high dimensional nonlinear GFCC algorithm has been used in a terminal interception control problem for an anti-tank missile. The nonlinear missile interception model is used directly in the controller design. A receding forecasting control index is used to get the guidance law. This makes the controller more robust to disturbances. The simulation results show that the terminal miss distance is less than 0.5 meters, which satisfies the interception requirements.