An Introduction towards application to dual control for nonlinear model predictive control
Speaker: La Huu Chuong

Time: 9h30, Tuesday, March 31, 2015

Location: Room 4, Building A14, Institute of Mathematics, 18 Hoang Quoc Viet, Cau Giay, Hanoi

Abstract: Nonlinear Model Predictive Control (NMPC) is an iterative optimization procedure to implement feedback control for nonlinear systems. At each NMPC iteration, based on available measurements, current states and parameters of the system are estimated. Using these estimates, an optimal control problem (OCP) is solved to compute the control inputs on the prediction horizon. However only the first part of the computed control is applied to the system. New measurements are taken and the procedure is repeated on the new time horizon.

The conventional NMPC scheme presented as above makes use of only the estimates, not their quality. When controlling uncertain processes, it is important to take into account the uncertainties of the current estimates as well as the possibility of gaining information to improve the future estimates. Dual control refers to strategies that attempt to balance the performance control and the information gain. By rapidly obtaining accurate estimates without worsening much the control objective, we can have better control action in the future.

This improves the overall performance. In this talk, a new approach to dual control is introduced. We treat the problem of dual control in the context of NMPC and make use of nonlinear Optimal Experimental Design (OED). Examples of the rocket car, moon lander illustrate the efficiency of our approach as well as various aspects of controlling dynamic systems under uncertainties.

Nominal control and some other approaches are also discussed and compared.

Back