Affordable Access

Data-based optimal control

Publication Date
  • Computer Science
  • Design
  • Mathematics


Title Data-Based Optimal Control Wouter Aangenent,1 Dragan Kostic´,2 Bram de Jager,3 Rene´ van de Molengraft,4 and Maarten Steinbuch5 1,3,4,5Department of Mechanical Engineering 2Embedded Systems Institute Technische Universiteit Eindhoven P.O. Box 513, 5600 MB Eindhoven, P.O. Box 513, 5600 MB Eindhoven, The Netherlands The Netherlands [email protected], [email protected], [email protected], [email protected], and [email protected] Abstract—This paper deals with data-based optimal control. The control algorithm consists of two complementary subsys- tems, namely a data-based observer and an optimal feedback controller based on the system’s Markov parameters. These parameters can be identified on-line using only input/output data. The effectiveness of the resulting controller is evalu- ated with a regulation and a tracking control experiment, performed on a direct-drive robot of spatial kinematics. I. INTRODUCTION Plant modelling and model-based control design are the conventional steps to obtain feedback controllers. The per- formance of the resulting controller is strongly influenced by the quality of the underlying model. The derivation of an accurate model using first principles or using sys- tem identification can be a time consuming and not a straightforward task. Moreover, even complex models will not cover all system dynamics, and adaptive and robust feedback control methods have not completely overcome the problem of modelling errors. The problems inherent to plant modelling are the motivation of research in the field of data-based control techniques, such as: unfalsified control [1], iterative feedback tuning [2], pulse response based control [3], Markov data-based LQG control [4], data-based predictive control [5], virtual reference feedback tuning [6], disturbance-based control [7], and simultaneous perturba- tion stochastic approximation [8]. In the above techniques, the plant modelling step is circumvented, and the control design is

There are no comments yet on this publication. Be the first to share your thoughts.


Seen <100 times