Learning Model-Free Reference Tracking Control with Affordable Systems

Abstract

This chapter validates and discusses the application of two intelligent learning control techniques, namely Model-free Value Iteration Reinforcement Learning (MFVIRL) and Virtual State-feedback Reference Tuning (VSFRT), for linear output reference model (ORM) tracking of three inexpensive lab scale systems which are interacted with by the help of modern software and hardware. The lab systems consist of an Electrical Braking System (EBS) emulator which is a very representative resistive-based dissipative device, a lab scale Active Temperature Control System (ACTS) which is another area of interest for home/industrial applications and a generic Voltage Control Electrical System (VCES) device, as representative complex, nonlinear and multidimensional systems. The control techniques are unique in the sense that they pertain to different paradigms such as artificial intelligence and classical control, however they share the same learning goal formulation. Herein, learning is based on a virtual state representation built from input–output (I/O) data measured from the systems by interaction, under exploration settings. The learned linear feedback controllers indirectly linearize the closed-loop system, by adequate selection of the linear ORM whose behavior is to be replicated by the closed-loop.