An adaptive dynamic programming approach for closely-coupled MIMO system control

Abstract

In this paper, we investigate the application of adaptive dynamic programming (ADP) for a real industrial-based control problem. Our focus includes two aspects. First, we consider the multiple-input and multiple-output (MIMO) ADP design for online learning and control. Specifically, we consider the action network with multiple outputs as control signals to be sent to the system under control, which provides the capability of this approach to be more applicable to real engineering problems with multiple control variables. Second, we apply this approach to a real industrial application problem to control the tension and height of the looper system in a hot strip mill system. Our intention is to demonstrate the adaptive learning and control performance of the ADP with such a real system. Our results demonstrate the effectiveness of this approach. © 2011 Springer-Verlag.