Perspectives on Stochastic Predictive Control with Autonomous Model Adaptation for Model Structure Uncertainty

Integrated stochastic optimal control and system learning to simultaneously reduce parametric and model structure uncertainty can create new avenues for achieving high-performance operation of uncertain systems using model predictive control. This paper presents a generic framework for stochastic optimal control with integrated (control-oriented) model structure adaptation, and discusses general solution methods and key research issues associated with the framework. The potential advantages of the proposed framework include autonomous maintenance of model predictive controllers as well as handling of multiple system models under closed-loop conditions, for example, for fault-tolerant control applications and dealing with systems with intrinsically uncertain dynamics.