Event-triggered Suboptimal Control Based Adaptive Reinforcement Learning

Abstract

The paper presents event-triggered suboptimal control based adaptive reinforcement learning. Linear quadratic optimal control requires the use of all state variables feedback, but in engineering practice, not all states can be measured or easy to be measured. Therefore, suboptimal control becomes very significant. Under event-triggered (ET) mechanism, we give the expression of suboptimal control, propose a novel triggering condition and prove the stability of close-loop system. Adaptive Q-learning is a kind of reinforcement learning, which is used to structure critic network. Finally, simulation example is represented to show the proposed is valid.