Optimal Control of Nonlinear Time-Delay Systems with Input Constraints Using Reinforcement Learning

In this paper, input-constrained optimal control policy for nonlinear time delay system is proposed in virtue of Lyapunov theories and adaptive dynamic programming method. The stability on delayed nonlinear systems is investigated based on linear matrix inequalities, upon which a sufficient stability condition is proposed. To implement the feedback control synthesis, a single neural network is constructed to work as critic and actor network simultaneously, which consequently reduces the computation complexity and storage occupation in programs. The weights of NN are online tuned and the weight estimate errors are proved to be convergent. Finally, simulation results are demonstrated to illustrate our results.