Optimal control of unknown nonlinear system under event-triggered mechanism and identifier-critic-actor architecture

This paper proposes an event-triggered adaptive control algorithm for general continuous-time systems with unknown system models. Unlike existing articles, the developed method does not require prior determination of the knowledge of system dynamics, and effectively reduces the update frequency of key signals through the introduction of event-trigger mechanism. Three neural networks (NNs) are designed in the identifier-critic-actor (ICA) architecture to learn the optimal control solution online. The unknown system is approximated by the identifier NN, the critic NN is designed to approach the optimal cost function, and the actor NN is designed to approach the optimal controller. Besides, under event-triggered control, the parameters of critic NN and actor NN as well as control signals are updated only at the trigger time determined by the event-trigger condition, which reduces effectively the computing burden and communication cost. The stability of event-trigger control and the convergence of parameters of three NNs are verified via Lyapunov method. Finally, two examples are presented to demonstrate the viability of the proposed algorithm.