Learning-Observer-Guided Decentralized Control of State-Unknown Interconnected Systems: Enabling Estimation-Aware Self-Triggered Transmission

This article concentrates on the decentralized self-triggered control problem for state-unknown nonlinear interconnected systems within reinforcement learning framework. To simultaneously estimate the partially unknown dynamics and unavailable state of nonlinear interconnected systems, a decentralized learning observer is designed for the constructed auxiliary subsystems via output data. From the resource-efficient orientation, a novel dynamic factor-enhanced self-triggered scheme (DFSTS) with dead-zone operation is proposed to predict the data transmission and controller updating moment. Distinct from the existing schemes, the distinguishing advantages of DFSTS are twofold: 1) the successive monitoring for event-triggered condition is not required; 2) the utilization of dynamic factor and dead-zone operation leads to the larger release interval. Moreover, a DFSTS-incorporated critic-sole neural network is presented to approximate the optimal cost function with the improved weight tuning policy, in which the constraint of initial admissible control signal is obviated. It is assured that the critic weight approximation error and the target system state are uniformly ultimately bounded under the designed intelligent control strategy. Eventually, the simulation results containing the effectiveness validation and comparison analysis illustrate the practicability and superiority of the learning-observer-guided decentralized self-triggered control algorithm.