This paper proposes an adaptive dynamic programming (ADP)-based decentralized event-triggered control strategy for large-scale nonlinear systems with event-triggered scheme to efficiently reduce communication cost and computational burden. Under the event-triggered mechanism, an local neural networks (NNs)-based observer is introduced to identify the mismatched interconnections and the estimation error is guaranteed to be uniformly ultimately bounded (UUB). Then, a decentralized triggering condition related to the approximation error of interconnections is designed to reduce the local controller updates with guaranteed overall stabilization of large-scale systems. By virtue of critic-only structure, the local optimal control policy can be approximated via aperiodic tuning rule using ADP. In addition, the closed-loop large-scale system ensures to be asymptotically stable with adaptive triggering threshold according to Lyapunov method. Finally, the simulation results justify the theoretical analysis and illustrate the effectiveness of the proposed event-triggered control (ETC) strategy.
KEYWORDSadaptive dynamic programming, critic NNs, event-triggered, NN-based observer
INTRODUCTIONNowadays, there are many interconnected large-scale systems existing in daily life, engineering area, and other practical applications, such as ecosystems, transportation system, communication systems, power systems, and so forth. Besides, with the growing demand of production quality and economic efficiency for these large-scale systems, investigating the optimal control for interconnected large-scale systems is a hot research topic. In general, interconnections coupled by subsystems pose big challenges to design the control approaches for large-scale systems because they often have negative effect on system performance and may even lead to system instability [1-3]. Therefore, more and more attention has been paid on how to design effectively optimal controllers to achieve good performance for interconnected large-scale systems [4,5].Different control strategies have been developed to address interconnected large-scale systems, which are broadly categorized into centralized control, distributed control and decentralized control. The classical centralized control requires information of all subsystems and control integrity is dominated by the centralized controller, so it is impractical for the lack of errors tolerance and growing dimension of large-scale systems [6]. The distributed control algorithm guarantee the overall