Data-driven control of a class of discrete-time linear complex dynamical networks

Abstract

The study presented in this poster deals with the model-free optimal synchronization control of the complex dynamical networks (CDNs) with unknown non-identical dynamics. The traditional synchronization control methods of the CDNs require the complete knowledge of system dynamics. However, there usually exists substantive data which contains the information of system state variables. In addition, we are surrounded by big data with the advent of rapid and low-cost data acquisition techniques. We can employ the data-driven method to acquire the control law. Thus, in this study, we propose a data-driven model-free optimal control scheme based on reinforcement learning (RL) to achieve the synchronization of the CDNs. First, a data-driven adaptive distributed observer is designed to estimate the reference node state for each node. A feedforward control law is designed to compensate the coupling dynamics of the CDNs. With the help of the adaptive distributed observer and the feedforward control law, the synchronization control of the CDNs transforms into the optimal control problem of an augmented systems, which is composed of the system dynamics with compensation and the reference node dynamics. Then, a model-free RL-based control method using measurable data is developed to solve the optimal control problem. Finally, the simulation results are provided to demonstrate the effectiveness of the developed approach.