Reinforcement Learning-Tuned Nonlinear Controller Parameters for Synchronization of Chua and Duffing Chaotic Systems

Reinforcement learning has recently emerged as a promising machine learning paradigm for the adaptive control of complex nonlinear systems by enabling online decision-making through continuous interaction with the environment. However, its application to the synchronization of chaotic systems using Lyapunov-based nonlinear controllers remains relatively unexplored, particularly with respect to comparing discrete-action and continuous-action reinforcement learning algorithms. To address this research gap, this study proposes reinforcement learning-assisted Lyapunov-based nonlinear controllers for the synchronization of chaotic systems. Specifically, two independent master–slave synchronization problems are considered, one based on Chua’s circuit and the other based on the Duffing oscillator. In each synchronization problem, Deep Q-Learning (DQN) and Deep Deterministic Policy Gradient (DDPG) are employed to adaptively adjust the nonlinear controller parameters online by maximizing the instantaneous reward at each sampling instant, thereby improving synchronization performance while reducing undesirable switching behavior in the control signals. The proposed approaches are evaluated through numerical simulations using reward, average reward, Integral Absolute Error (IAE), Integral Square Error (ISE), Integral Time Absolute Error (ITAE), Integral Time Square Error (ITSE), and control effort as performance metrics. The simulation results demonstrate that DQN-assisted approach consistently achieves higher reward and average reward values than DDPG-assisted approach for both synchronization problems, indicating a more effective reward-maximizing policy under identical operating conditions. For the Chua master–slave synchronization problem, DQN-assisted nonlinear controller provides superior synchronization performance, whereas for the Duffing master–slave synchronization problem, DDPG-assisted nonlinear controller achieves better performance in terms of the error-based performance indices and control effort. Furthermore, robustness analyses under external disturbances demonstrate that both proposed approaches successfully suppress disturbances and maintain stable synchronization between the corresponding master and slave systems. Overall, the results confirm that integrating Lyapunov-based nonlinear control with reinforcement learning provides an adaptive, robust, and effective framework for the synchronization of nonlinear chaotic systems, offering improved synchronization performance and enhanced practical applicability.