A deep reinforcement learning method to control chaos synchronization between two identical chaotic systems

We propose a model-free deep reinforcement learning method for controlling the synchronization between two identical chaotic systems, one target and one reference. By interacting with the target and the reference, the agent continuously optimizes its strategy of applying perturbations to the target to synchronize the trajectory of the target with the reference. This method is different from previous chaos synchronization methods. It requires no prior knowledge of the chaotic systems. We apply the deep reinforcement learning method to several typical chaotic systems (Lorenz system, Rössler system, Chua circuit and Logistic map) and its efficiency of controlling synchronization between the target and the reference is demonstrated. Especially, we find that a single learned agent can be used to control the chaos synchronization for different chaotic systems. We also find that the method works well in controlling chaos synchronization even when only incomplete information of the state variables of the target and the reference can be obtained.