Adaptive Deep Reinforcement Learning for Virtual Synchronous Generator Control in Ship Power Systems

The impact of renewable energy resources in the future of marine power systems is significant and inevitable. Integration of distributed energy resources (DERs) into ship electrical systems will eventually lead to lack of overall inertia. As a result, undesirable dynamic performance and stability issues will arise during large disturbances. To mitigate the impact of low-inertia systems, virtual synchronous generators (VSG) have been modeled and utilized in recent years. In this paper, a deep reinforcement learning-based adaptive control approach is introduced for virtual synchronous generator control. As a part of this approach, we utilized weakly coupled deep Q-network (WCDQN) algorithm to adaptively adjust the inertia and damping parameters of the VSG controller as well as the parameters of the variable-speed diesel-electric propulsion system. To verify the robust performance of the proposed method, it is applied to a real diesel-electric testbed based on the national security multi-mission vessel (NSMV) training ship power system. The outcome compared the dynamic performance of the system with other recent approaches.