Deep Deterministic Policy Gradient Control and Optimal for Grid-forming Converter

In the new energy grid connected system, the grid-forming inverter can improve the stability of weak grid or island operation mode, VSG control can enhance system stability and improve dynamic response. but it is difficult to adjust the control parameters in the dynamic process. Due to the self-learning ability of deep reinforcement learning (DRL), especially the advantage of deep deterministic strategy gradient (DDPG) in high-dimensional continuous control of power converter, this paper proposed an adaptive VSG controller based on DDPG, which can realize real-time optimization for the virtual inertia J and virtual damping D. In addition, it can also optimize the PI parameters of PCC voltage loop according to the change of system load, which can reduce the fluctuation of PCC voltage when the system load fluctuates, and improve the response speed and steady-state performance of PCC voltage recovery. The verification results confirm the effectiveness of PCC voltage stability.