Reinforcement Learning for Optimal Primary Frequency Control: A Lyapunov Approach

As more inverter-connected renewable resources are integrated into the grid, frequency stability may degrade because of reduction in mechanical inertia and damping. A common approach to mitigate this degradation performance is use power electronic interfaces for primary control. Since can realize almost arbitrary responses changes, they not limited reproducing linear droop behaviors. To fully leverage their capabilities, reinforcement learning (RL) has emerged as a popular method design nonlinear controllers optimize host objective functions. Because both synchronous generators would be significant part grid near intermediate future, learned controller former should stabilizing with respect dynamics latter. overcome challenge, we explicitly engineer structure neural network-based such that guarantee system by construction, through Lyapunov function. recurrent network architecture used efficiently train controllers. The resulting only local information outperform optimal well other state-of-the-art approaches.