Title: SAVER: Safe Learning-Based Controller for Real-Time Voltage Regulation

Abstract: Fast and safe voltage regulation algorithms can serve as fundamental schemes for achieving a high level of renewable penetration in the modern distribution power grids. Faced with uncertain or even unknown distribution grid models and fast-changing power injections, model-free deep reinforcement learning (DRL) algorithms have been proposed to find the reactive power injections for inverters while optimizing the voltage profiles. However, such data-driven controllers can not guarantee satisfaction of the hard operational constraints, such as maintaining voltage profiles within a certain range of the nominal value. To this end, we propose SAVER: SAfe VoltagE Regulator, which is composed of an RL learner and a specifically designed, computational efficient safety projection layer. SAVER provides a plug-and-play interface for a set of DRL algorithms that guarantees the system voltages to be within safe bounds. Numerical simulations on real-world data validate the performance of the proposed algorithm.