Least Effort Attack for Inverter-based Microgrids by Deep Reinforcement Learning

In inverter-based microgrids, the droop controllers can be adjusted by system operator, where the flexibility increases the risk of cyber-attack in this system. This paper proposes a deep reinforcement learning (DRL) method to identify the least effort attack for inverter-based microgrids. First, the linearized system model of microgrids under the operation point is derived. Then the small-signal stability regarding to the parameter space of droop control gains is analyzed. Finally, a deep reinforcement learning approach based on deep deterministic policy gradient is developed to find the least effort attack path of this system. Simulation study in a 3-inverter system is conducted to validate the proposed method.