This is the code repo for paper Understanding the Safety Requirements for Learning-based Power Systems Operations.
We demonstrate for the power system control tasks, how deep reinforcement learning (DRL) algorithms perform under adversarial attacks. We show that small perturbations over state observations can cause model-free learning algorithms output unsafe decisions.