Real-time security margin control using deep reinforcement learning

This paper develops a real-time control method based on deep reinforcement learning aimed to determine the optimal actions maintain sufficient secure operating limit. The limit refers most stressed pre-contingency point of an electric power system that can withstand set credible contingencies without violating stability criteria. developed uses hybrid scheme is capable simultaneously adjusting both discrete and continuous action variables. performance evaluated modified version Nordic32 test system. results show quickly learns effective policy ensure for range different scenarios. also compared rule-based look-up table adapted spaces. managed achieve significantly better all defined sets, indicating possibility variables resulted in more flexible efficient policy.