Multi‐Objective Real‐Time Control of Operating Conditions Using Deep Reinforcement Learning

The secure operation of modern power grids is often challenged by the increasing penetration of renewable resources and natural disasters due to their intermittent and uncertain nature. Nevertheless, deriving fast and effectively coordinated control actions for mitigating risks of real-time operating conditions remains a grand challenge. This chapter presents a novel Deep Reinforcement Learning (DRL)-based methodology to achieve multi-objective real-time power flow control in a real-world environment. State-of-the-art DRL algorithms, e.g., deep Q network (DQN), Proximal Policy Optimization (PPO), and Soft Actor-Critic (SAC), are used to train AI agents that learn their optimal control policies by interacting with high-fidelity power grid simulators for controlling line flows, bus voltages, and transmission losses without violating various constraints. The effectiveness of the proposed method is verified by numerical case studies conducted on the IEEE 14-bus, the Illinois 200-bus model, and the actual Energy Management System (EMS) environment in the control center of State Grid Jiangsu Electric Power Company, indicating well-trained DRL agents can autonomously mitigate operational risks in real-time environment.