Deriving Fast AC OPF Solutions via Proximal Policy Optimization for Secure and Economic Grid Operation

Solving AC optimal power flow (OPF) with operational security constraints remains an important but challenging optimization problem for secure and economic operations of the power grid. With the ever-increasing penetration of renewable energy, fast and large fluctuations in power and voltage profiles are observed on a daily basis; thus, fast and accurate control actions derived in real-time are needed to ensure system security and economics. This paper presents a novel method to derive fast OPF solutions using state-of-the-art deep reinforcement learning (DRL) techniques, which can greatly assist grid operators in making rapid and effective decisions. Imitation learning (IL) is adopted to generate initial weights for the neural network (NN); and more importantly, proximal policy optimization (PPO) algorithm is utilized to train and test stable and robust AI agents. The training and testing procedures are conducted on both the IEEE 14-bus and the Illinois 200-bus systems. The results demonstrate the effectiveness of the proposed methodology and illustrate its great potential for assisting grid operators in real-time operation.