DeepOPF: deep neural network for Running the Power Grids

It is well-known that an efficient optimal power flow (OPF) solver is crucial for cost-effective and reliable operation in modern power systems. However, in many cases, conventional iterative OPF solvers are not fast enough for the purpose. In this seminar, I will introduce our work on Deep Neural Networks (DNN)-based OPF solver, named DeepOPF, which is one of the leading studies on leveraging machine learning for designing fast OPF solvers. In DeepOPF, we propose a novel “Predict-and-reconstruct” framework, where we first train a DNN to predict a set of independent operating variables and then obtain the remaining ones by directly solving the power flow equations. Simulation results on IEEE standard cases show that for the standard OPF problem and its variants, our approach can generate feasible solutions with negligible optimality loss while speeding up the computation time by up to two orders of magnitude as compared to a state-of-the-art solver. At last, I will conclude the talk with our recent progress and future directions on this topic.