DeepOPF: A Feasibility-Optimized Deep Neural Network Approach for AC Optimal Power Flow Problems

To cope with increasing uncertainty from renewable generation and flexible load, grid operators need to solve alternative current optimal power flow (AC-OPF) problems more frequently for efficient reliable operation. In this article, we develop a deep neural network (DNN) approach, called DeepOPF, solving AC-OPF in fraction of the time used by conventional iterative solvers. A key difficulty applying machine learning techniques lies ensuring that obtained solutions respect equality inequality physical operational constraints. Generalized prediction-and-reconstruction procedure our previous studies, DeepOPF first trains DNN model predict set independent operating variables then directly compute remaining ones equations. Such an approach not only preserves power-flow balance constraints but also reduces number be predicted DNN, cutting down neurons training data needed. employs penalty zero-order gradient estimation technique process toward guaranteeing We drive condition tuning size according desired approximation accuracy, which measures its generalization capability. It provides theoretical justification using problems. Simulation results IEEE 30/118/300-bus synthetic 2000-bus test cases demonstrate effectiveness approach. They show speeds up computing two orders magnitude as compared state-of-the-art solver, at expense $< $ 0.2% cost difference.