Deep Integration and Learning of Measurement Data in Active Distribution Power Networks for Load Flow Estimation Considering Incomplete Information

Abstract

To reach the goals of carbon peaking by 2030 and carbon neutrality by 2060 in China, a growing penetration of PV generation is integrated into the modern distribution power grid, causing challenges in operating the grid to meet security requirements due to the increased dynamics and stochastics. A deep neural network (DNN)-based method is proposed in this paper for estimating load flow solutions in active distribution power networks that typically suffer from incomplete measurements. Adaptive DNN models are trained from massive historical observations obtained from actual measurements or high-fidelity simulations, whose hyperparameters are tuned automatically. Comprehensive case studies are conducted on the IEEE 123-node feeder model with renewable generation considering various levels of missing information as inputs, which validate the effectiveness and robustness of the proposed method.