Graph Attention Enabled Convolutional Network for Distribution System Probabilistic Power Flow

Abstract

Probabilistic power flow (PPF) is pivotal to quantifying the state uncertainties of distribution power systems. However, it is very challenging due to underlying complex correlations among renewable outputs. To address this problem, a graph attention enabled convolutional network (GAECN) is proposed to approximate PPF in this article. Specifically, the graph convolutional layer of GAECN is used to aggregate the correlations among the nodal power injections during the training process. Within this layer, a full self-adaptive graph convolutional operation is proposed to automatically capture and learn the implicit correlation for achieving significantly enhanced accuracy of PPF. This layer is then followed by the convolutional neural network to capture the uncertain generation of renewable energy to achieve the robust computation of system state variable distributions. The simulation results demonstrate the accuracy and efficiency of the proposed method in IEEE 33, PG&E 69-node, 118-node, and practical 76-node distribution systems.