Probabilistic Power Flow of Distribution System Based on a Graph-Aware Deep Learning Network

Quantifying the uncertainties in the distribution system is critical for economic load dispatch yet of great challenge. To address this issue, a graph-aware deep learning network (GADLN) for probabilistic power flow (PPF) calculation is proposed considering the unknown correlation distribution pattern among the wind and solar power generation. By fully utilizing the convolutional operation to aggregate the correlation among nodal active and reactive power injections, the distribution features of the distribution system state variables brought by uncertain wind, solar power, and load demand can be well captured. In this regard, the proposed GADLN can achieve enhanced effectiveness and efficiency without prior knowledge about the correlation of wind and solar power profiles. The case studies are carried out and compared with the state-of-art based on the IEEE 33-node system. Simulation results show that the proposed model outperforms the state-of-art in terms of PPF calculation efficiency and accuracy.