GraphSAGE-Based Probabilistic Optimal Power Flow in Distribution System

The large-scale penetration of wind and photovoltaic (PV) power generation brings significant uncertainties to the distribution system power flows. In this paper, a data-driven Graph Sample Aggregate (GraphSAGE) - based deep learning method is proposed to solve the probabilistic optimal power flow (POPF) problems for distribution systems, whose key concept is to flexibly aggregate and transform the nodal and branch features of the distribution system. By fully extracting the features via aggregation, the proposed method is capable of directly learning the implicit correlation among renewable and load uncertainties to improve the calculation accuracy as well as reducing the time and efforts on mathematical modelling. Cases studies on IEEE 33-node system are conducted to compare the effectiveness of the GraphSAGE-based method with other classical deep learning network based methods. Numerical simulation results further prove the effectiveness and accuracy of the GraphSAGE-based method on POPF problems.