Key points are not available for this paper at this time.
Abstract A deep learning-based power flow calculation model can directly fit the mapping relationship between the initial and final values of the power flow in a system, providing extremely fast computational speeds without causing ill-conditioned power flow issues. However, existing deep learning power flow calculation methods are often based on homogeneous graph neural networks, which do not differentiate between different types of nodes in power flow calculations and neglect the influence of line admittance on power flow distribution. A power flow calculation method based on a heterogeneous edge graph convolutional neural network is proposed in response to this issue. The proposed model underwent comprehensive simulations on a dataset expanded from the IEEE 57-node system, generating 10, 000 samples of different system configurations. Simulation experiments validated that the proposed model’s computational time is approximately one-tenth of the Newton-Raphson method, with a power flow distribution calculation accuracy reaching 98.44%. Comparative experiments further confirmed the effectiveness of the proposed model’s improvements in graph convolution.
Wu et al. (Sat,) studied this question.