GAT- DNet: High Fidelity Graph Attention Network for Distribution Optimal Power Flow Pursuit

The optimal operation of the distribution grid aims to efficiently manage the flow of electricity from sources to end-users, ensuring a resilient and sustainable grid. To perform optimal operation, Optimal Power Flow (OPF) plays a pivotal role in solving a complex optimization problem due to the system operational constraint and the provided OPF solutions. Traditional OPF algorithms are based on mathematical programming techniques, which can be computationally expensive and difficult to implement for complex large-scale networks. The most recent development in learning-based computational models has paradigm shifted towards data-driven approaches to OPF. This paper proposes a high-fidelity Graph Attention Networks (GAT) model that leverages the attention mechanism and graph convolution feature mapping to learn informative node representations for OPF solutions. We evaluated the proposed model on the IEEE-33 bus power distribution system, which is a representative example of a real-world distribution system. The proposed model outperformed the state-of-the-art MPGCN and DNN models in terms of loss function and performance evaluation indices. The proposed GAT model showcases its effectiveness and promises results for addressing the OPF problem in the distribution grid, as evidenced by performance evaluation metrics: MSE (0.0167), RMSE (0.1294), and MAE (0.0700), respectively.