PowerGNN: A Topology-Aware Graph Neural Network for Electricity Grids

TL;DR

The paper addresses accurate power system state forecasting under high renewable penetration by exploiting grid topology through a topology-aware spatio-temporal Graph Neural Network. It introduces a model that blends GraphSAGE-based spatial convolutions with per-node GRUs to learn joint spatial and temporal dependencies, using a physics-informed graph representation of buses and lines. Evaluated on the NREL-118-bus system with realistic renewable profiles, the approach outperforms fully connected networks, linear regression, and rolling mean baselines, achieving average RMSEs of $0.1567$, $0.1714$, $0.1292$, and $0.1460$ for voltage magnitude, voltage angle, active power, and reactive power, respectively. The work demonstrates robustness across locations and operating conditions and shows potential for improved congestion prediction and operational decision support in future grids.

Abstract

The increasing penetration of renewable energy sources introduces significant variability and uncertainty in modern power systems, making accurate state prediction critical for reliable grid operation. Conventional forecasting methods often neglect the power grid's inherent topology, limiting their ability to capture complex spatio temporal dependencies. This paper proposes a topology aware Graph Neural Network (GNN) framework for predicting power system states under high renewable integration. We construct a graph based representation of the power network, modeling buses and transmission lines as nodes and edges, and introduce a specialized GNN architecture that integrates GraphSAGE convolutions with Gated Recurrent Units (GRUs) to model both spatial and temporal correlations in system dynamics. The model is trained and evaluated on the NREL 118 test system using realistic, time synchronous renewable generation profiles. Our results show that the proposed GNN outperforms baseline approaches including fully connected neural networks, linear regression, and rolling mean models, achieving substantial improvements in predictive accuracy. The GNN achieves average RMSEs of 0.13 to 0.17 across all predicted variables and demonstrates consistent performance across spatial locations and operational conditions. These results highlight the potential of topology aware learning for scalable and robust power system forecasting in future grids with high renewable penetration.

Figures (3)

• Figure 1: Distribution of average RMSE across all buses for each model. Each box summarizes the mean RMSE across four state variables at each of the 118 buses. The GNN exhibits the lowest median error and the least variance, indicating both accuracy and robustness.
• Figure 2: Distribution of prediction errors by feature. Probability density functions of absolute errors for each state variable and model. The GNN model (blue) consistently shows more concentrated distributions near zero error compared to other models, particularly for voltage-related variables. Error values are clipped at 1.5 for better visualization of the distribution shapes.
• Figure 3: RMSE by bus, state variable, and model. Heatmap visualization of prediction errors across all 118 buses (y-axis) for each state variable (columns) and model (x-axis within each column). Color intensity indicates error magnitude, with darker colors representing lower errors. Notable clusters of elevated errors appear at specific buses across multiple models, while the GNN maintains more consistent performance throughout the network.