Generalization of Graph Neural Network Models for Distribution Grid Fault Detection

2025-10-03

AI Summary

To address insufficient generalization of fault detection under dynamic distribution network topologies—such as reconfiguration and distributed energy resource (DER) integration—this paper systematically evaluates multiple graph neural network (GNN) architectures within an RNN-GNN hybrid framework. It pioneers the application of GraphSAGE, Graph Attention Network (GAT), and GATv2 to power system fault diagnosis, integrating them with GRU to form unified spatiotemporal models (RGNNs). Experiments on the IEEE 123-bus system demonstrate that RGATv2 achieves superior cross-topology generalization: its F1-score degrades by only ~12% under topology variations, substantially outperforming conventional RNNs and baseline GNNs (e.g., GCN, RGCN). The key contribution lies in empirically establishing the enhanced robustness of advanced attention-based GNNs—particularly GATv2—against dynamic grid topologies, thereby identifying GATv2 as the optimal GNN component for the current RGNN paradigm.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Computer Vision: Generative Adversarial Networks (GANs) for Vision

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Federated Web and WoT systems, including distributed, federated and edge-based data processing
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Fault detection in power distribution grids is critical for ensuring system reliability and preventing costly outages. Moreover, fault detection methodologies should remain robust to evolving grid topologies caused by factors such as reconfigurations, equipment failures, and Distributed Energy Resource (DER) integration. Current data-driven state-of-the-art methods use Recurrent Neural Networks (RNNs) for temporal modeling and Graph Neural Networks (GNNs) for spatial learning, in an RNN+GNN pipeline setting (RGNN in short). Specifically, for power system fault diagnosis, Graph Convolutional Networks (GCNs) have been adopted. Yet, various more advanced GNN architectures have been proposed and adopted in domains outside of power systems. In this paper, we set out to systematically and consistently benchmark various GNN architectures in an RNN+GNN pipeline model. Specifically, to the best of our knowledge, we are the first to (i) propose to use GraphSAGE and Graph Attention (GAT, GATv2) in an RGNN for fault diagnosis, and (ii) provide a comprehensive benchmark against earlier proposed RGNN solutions (RGCN) as well as pure RNN models (especially Gated Recurrent Unit (GRU)), particularly (iii) exploring their generalization potential for deployment in different settings than those used for training them. Our experimental results on the IEEE 123-node distribution network show that RGATv2 has superior generalization capabilities, maintaining high performance with an F1-score reduction of $sim$12% across different topology settings. In contrast, pure RNN models largely fail, experiencing an F1-score reduction of up to $sim$60%, while other RGNN variants also exhibit significant performance degradation, i.e., up to $sim$25% lower F1-scores.

Problem

• Benchmarking GNN architectures for robust distribution grid fault detection
• Evaluating generalization of fault detection models across changing grid topologies
• Comparing RGNN models against traditional RNN approaches for power system reliability

Innovation

• Combining Graph Attention Networks with recurrent neural networks
• Proposing GraphSAGE and GAT variants for fault diagnosis
• Systematically benchmarking generalization across topology changes