A survey of Graph Neural Networks for smart-grid cybersecurity
This is an ongoing survey of graph neural network (GNN) methods applied to cybersecurity in smart grids. The grid is a graph of buses, lines and substations, and attacks against it exploit that structure as much as the signal. GNNs learn over the structure directly, which has made them a popular class of models for detecting and localizing false data injection (FDI), load redistribution and topology-based intrusions.
Scope
The survey covers roughly 100 papers from 2021 to 2026 in four areas:
- Detection — binary and multi-class classifiers for attack vs. normal operation, including spectral GNNs, GAT/GCN variants, and spatiotemporal fusions (GC-LSTM, DST-GNN).
- Localization — node- and edge-level identification of compromised measurements, increasingly via attention and physics-informed constraints (PING, Hodge-based approaches).
- Resilience and response — reinforcement-learning-augmented GNNs, load-restoration GNNs, and adversarial training as a defense layer.
- Explainability — what a grid operator can do with a GNN’s output under time pressure.
Status
Draft v3 is under revision. The taxonomy figure and PRISMA diagram are done. Remaining work is tightening the discussion of physics-informed approaches and filling coverage gaps in three recent FDIA-localization papers.
Why this matters
A missed attack on the grid costs outages, and a false alarm can cause disconnections that propagate. GNNs have limits, but they are the first class of model that works on the grid’s own graph structure instead of flattening it into a vector.
The PDF and a link to the related code will be posted here when the draft is submitted.