MxGPS: Multiplex Graph Transformers for a Power Grid Foundation Model

2026-07-15

AI Summary

This work addresses the sharp degradation in generalization performance of graph neural networks for power systems under topological changes, a phenomenon attributed to “topological overfitting.” To mitigate this, the authors propose MxGPS, a multi-path Graph Transformer architecture that employs a shared node encoder to jointly train multiple tasks—including static state estimation and AC power flow—in parallel. This design compels the model to learn the underlying physical principles of power grids rather than topology-specific patterns. Integrating self-supervised pretraining, multi-task fine-tuning, and cross-branch attention mechanisms, MxGPS achieves a 0% boundary violation rate on four zero-shot topologies with only 1.6 million parameters. Under topological shifts, its performance degrades by merely 39%, substantially outperforming existing methods, which exhibit performance drops ranging from 190% to 1400%.

Abstract

Single-task fine-tuning of graph neural networks (GNNs) for power grid problems exhibits a systematic failure mode: models that achieve the lowest in-distribution error degrade the most under topology shift. We term this topology overfitting: the tendency of task-specific gradient signals to encode relational structure particular to the training topologies rather than the underlying physics, causing models to fail on unseen grids despite strong in-distribution performance. To expose and address this failure mode, we introduce MxGPS (Multiplex GPS), a multiplex graph transformer that runs K task-specialised GPS branches over a shared node encoder, jointly trained on Static State Estimation (SSE) and AC Power Flow (PF) via a self-supervised pre-training and multi-task fine-tuning protocol, with a cross-branch attention module evaluated in ablation. The joint SSE+PF objective forces the shared encoder to simultaneously satisfy complementary gradient signals, preventing it from overfitting to topology-specific relational structure. Under a 3-fold sliding-window cross-validation spanning four unseen topologies (14-, 24-, 162-, and 300-bus), MxGPS attains 0% boundary violation rate (BVR) on all four zero-shot Power Flow topologies. Critically, models with substantially lower in-distribution PF error degrade by 190% to 1400% under topology shift, whereas MxGPS degrades by only 39%, an inversion that directly implicates topology overfitting as the failure mechanism rather than insufficient model capacity. With only 1.6M parameters (12x fewer than the GridFM reference baseline), MxGPS demonstrates that multi-task joint training is a principled and parameter-efficient mechanism for topology-agnostic generalisation in power grid foundation models.

Problem

• topology overfitting
• power grid
• graph neural networks
• generalization
• foundation model

Innovation

• topology overfitting
• multiplex graph transformer
• multi-task pre-training
• power grid foundation model
• zero-shot generalization