Power Failure Cascade Prediction using Graph Neural Networks

TL;DR

This work tackles predicting power failure cascades by introducing a flow-free graph neural network that forecasts cascade dynamics across generations using topology $G=(V,E)$, initial contingencies, and node injections. The model processes data through an Initial, Attention, Averaging, and Final stage to output per-edge failure-step probabilities, bypassing repeated DC power-flow computations. Evaluations on IEEE 89- and 118-bus systems show the GNN outperforms load-specific influence models across graph- and branch-level metrics, while delivering nearly two orders of magnitude faster predictions than flow-based simulators. The approach enables fast, topology-aware risk assessment and scalable cascade predictions suitable for real-time or large-scale planning, with demonstrated generalization to variable loading via random scaling factors ${\alpha}$.

Abstract

We consider the problem of predicting power failure cascades due to branch failures. We propose a flow-free model based on graph neural networks that predicts grid states at every generation of a cascade process given an initial contingency and power injection values. We train the proposed model using a cascade sequence data pool generated from simulations. We then evaluate our model at various levels of granularity. We present several error metrics that gauge the model's ability to predict the failure size, the final grid state, and the failure time steps of each branch within the cascade. We benchmark the graph neural network model against influence models. We show that, in addition to being generic over randomly scaled power injection values, the graph neural network model outperforms multiple influence models that are built specifically for their corresponding loading profiles. Finally, we show that the proposed model reduces the computational time by almost two orders of magnitude.

Figures (8)

• Figure 1: Block diagram of the GNN model.
• Figure 2: Failure size error rates $l_{size}^\alpha$ of various models for IEEE89 (left) and IEEE118 (right) against scaling values $\alpha$.
• Figure 3: Final state error rates $l_{state}^\alpha$ of various models for IEEE89 (left) and IEEE118 (right) against scaling values $\alpha$.
• Figure 4: Failure step error rates $l_{failure\text{-}step}^\alpha$ of various models for IEEE89 (left) and IEEE118 (right) against load scaling $\alpha$.
• Figure 5: Branch final state prediction error rate $l_{state,e}$ for IEEE89 (left) and IEEE118 (right) averaged over all load scaling values $[1,2]$ against branch failure frequencies $l_{freq,e}$.