Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Generating adversarial inputs for a graph neural network model of AC power flow

Robert Parker

TL;DR

The paper addresses robustness of neural surrogate models for AC power flow by formulating two optimization problems to generate adversarial inputs against a CANOS-PF graph neural network on the IEEE 14-bus system. The maximum-error and constrained-error formulations identify input perturbations that maximize NN versus ground-truth discrepancies or enforce output-bound violations, respectively, solved via an interior-point method. Empirical results show sizable errors (up to $3.4$ p.u. in reactive power and $0.08$ p.u. in voltage) and small input perturbations (often two variables) can induce adversarial outputs, underscoring the need for verification and adversarially robust training. The findings motivate hardened validation methods for neural surrogates in power systems and suggest that training data near operational bounds could improve resilience, with implications for larger networks and more flexible input perturbations.

Abstract

This work formulates and solves optimization problems to generate input points that yield high errors between a neural network's predicted AC power flow solution and solutions to the AC power flow equations. We demonstrate this capability on an instance of the CANOS-PF graph neural network model, as implemented by the PF$Δ$ benchmark library, operating on a 14-bus test grid. Generated adversarial points yield errors as large as 3.4 per-unit in reactive power and 0.08 per-unit in voltage magnitude. When minimizing the perturbation from a training point necessary to satisfy adversarial constraints, we find that the constraints can be met with as little as an 0.04 per-unit perturbation in voltage magnitude on a single bus. This work motivates the development of rigorous verification and robust training methods for neural network surrogate models of AC power flow.

Generating adversarial inputs for a graph neural network model of AC power flow

TL;DR

The paper addresses robustness of neural surrogate models for AC power flow by formulating two optimization problems to generate adversarial inputs against a CANOS-PF graph neural network on the IEEE 14-bus system. The maximum-error and constrained-error formulations identify input perturbations that maximize NN versus ground-truth discrepancies or enforce output-bound violations, respectively, solved via an interior-point method. Empirical results show sizable errors (up to p.u. in reactive power and p.u. in voltage) and small input perturbations (often two variables) can induce adversarial outputs, underscoring the need for verification and adversarially robust training. The findings motivate hardened validation methods for neural surrogates in power systems and suggest that training data near operational bounds could improve resilience, with implications for larger networks and more flexible input perturbations.

Abstract

This work formulates and solves optimization problems to generate input points that yield high errors between a neural network's predicted AC power flow solution and solutions to the AC power flow equations. We demonstrate this capability on an instance of the CANOS-PF graph neural network model, as implemented by the PF benchmark library, operating on a 14-bus test grid. Generated adversarial points yield errors as large as 3.4 per-unit in reactive power and 0.08 per-unit in voltage magnitude. When minimizing the perturbation from a training point necessary to satisfy adversarial constraints, we find that the constraints can be met with as little as an 0.04 per-unit perturbation in voltage magnitude on a single bus. This work motivates the development of rigorous verification and robust training methods for neural network surrogate models of AC power flow.
Paper Structure (9 sections, 2 equations, 3 figures, 3 tables)

This paper contains 9 sections, 2 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Problem formulation
  4. Test problem
  5. Computational results
  6. Neural network training results
  7. Maximum-error results
  8. Constrained-error results
  9. Conclusion

Figures (3)

  • Figure 1: Maximum absolute error obtained for each bus. Left: PQ buses. Right: PV and reference buses. In each case, the opposite set of buses is shaded gray. Generated using PowerPlots powerplots.
  • Figure 2: Errors achieved by Problem \ref{['eqn:max-error']} with maximization and minimization objective senses. Maximized error is omitted for bus 3 because the optimization algorithm failed to converge for this case.
  • Figure 3: Histogram of $l_1$ perturbation magnitudes necessary to satisfy adversarial constraints.