New method generates adversarial inputs for AC power flow GNNs

Researchers have developed a method to generate adversarial inputs for graph neural network models used in AC power flow simulations. This technique involves formulating and solving optimization problems to identify input points that cause significant discrepancies between the neural network's predicted solution and the actual AC power flow equations. The study demonstrated this on the CANOS-PF model using a 14-bus test grid, showing errors up to 3.7 per-unit in reactive power and 0.08 per-unit in voltage magnitude. The findings highlight the need for robust verification and training methods for these neural network surrogate models. AI

IMPACT Highlights potential vulnerabilities in AI models used for critical infrastructure, necessitating improved robustness and verification techniques.

RANK_REASON Academic paper detailing a new method for generating adversarial inputs for GNNs in power flow simulation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• AC power flow
• CANOS-PF
• graph neural network
• PFΔ
• Robert Parker

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →