A Distillation-Based Attack Against Adversarial Training Defense for Smart Grid Federated Learning
Atef H. Bondok, Mohamed M. E. A. Mahmoud, Mahmoud M. Badr, Mostafa M. Fouda, Maazen Alsabaan · 2024
In the advanced metering infrastructure (AMI) of the smart grid, smart meters (SMs) are deployed to collect fine-grained electricity consumption data, enabling billing, load monitoring, and efficient energy management. However, some consumers engage in fraudulent behavior by hacking their meters, leading to either traditional electricity theft or more sophisticated evasion attacks. Evasion attacks aim to illegally reduce electricity bills while deceiving theft detection mechanisms. The current methods for identifying such attacks raise privacy concerns due to the need for access to consumers' detailed consumption data to train detection mechanisms. To address privacy concerns, federated learning (FL) is proposed as a collaborative training approach across multiple consumers. Adversarial training (AT) has shown promise in countering evasion threats on machine learning models. This paper, first, investigates the susceptibility of traditional electricity theft classifiers trained by FL to evasion attacks for both independent and identically distributed (IID) and Non-IID consumption data. Then, it investigates the effectiveness of AT in securing the global electricity theft detector against evasion attacks, assuming no misbehavior from the participant consumers in the FL process. After that, we introduce a novel attack, called Distillation, which can be launched during the AT process to make the global model susceptible to evasion at inference time. Finally, extensive experiments are conducted to validate the severity of the proposed attack.