Adversarial attacks on deep learning models in smart grids
Jingbo Hao, Yang Tao
Abstract
A smart grid may employ various machine learning models for intelligent tasks, such as load forecasting, fault diagnosis and demand response. However, the research on adversarial machine learning has attracted broad interest recently with the rapid advancement of deep learning techniques, which poses an evident threat to those deep learning models deployed in smart grids. In the face of the emergent problem, we make a compact survey of the adversarial attacks against deep learning models in smart grids. The research status of deep learning applications in smart grids and adversarial machine learning is briefly summarized firstly. Adversarial evasion and poisoning attacks in smart grids are analyzed and exemplified respectively with focus. To mitigate the threat typical countermeasures against adversarial attacks are also presented. From the survey it can be concluded that the threat of adversarial attacks in smart grids will be a kind of long-term existence and need continuous attention.