Anomaly detection in smart grids using machine learning

Abstract

Today’s smart power system is threatened by an increasing number of cyberattack events, fast and accurate detection of attack events is essential for the safe and reliable operation of the smart grid. In this dissertation, anomaly detection strategies based on reinforcement learning (RL) are proposed. Moreover, a cyber attack location method in the power system is presented. Various tests are performed with IEEE 14-bus systems, and results illustrate the effectiveness of the proposed algorithms of accurate and delay reduction of detection against cyber attacks targeting the smart grid. In addition, the impact of multiple solutions to the security-constrained economic dispatch is investigated, and apply an approach to determine the optimal attack vector, which can lead to a significant increase in the operation cost. The optimal false data injection (FDI) attack vector can be determined by solving a bi-level linear programming problem (LP), but in this approach, the attack vector only needs to solve one LP. The result of the simulation test on the IEEE 14-bus system verifies the effectiveness of this model.