FDI attack detection using extra trees algorithm and deep learning algorithm-autoencoder in smart grid

Abstract

Today's smart grid (SG) combines the classical power system with the information technology, leading to a cyber-physical system (CPS). Its strong dependencies on digital communication technology bring up new vulnerabilities. Among the existing cyber-attacks, false data injection (FDI) attack is targeted at compromising power system state estimation by injecting false data into meter measurements. Such a malicious attack cannot be identified by the traditional bad data detection (BDD) techniques. According to this problem, finding a way to detect this kind of attack is necessary. Therefore, to overcome this problem, extremely randomized trees algorithm is proposed in this paper because of its high accuracy and fastness compared with other algorithms like support vector machine (SVM), random forest, and k-nearest neighbor (KNN). It is evident that as the system size increases, the computational complexity increases. Thus, a stacked autoencoder is designed along with extremely randomized trees classifier to tackle with dimensionality issue. Autoencoder is a deep neural network which can present a new representation of data in lower dimension. Performance evaluation on the standard IEEE 14-bus, IEEE 30-bus, IEEE 57-bus and 118-bus systems verifies that the proposed model outperforms other algorithms in the literature by improving the detection accuracy and reducing false positive rate.