Semi-Supervised Outlier Detection and Deep Feature Extraction for Detecting Cyber-Attacks in Smart Grids Using PMU Data

Qi, Ruobin; Rasband, Craig; Zheng, Jun; Longoria, Raul G.

doi:10.1007/978-3-030-43020-7_67

Cited by 1 publication

(1 citation statement)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The proposed method contains two main components: deep representation learning and semi-supervised anomaly detection [11][12][13][14][15]. The first step of the proposed method is to prepare a training dataset that contains only examples of normal events.…”

Section: Methodsmentioning

confidence: 99%

Detection of cyber-attacks on the power smart grids using semi-supervised deep learning models

Shchetinin

Velieva

2022

Discrete and Continuous Models

View full text Add to dashboard Cite

Modern smart energy grids combine advanced information and communication technologies into traditional energy systems for a more efficient and sustainable supply of electricity, which creates vulnerabilities in their security systems that can be used by attackers to conduct cyber-attacks that cause serious consequences, such as massive power outages and infrastructure damage. Existing machine learning methods for detecting cyber-attacks in intelligent energy networks mainly use classical classification algorithms, which require data markup, which is sometimes difficult, if not impossible. This article presents a new method for detecting cyber-attacks in intelligent energy networks based on weak machine learning methods for detecting anomalies. Semi-supervised anomaly detection uses only instances of normal events to train detection models, which makes it suitable for searching for unknown attack events. A number of popular methods for detecting anomalies with semisupervised algorithms were investigated in study using publicly available data sets on cyber-attacks on power systems to determine the most effective ones. A performance comparison with popular controlled algorithms shows that semi-controlled algorithms are more capable of detecting attack events than controlled algorithms. Our results also show that the performance of semi-supervised anomaly detection algorithms can be further improved by enhancing deep autoencoder model.

show abstract

Section: Methodsmentioning

confidence: 99%