Robust Hierarchical Deep Learning Based Intrusion Detector for Smart Grid Attacks on IEC104 Protocol

Abstract

Smart grids play a crucial role in managing and distributing electricity in modern power systems. However, the increased connectivity and reliance on communication within a smart grid also amplify the vulnerability to cyber threats. Machine learning (ML) can potentially revolutionize cybersecurity in smart grids and secure protocols like the IEC61850, a widely adopted international standard for communication networks in electric power systems. However, attackers have begun targeting ML-based intrusion detection systems (IDS) to evade detection using adversarial attacks to mislead the models, which can have severe consequences for smart grid security. This thesis explores the problem of the cyber attacks that intrude on the communication network of smart Grids specifically for the IEC61850- 104 protocol. We propose a machine learning-based intrusion detection framework for the IEC104 protocol. The framework employs an artificial neural network (ANN) to analyze network traffic and identify anomalies indicative of cyber attacks. We also evaluate the robustness of the ANN model against adversarial attacks, including the Fast Gradient Sign Method (FGSM), Projected Gradient Descent (PGD), and Carlini andWagner (CW) attacks. Our work firstly labels and cleans a recently published streaming Industrial Control Systeme (ICS) dataset that captured information from two headers, IEC 104 and IEC Manufacturing Message Specification (MMS). Our results demonstrate that the proposed framework can accurately detect cyber attacks while remaining robust to adversarial attacks. Our proposed framework can enhance the security of smart grids by providing an efficient and resilient IDS that can detect and mitigate cyber attacks, even in the presence of adversarial attacks. This framework can be a valuable tool for power system operators and security professionals to secure smart grids against cyber threats.

Date of Award	2023
Original language	American English
Awarding Institution	HBKU College of Science and Engineering