Eee projects on cyber attack detection in smart inverter using machine learning — Simulation & Power Modeling

Abstract — Eee projects on cyber attack detection in smart inverter using machine learning

including scaling attacks, additive attacks, and their combinations, to test the selected approaches. Our results indicate that GNN-based methods outperform k-means and autoencoder in detection. Additionally, GNNs show promise in accurately localizing attacks for simple scenarios, although they still face challenges in more complex cases, especially ones that involve combinations of scaling and additive attacks.

Analytical & State-Space Control Methodology

Cyber-attacks on power systems can have devastating effects, disrupting essential services and causing significant economic losses . As power systems become increasingly interconnected and digitized, they become more vulnerable to sophisticated cyber-attacks, as demonstrated by real cyber-attacks [ 2 , 3 ] . Detecting cyber-attacks on power systems in a timely manner is of paramount importance because it allows for swift mitigation measures, minimizing the impact of the attack and ensuring the resilience and reliability of the power grid [ 4 , 5 ] . Increasing cloud-based communication and control systems, fast-evolving ransomware threats, and the convergence of information technology (IT) and operations technology (OT) systems were identified as among the emerging cybersecurity challenges faced by the power grid in a recent report by [ 7 , 8 ] , or a hybrid (combined IT/OT) intrusion detection but with static, pre-determined, OT rules [ 9 , 10 ] . While rule-based cyber intrusion detection engines rely on operators’ OT knowledge to establish a ‘baseline’, these suffer from a lack of adaptability, especially as new energy resources with evolving controls and communications are integrated into the grid at an increasing rate .

Simulink Waveforms & Power Quality Findings

There has been various research focusing on OT-based automated cyber-attack detection for power systems using different algorithmic (non-rule-based) methods, including model-based, machine learning, and deep learning methods. A broad class of the cyber-detection methods use physics-based models with an outlier detection algorithm [ 12 , 13 , 14 ] . Different machine learning-based methods, on the other hand, have deployed time-series modeling approach , decision trees , ensemble methods , or unsupervised k-nearest neighbor approach . More recently, various deep learning-based intrusion detection approaches are being investigated, e.g., the Convolutional Neural Network (CNN) approach , deep reinforcement learning .

Scope & Practical Applications

• High-efficiency DC-DC converters and isolated resonant topologies (LLC / DAB)
• Sinusoidal and Space Vector PWM inverters for grid integration & industrial motor drives
• MOSFET / IGBT gate driver optimization with active Miller-clamp shoot-through protection
• Power quality conditioning with Total Harmonic Distortion (THD) strictly compliant with IEEE 519
• Complete Simscape .SLX model files, parameter calculation scripts, and IEEE dissertation reports