Abstract
Data integrity attacks on the various meter readings found in smart grid systems can be executed to be undetectable by current detection algorithms used in smart grid systems. These unobservable cyberattacks present a potentially dangerous threat to grid operations. Data integrity attacks that involve the compromise of various meter readings such as voltage and current levels can lead to threats ranging from trivial problems such as energy usage miscalculations to dire consequences resulting from the breakdown of the entire smart grid system through overloading the generators. An efficient detection algorithm to detect these attacks on various sensors embedded in the smart grid system is proposed. Due to manufacturing imperfections, discretizing the sensor readings produces variations in the readings that are unique to each sensor. A fingerprint of this sensor noise (variations in readings) is modeled through the use of machine learning techniques. Under a malicious spoofing attack, the noise pattern deviates from the fingerprinted pattern and hence enabling the proposed detection scheme to identify these attacks. A novel ensemble learning method is used to identify the Intelligent Electronic Device (IED). Experiments are performed on the Electric Power and Intelligent Control (EPIC) testbed. It is shown that a set of IEDs under the different stages of the power generation process can be uniquely identified with an accuracy greater than 90% based on the fingerprint.