Machine Learning Based Intrusion Detection Scheme to Detect Replay Attacks in Smart Grid

Smart grid involves bi-directional communication among producer and consumer. Further, advanced metering infrastructure in smart grid facilitates transformation of consumer as prosumer through intelligent scheduling. As smart grid incorporates automatic control, communication techniques and information technology, protection against cyber attack is mandate. Though variety of cyber attacks is feasible in smart grid, the proposal addresses the detection mechanism for one of the predominant attacks namely replay attack using machine learning. In replay attack, legitimate node is compromised by the attacker and hence, the node will be sending older messages repeatedly at frequent intervals. In such scenario, the receiving node is busy and is not able to provide service for other legitimate nodes. Further, the older message may cause malfunctioning of control system which is undesirable. To evade this, machine learning based intrusion detection scheme for replay attack detection is proposed. The proposal is implemented in real time using voltage and current sensors, Arduino, Zigbee and Raspberry Pi and the data is monitored through Thing Speak. Further, the data from the Zigbee is collected through XCTU and is fed as input to the pool of machine learning algorithms implemented in MATLAB. The time stamp of the node is used to classify the data either as normal or malicious. Among the algorithms, fine Gaussian Support Vector Machine exhibits better performance metrics when compared with rest of the algorithms which clearly depicts the suitability of the model.