Detection of Cyber-Attacks in Smart Power Transmission System using Mathematical Morphology and Autoencoder

Cyber-security of the transmission protective relays is essential for the safety and reliable operation of the smart grid. The increased adoption of the IoT-based smart relays and lack of encryption in the IEC 61850 protocol-based substation automation system has raised the risk of false data injection attacks on the protection relays. The attackers of the power system can easily trip the potential transmission lines by tampering with the measurement data that is vital for the protective relay operation. This work proposes a mathematical morphology and deep learning autoencoder-based technique to detect such attacks on overcurrent relays. The mathematical morphology extracts the most significant features of the measured data, which are then supplied as an input to the autoencoder, distinguishing the attacked measurements from the actual fault measurements. The autoencoder distinguishes the maliciously injected faulty data from the actual fault data by calculating the reconstruction error. To validate the proposed technique, a two-bus benchmark system is used. The result shows that the proposed method is capable of attack detection with excellent precision and satisfactory recall rates for the overcurrent protection scheme. Also, the effect of the severity of different attacks on the performance of the proposed technique is analyzed and is being found that the performance of the proposed technique remains satisfactory even during low severity conditions.