Sparse auto-encoder assisted data attack resilient protection scheme for cyber-physical microgrids based on identification of critical sensors

The wider penetration of information and communication technology tools for better monitoring of microgrids has increased their vulnerability to malicious intrusion in the cyber layer through false data injection attacks. Being one of the most significant operations in a cyber-physical microgrid, a line fault protection mechanism is a major target for intruder seeking to disrupt the functionality of the entire microgrid. Through the sensor data fed to the protection algorithm, a fault scenario may be falsified as a healthy scenario and vice-versa. In this regard, this paper aims to develop a protection algorithm for fault detection/classification and faulty section identification with increased robustness to data-attacks. The proposed scheme, which combines critical sensor identification through constrained optimization with multistage classification using a stacked sparse autoencoder deep neural network, not only detects and classifies line faults and identifies faulty sections, but also effectively detects malicious false data injections in sensor information. The inclusion of data acquired from only the critically identified sensors reduces the financial implications related to sensor installation and along with imparting immunity to the protection tasks against cyber intrusions.