A Machine Learning Approach for Anomaly Detection to Secure Smart Grid Systems

The rapid industrial growth in cyber-physical systems has led to upgradation of the traditional power grid into a network communication infrastructure. The benefits of integrating smart components have brought about security issues as attack perimeter has increased. In this chapter, firstly, the authors train the network on the results generated by the uncompromised grid network result dataset and then extract valuable features by the various system calls made by the kernel on the grid and after that internal operations being performed. Analyzing the metrics and predicting how the call lists are differing in call types, parameters being passed to the OS, the size of the system calls, and return values of the calls of both the systems and identifying benign devices from the compromised ones in the test bed are done. Predictions can be accurately made on the device behavior in the smart grid and calculating the efficiency of correct detection vs. false detection according to the confusion matrix, and finally, accuracy and F-score will be computed against successful anomaly detection behavior.