Machine learning classifier-based detection of cyber-attack on power system: Comparative analysis

In digital substations, organized cyber-attacks can severely impact the protection and security of the system and may cause generation and line disconnections, triggering cascading failures in the modern power system. In this study, the modeling of cyber-attack and the formulation of the detection problem as the classification problem for the machine learning classifier have been discussed. Two types of classification binary and multi-class have been taken up for analysis. Two publicly available datasets were used to train the classifier. Comparative analysis was performed in detail for different classifiers in different attack scenarios using Python, Matlab, and WEKA 3 software. The standard IEEE-24 bus system has been used for the dataset generation. It has been observed from the results that the Random forest Classifier works best in most of the cases followed by the Multi-Layer Perceptron Classifier.