Intrusion detection for cybersecurity in the smart grid distribution

Abstract

This thesis studies the smart grid. The main entity that this intrusion detection system will cover will be the distribution system. A simulated model is built to attain data needed for the study. It proposes a new model for an intrusion detection system for smart grids. The data generated from the simulated system is used to build the Intrusion Detection System. Various linear and non-linear models are tested and analyzed to show the best fit for the IDS to be most efficient. A new proposed algorithm based on machine learning is introduced. For the most efficiency and accuracy, an IDS is built based on a combination of machine learning with a customized cost function. The measured performance show that the LSTM with varying epsilon present the best result for it has the most precision and the least errors among all the cost functions studied.