A dimensional transformation scheme for power grid cyber event detection

As the technologies used for the safe and efficient delivery of power become more sophisticated, the amount of system state parameters being recorded increases. This data not only provides an opportunity for monitoring and diagnostics of the power system but also creates an environment wherein security can be maintained. Being able to extract relevant information from this pool of data in a reasonable amount of time is one of the key challenges still yet to be obtained in the smart grid. New power grid security applications can be created that use the statistical patterns in the reported data as a metric for security. Anomalies detected by the developed security metrics can then be alerted upon as a possible cyber intrusion. This article is an examination into the utilization of principal component analysis along with a Naive-Bayes classifier for the identification of spoofed power system state parameters. Examination targets a 5 Bus power system with results indicating successful classification of simulated cyber attacks or true positive classification at a rate of 92%. These findings also indicate a dependency on which variables were compromised, providing an initial formalization into the stealthiness of state estimation attacks.