ANOMALY DETECTION FOR THE NAVAL SMART GRID SYSTEM USING AUTOENCODER NEURAL NETWORKS

In 2019, the Naval Facilities Engineering Command (NAVFAC) deployed its first smart grid infrastructure in Norfolk, VA, enabling shore commands to meet energy goals set by the secretary of the Navy. However, with increased functionality and control comes increased vulnerability to malicious cyber activity. This research aims to address anomaly detection using an autoencoder neural network as an intrusion detection mechanism on the NAVFAC smart grid. We built and experimented with multiple autoencoder structures to identify an optimal model that provides the best results in terms of precision, recall, and accuracy for the data sets used. We trained our autoencoder on NAVFAC-provided advanced metering infrastructure (AMI) data. We used the NAVFAC smart grid data set to simulate 14 different false data injection attacks (FDIA). Our experiments, performed with Python and TensorFlow, showed that an autoencoder is an effective instruction detection system (IDS) when the threshold is tuned correctly. Moreover, our results show that the activation function and optimizer used may affect performance. Thus, the “best” autoencoder depends on the customer’s needs and the threat environment.ONR-ESTEP (Arlington, VA 22203)Lieutenant, United States NavyApproved for public release. Distribution is unlimited