FDIA Detection Methods on a Navy Smart Grid AMI Data Set using Autoenocoder Neural Networks: A Case Study
Preston Musgrave, Preetha Thulasiraman · 2022
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. In this paper we aim to address anomaly detection in the Navy smart grid using autoencoder neural networks. Our specific focus is on anomalies in the sensor data originating within the advanced metering infrastructure (AMI) of the Navy smart grid. An efficient autoencoder model was developed through benchmarking experiments with open source Modbus data sets. We then used NAVFAC provided AMI data to train the autoencoder model to detect 14 different false data injection attacks (FDIA). Twenty six AMI data features were manipulated to simulate FDIAs. Accuracy, precision and recall scores were used to quantify model performance. We compare the performance of our autoencoder with that of a deeper neural network model to show that a smaller, leaner model can be equally effective in classifying and identifying FDIAs. Our experiments showed that our autoencoder model achieved an average 90%-95% on precision, accuracy and recall scores over multiple combinations of optimizers and activation functions. This work is a case study on the use of unsupervised machine learning methods for data anomaly detection on the Naval smart grid.