부산대학교 도서관

The energy communications network deployed at Marine Corps Air Station, Miramar, San Diego, is a hybrid system developed to provide remote control and monitoring of distributed energy resources (DER). The network consists of a hardwired system and the integration of a Verizon 5G Non-Stand Alone (NSA) network that provides wireless connectivity from the DERs to the Energy and Water operations center (EWOC) on base. The network connectivity of the DERs exposes them to specific network dynamics, including potential cyber penetration by an adversary. The objective of this paper is to develop a cyber anomaly detection model for this hybrid energy communications network using an auto encoder neural network. The auto encoder classifies distributed denial of service (DDoS) attacks against the network infrastructure. We train the auto encoder model on two traffic data sets: 1) Modbus TCP/IP data from the hardwired network apparatus of the energy communications infrastructure and 2) experimentally generated 5G data that mimics traffic on the energy communications network. We present a foundational approach for detecting anomalous behavior on this network and evaluate the effects of applying various combinations of model configurations to the auto encoder. We highlight the results from the best model configurations that provide the highest level of DDoS prediction accuracy.