Date(s) - 16 Oct 2019
1:10 PM - 2:00 PM
3043 ECpE Building Addition
Speaker: Vivek Kumar Singh, ECpE Graduate Student
Adviser: Manimaran Govindarasu
Title: Anomaly Detection for Wide-Area Protection in Smart Grid
Abstract: Cyber physical security (CPS) research for the smart grid is currently one of the nation’s top research and development (R&D) priorities. The existing vulnerabilities in the legacy grid infrastructure make it particularly susceptible to different types of cyberattacks. Wide-area protection scheme (WAPS), also known as a remedial action scheme (RAS) or system protection scheme (SPS), relies on the interconnected networks and data sharing devices that are exposed to a multitude of vulnerabilities. This talk discusses a novel approach to develop a unified anomaly detection system (ADS) by leveraging the information obtained from PMU measurements from different substations. In particular, we have applied the state-of-the-art machine learning technique, such as decision tree (DT), to detect anomalies, including cyberattacks and line faults with an objective to develop the attack and fault-resilient WAPS. Further, we have shown the performance evaluation of proposed approach using network (cyber) based federation testbed in terms of network latency, communication bandwidth, and accuracy rate. In general, the cyber physical system (CPS) federated testbed works as a driving force to enable the pipeline from state-of-the art research work through the transition to industry by experimental testing and validation. We have developed a cyber (network) federated testbed by utilizing the resources available at the Iowa State University Power Cyber (ISU PCL) Laboratory to emulate the substation networks; and the US Army Research Laboratory (ARL); to emulate the regional control center network. Further, by incorporating the knowledge and observations during the testbed experiments, we have demonstrated that the proposed decision tree-based anomaly detection algorithm is showing consistent performance as compared to other machine learning classifiers. We have also computed the maximum latency for incoming synchrophasor data packets, which is well within the timing requirement of overloading based WAP scheme. The same approach presented and testbed here for developing ADS could be applied in other wide area applications, such as wide-area voltage control, oscillation monitoring, etc.