Enabling cybersecurity, situational awareness and resilience in distribution grids through smart devices and deep-learning Grant uri icon


  • The power grid is becoming an easy target to cyberattacks as more generation and control devices operate outside the traditional utilityĆ¢ s administrative domain when employing more distributed generation (DG) units at the grid edge. The situation gets further exacerbated because legacy distribution systems have very limited visibility beyond the substation. In fact, observability of the system cannot be guaranteed due to the limited amount of sensing done currently at the distribution system level. The energy infrastructure is fast evolving from a utility-centric structure to a distributed smart grid, and therefore, utilities must implement a supervisory structure via wide-area communication to remotely control DGs and to enhance their situational awareness. On the other hand, situational awareness will use data from various sources, e.g. embedded sensors in smart inverters, smart transformers, smart meters, phase measurement units, etc., and thus, it will also be susceptible to cyberattacks. A severe cyberattack typically spreads throughout the grid gradually. The detection of such an attack is extremely difficult in its early stages using conventional protection schemes that are normally designed for detection of transients and sudden deviations from normal operation. In this project, a multi-level cyberattack detection approach that combines device-level security using self-learning fleet of smart devices (i.e. PV inverters) with a central anomaly detection system is proposed. At the device-level of this architecture, clusters of smart devices will be considered to detect normal and abnormal operation based on healthy operational regions derived using aggregated reference models (ARM) describing the dynamics of each cluster and a semi-supervised learning process. The project is aimed to design an architecture that has better immunity to cyber-intrusions. The normal operational region of each cluster has boundaries, which are adaptively tuned. The smart devices are used in an active technique to detect abnormal conditions and distinguish those abnormalities from normal transients (e.g., PV partial shading, flickers, etc.) via a deep-learning process............

date/time interval

  • 2020 - 2023