Topology Change Estimation in Power Distribution Networks: A Machine Learning Approach

A critical component in distribution grid monitoring and control is the knowledge of the distribution network and the capability to detect and localize any topology changes, including line outages and additions, in near real-time. Traditionally, such a requirement is achieved with the breaker status and smart meter measurements. However, the slower data rate and inadequate wireless protection of such measurements degrade the situational awareness and make the system vulnerable to cyber-attacks. In this context, phasor measurement units (PMUs) are a robust and secure alternative, having a faster data rate and an integrated advanced cyber-protection mechanism. In this article, we propose a comprehensive solution to (1) detect physical topology changes, such as branch outages and additions, (2) detect change instants, and (3) localize the changes, entirely from phasor measurements, and without any prior information about the network. Furthermore, our proposed strategy can detect and localize multiple branch outages or additions, or a combination of both at the same time. We employ the expectation-maximization technique for change detection, and the outcome of the same is utilized in (1) a Kernel-density estimator module to detect change instants and (2) a data-driven topology learning technique to localize the changes. We perform real-time hardware-in-the-loop experiments on a modified IEEE 33-bus, 226-bus, and 1262-bus network emulated on a real-time simulator. Our experimental validation corroborates the utility and scalability of our proposed methodology.