Smart meter-based outage detection method for power distribution systems

This paper proposes a new data-driven method for power outage detection. By capturing the changes in data distribution of smart meters (SM), it can detect power outages in partially visible distributed systems. First, a mechanism based on breadth-first search (BFS) is proposed, which decomposes the network into a set of regions to find the location information where power outages are most likely to occur. Then, the SM data for each region, generating a generative adversarial network (GAN), is used in order to extract unsupervised manner implicit temporal behavior under normal conditions. After network training, anomaly scoring technology is used to determine whether the real-time measurement data is the data of a power outage event. Finally, in order to infer the location of a power outage in a multi-area network, a regional coordination process with interdependence be-tween cross-regions is used. At the same time, the concept of entropy is used to provide performance analysis for the algorithm in this paper. This method has been verified on the distribution feeder model with actual SM data. Experimental results show that the algorithm is effective and feasible.