Automated feeder routing for underground electricity distribution networks based on aerial images

Abstract

The process towards a carbon-neutral society requires a substantial amount of investment into electricity distribution networks for integrating more sustainable technologies such as photovoltaic systems, electric vehicles, heat pumps, and others. In order to execute the large number of network expansion measures associated with this effort, strategic and operational planning of distribution networks benefit from higher degrees of automation. A key component in these planning processes is the routing of feeder segments, i.e., the task of finding feasible and cost-efficient paths for underground power lines between two geographic locations, such as residential buildings and transformer stations. In this paper, we present FEEDAIR, a novel and innovative approach to automate feeder routing based exclusively on aerial images. FEEDAIR integrates deep neural networks, which encode geoinformation from aerial images, and A* search, which identifies effective feeder routes based on this encoding. We train and evaluate FEEDAIR using a large data set of more than 7000 real-world electricity distribution network feeder segments. Thereby, we compare FEEDAIR to state-of-the-art approaches from related work and find that the feeder segments generated by FEEDAIR are on par or superior with regard to several characteristics and metrics. FEEDAIR achieves this performance by relying exclusively on aerial images, whereas competing approaches require data on geocoded land cover and estimated construction cost, which is not as widely available and is frequently based on potentially subjective expert opinion.