Topology detection in distribution networks with machine learning

Accurate estimation of the state and topology of the distribution grid is hindered by the limited placement of real-time flow meters and breaker statuses at distribution grid lines. In recent years, increasing presence of smart devices and sensors at households have made measurements of consumption and voltages available at distribution buses. This chapter discusses greedy algorithms to learn the grid topology using voltage measurements collected at a subset of the buses in the distribution grid. The distribution grids are operated in a radial topology. This topological restriction leads to provable trends in voltage second moments (covariances) and enables the design of our learning algorithms. For the case where voltage measurements are available at all grid buses, our framework does not require any additional information related to line impedances of grid lines or consumption statistics at buses to estimate the operational topology. Further in presence of such information, we demonstrate guaranteed topology learning in scenarios with varying fraction of “missing” buses that have no voltage measurements. The efficiency of the algorithms is highlighted by their computational complexity that scales polynomially in the number of grid buses.