Learning Parameters and Topology in Unbalanced Distribution Grids

The methods proposed in this paper aim at learning the parameters and topology of a distribution grid by making use of available system measurements. The methods that are used rely on the formulation of the underlying physical equations as linear systems and use the resulting problem as a regression problem by applying the ordinary least squares method. Modelling of distribution grids is done by taking assymetrical behaviour caused by single phased loads such as electric vehicles, photovoltaics and heatpumps into account. The developed methods are applied to one medium voltage and one low voltage benchmark distribution grid from the CIGRE networks. Effects of noise on the recovery task are investigated. It can be shown, that given a sufficient amount of measurements, the original system can be recovered. An outlook regarding extensions of the methodology to systems with lower observability is given.