Heuristic graph partitioning with preferred cluster sizes and application to the generation of realistic distribution grid topologies

In distribution power systems studies, detailed data about grid topologies are often difficult to find. Apart from published benchmark grids, accurate models of real world distribution grids in cities are not typically kept and mantained by DSOs. The emerging need for testbeds for electrical simulations interfacing with actively controlled agents on realistic grids can be met with tools for generating grid topologies. In this article, we propose a heuristic graph partitioning technique that takes into account desired cluster sizes and preferential routing, and is thus applicable to the problem of automatically partitioning communities in subgroups that can be served by MV-LV substations of commonly available size. The technique is benchmarked on test graphs and is then applied to synthetic grid topology generation from publicly available geographical data of the town of Arosio, Switzerland.