A crowdsourcing approach for the inference of distribution grids

Maintaining a complete and up-to-date model of the distribution grid is a challenging task, and the scarcity of open models represents a significant bottleneck for researchers in this area. In this work, we address these challenges by introducing a crowdsourcing framework for the collection of open data on distribution grid devices and an algorithm to infer the topological model of the distribution grids. We use the crowd and smartphones to collect an image and the geographical position of power distribution grid devices. Since power distribution lines are usually underground and cannot be mapped, we use spatial data analytics on the collected data in combination with other open data sources to infer the topology of the distribution grid. This paper describes and evaluates our crowdsourcing and inference approach. To evaluate our approach, we organized and conducted a crowdsourcing campaign to map and infer a sizeable district in Munich, Germany. The results are compared with the ground truth of the distribution system operator. Our field experiments show that using the crowd to recognize power distribution elements, a precision of up to 82% and a recall of up to 65% can be obtained. The numerical evaluation of our inference algorithm demonstrates that the model we inferred based on the acquired official DSO grid dataset achieves a power length accuracy of 88% compared to the ground truth. These results confirm our approach as a practical method to infer real power distribution grid models.