Learning Topology of Distribution Grids using only Terminal Node Measurements
For power system operators, this enables fast estimation of grid topology with limited real-time measurements, addressing a critical problem in distribution grid management.
The paper proposes a learning algorithm that uses measurements only from terminal (leaf) nodes to estimate the radial structure of distribution grids, achieving structure learning under minimum permissible observability where only half of the nodes are measured.
Distribution grids include medium and low voltage lines that are involved in the delivery of electricity from substation to end-users/loads. A distribution grid is operated in a radial/tree-like structure, determined by switching on or off lines from an underling loopy graph. Due to the presence of limited real-time measurements, the critical problem of fast estimation of the radial grid structure is not straightforward. This paper presents a new learning algorithm that uses measurements only at the terminal or leaf nodes in the distribution grid to estimate its radial structure. The algorithm is based on results involving voltages of node triplets that arise due to the radial structure. The polynomial computational complexity of the algorithm is presented along with a detailed analysis of its working. The most significant contribution of the approach is that it is able to learn the structure in certain cases where available measurements are confined to only half of the nodes. This represents learning under minimum permissible observability. Performance of the proposed approach in learning structure is demonstrated by experiments on test radial distribution grids.