Topology Identification for Distribution Power Grid with Distributed Generation: A Graph Signal Processing Approach

This paper proposes a graph signal processing (GSP) approach for distributed generation (DG) penetrated-distribution power grid topology identification, using only voltage magnitude measurements from smart meters. The approach employs a twophase process: a transform-learning phase to jointly determine an optimal orthonormal basis that sparsifies the voltage signals, and a topology inference and refinement stage that uses convex optimization to infer the graph Laplacian, which is subsequently refined into a sparse, connected adjacency matrix. The proposed approach is validated via simulations on the IEEE 33-bus system with multiple DGs. To confirm its practical feasibility on the edge side, the algorithm is also implemented in C++ and deployed on a resource-constrained smart meter. Results demonstrate that the proposed approach is effective for topology identification, with the meter implementation achieving an accuracy comparable to the PC-based result and an execution time around 30 seconds.