Topology Change Aware Data-Driven Probabilistic Distribution State Estimation Based on Gaussian Process

This paper addresses the distribution system state estimation (DSSE) with unknown topology change. A specific kernel that can transfer across tasks is adopted to find relevant patterns from samples under different topologies and induce knowledge transfer. This enables the proposed method to achieve effective inductive reasoning when only limited data are available under a new topology. The Bayesian inference inherently allows us to quantify the uncertainties of the DSSE results. Comparative results with other methods on IEEE test systems demonstrate the improved accuracy and robustness against topology change.