Replicating Power Flow Constraints Using Only Smart Meter Data for Coordinating Flexible Sources in Distribution Network

Abstract

The increasing integration of distributed energy resources necessitates effective coordination of flexible sources within distribution networks. Traditional model-based approaches require accurate topology and line parameters, which are often unavailable. Neural constraint replication can bypass this requirement, but it relies on complete nodal and branch measurements. However, in practice, only partial buses are monitored, while branches often remain unmeasured. To address this issue, this paper proposes a topology identification-incorporated neural constraint replication to replicate power flow constraints with only partial nodal measurements. Utilizing the additive property of line parameters, we develop a recursive bus elimination algorithm to recover topology and line impedance from power injection and voltage measurements on limited buses. We then estimate missing voltage and branch flow measurements based on the recovered model information. By combining observed and estimated measurements to construct training sets, we train neural networks to replicate voltage and branch flow constraints, which are subsequently reformulated into mixed-integer linear programming forms for efficient solving. Monte-Carlo simulations on various test systems demonstrate the accuracy and computational efficiency of the proposed method, even with limited nodal measurements.