Distribution System Flexibility Characterization: A Network-Informed Data-Driven Approach

TL;DR

This paper introduces a network-informed, data-driven method to efficiently characterize the feasible power output region of distribution systems with DERs, improving accuracy and computational efficiency over existing sampling approaches.

Contribution

It develops a novel sampling approach leveraging network physics and legacy data, and trains a classifier to accurately estimate feasible power regions.

Findings

High-quality data sampling improves feasibility region estimation.

The classifier achieves high accuracy in real-world network validation.

The method reduces computational effort compared to traditional sampling.

Abstract

A distribution system can flexibly adjust its substation-level power output by aggregating its local distributed energy resources (DERs). Due to DER and network constraints, characterizing the exact feasible power output region is computationally intensive. Hence, existing results usually rely on unpractical assumptions or suffer from conservativeness issues. Sampling-based data-driven methods can potentially address these limitations. Still, existing works usually exhibit computational inefficiency issues as they use a random sampling approach, which carries little information from network physics and provides few insights into the iterative search process. This letter proposes a novel network-informed data-driven method to close this gap. A computationally efficient data sampling approach is developed to obtain high-quality training data, leveraging network information and legacy learning experience. Then, a classifier is trained to estimate the feasible power output region with high accuracy. Numerical studies based on a real-world Southern California Edison network validate the performance of the proposed work.