Screening Distribution Networks for Loss Minimum Reconfiguration: Leveraging Demonstration Insights for Statistical Deployment Prioritisation

Abstract

Loss minimisation reconfiguration technology, which aims to reduce power losses by operating sectionalising switches in distribution systems (DSs), plays a critical role in modern power systems. However, implementing this technology requires comprehensive modelling that reflects the physical characteristics of actual distribution networks making its widespread deployment challenging. To effectively disseminate this technology, we propose a statistical screening approach to prioritise DSs based on their expected loss reduction potential. The screening framework is implemented using multiple statistical and machine learning models, including linear regression, support vector regression, random forest, an uncertainty–aware selection strategy and the Bradley–Terry model. Our method leverages these models to estimate the potential for loss reduction from routinely available physical and operational data by system operators using results from a limited number of deployments. Specifically, we analyse the results of a technical demonstration on loss minimisation in distribution networks conducted by TEPCO between 2019 and 2020 and propose a screening approach that accounts for complex spatiotemporal variations in power flows. Numerical experiments over a 20-year deployment horizon demonstrate that the proposed screening framework substantially outperforms random system selection, achieving a cumulative loss reduction that is 369.8% higher than random selection after 20 years while effectively identifying high-priority DSs.