Low voltage user phase reconfiguration as a planning problem

TL;DR

This paper introduces static phase reconfiguration methods for low voltage distribution networks to improve phase balance, addressing the limitations of dynamic reconfiguration and providing practical solutions for grid operators.

Contribution

It proposes and compares three static reconfiguration methods—MINLP, MIQP, and GA—for effective phase balancing in LV networks, with MIQP showing superior scalability and performance.

Findings

MIQP efficiently mitigates phase imbalance.

MIQP outperforms MINLP and GA in scalability.

Static reconfiguration is a practical alternative to dynamic methods.

Abstract

Considerable levels of phase imbalance in low voltage (LV) distribution networks imply that grid assets are suboptimally utilized and can cause additional losses, equipment failure and degradation. With the ongoing energy transition, the installation of additional single-phase distributed energy resources may further increase the phase imbalance if no countermeasures are taken. Phase reconfiguration is a cost-effective solution to reduce imbalance. However, dynamic reconfiguration, through real-time phase swapping of loads using remotely controlled switches, is often impractical because these switches are too costly for widespread installation at LV users. Approaching phase reconfiguration as a planning problem, i.e. static reconfiguration, is an underaddressed but promising alternative. Effective static approaches that allow appropriate imbalance objectives are currently lacking. This paper presents reliable and expressive static phase reconfiguration methods that grid operators can easily integrate into routine maintenance for effective phase balancing. We present and compare three static methods, an exact mixed-integer nonlinear formulation (MINLP), a mixed-integer quadratic approximation (MIQP), and a genetic algorithm (GA), each supporting different imbalance objectives. The MIQP approach, despite using proxy objectives, efficiently mitigates the different types of imbalance considered, and outperforms both MINLP and GA in scalability and consistency.