Skip to Content
EnergiesEnergies
  • Article
  • Open Access

20 November 2026

27 Pages

A Digital Engineering Framework for Undervoltage Risk Screening in Renewable-Rich Smart-City Distribution Networks

,
,
,
and
1
School of Computing, Engineering and Digital Technologies, Teesside University, Middlesbrough TS1 3BX, UK
2
Department of Electrical, Electronic and Computer Engineering, University of Pretoria, P.O. Box 14679, Hatfield, Pretoria 0028, South Africa
*
Authors to whom correspondence should be addressed.
Energies2026, 19(22), 5613;https://doi.org/10.3390/en19225613 
(registering DOI)
This article belongs to the Special Issue Digital Engineering for Smart and Sustainable Cities

Abstract

Deeper penetration of renewable energy sources is seen as a vital component in the transition toward smarter, cleaner, and more intelligent cities, but injection of renewable energy sources directly into the electrical distribution network can oftentimes have unintended impacts, such as over- or undervoltage. This study explores and evaluates a digital tool for steady-state voltage screening for renewable distribution planning. Benchmark scenarios informed by South African operational records and modelled photovoltaic profiles are used, along with IEEE 33- and 69-bus feeder models, for realistic validation purposes. Eskom indicators and Tshwane/Hatfield load-shedding histories condition 10,000 test scenarios for each of the busses. AC power flow labels minimum voltages below 0.95 p.u.; all 20,000 cases converged. Ridge-logistic regression using seven operational features achieved mean precision–recall areas (PR-AUCs) of 0.9782 and 0.9476 over repeated holdouts. Applying the IEEE-33 model to IEEE-69 achieved PR-AUC 0.9376 and classification recall 0.9553, demonstrating model reuse within a shared scenario-modelling framework rather than independent-distribution generalisation. At a 30% retained-case budget, operational rankings captured 96.58%, 96.84%, and 95.66% of undervoltage cases under IEEE-33 out-of-fold, IEEE-69 out-of-fold, and model-reuse evaluation, respectively. Scenario descriptors only achieved the highest capture, 99.85%, conditional on known scenario-construction inputs. These descriptor-based results are not estimates of screening performance on real feeders; neither feature representation has field validation. The retrospective rankings select 70% fewer cases for subsequent assessment without demonstrating end-to-end acceleration, dynamic stability, or upper-voltage compliance. Urban and smart-city applications remain prospective but positive.

Article Metrics

Citations

Article Access Statistics

Article metric data becomes available approximately 24 hours after publication online.