Machine Learning Surrogates for Congestion Prediction and Risk-Constrained Battery Energy Storage Scheduling in PV Integrated Distribution Networks

The increasing integration of distributed photovoltaics (PV) generation introduces bidirectional power flows and additional challenges for congestion management in distribution networks. Effective battery energy storage (BES) scheduling can help address these challenges, provided that network constraints and forecast uncertainty are adequately represented. This study develops a probabilistic congestion prediction framework integrated with risk-constrained day-ahead BES scheduling for an IEEE 33-bus feeder incorporating PV generation and five batteries. Linear support-vector regression (SVR) and elastic-net models predict signed branch powers and bus voltages, while branch currents are reconstructed using circuit relationships. Across 15 test days, current reconstruction reduces mean absolute current error from 3.57 to 0.42 A for SVR and from 3.56 to 0.23 A for elastic net. Congestion-detection recall improves from 0.773 to 0.917 and 0.934, respectively. The SVR model is incorporated into a convex BES scheduler using conditional value-at-risk constraints and leave-one-day-out forecast scenarios. Newton–Raphson verification of 50 daily schedules shows that congestion decreases from 24 to 17 days and from 29 to 19 h, whereas scheduling based on average forecasts leaves the batteries idle. Conducting an additional leave-one-day-out evaluation, which excludes the target day from both scaler fitting and surrogate training, maintains these overall congestion reductions. Nevertheless, two daily peak currents increase, and one new violation occurs. These results demonstrate the potential of network-informed BES scheduling to alleviate congestion and improve the utilization of storage flexibility in PV integrated distribution networks. By addressing operational constraints associated with distributed renewable generation, the proposed approach contributes to the reliable and sustainable operation of distribution networks, while highlighting the limitations imposed by forecast uncertainty.

Authors

Institutions

Publication Details

Journal
Sustainability
Published
2026-11-13
DOI
https://doi.org/10.3390/su182211573
Primary Topic
Optimal Power Flow Distribution
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Machine Learning Surrogates for Congestion Prediction and Risk-Constrained Battery Energy Storage Scheduling in PV Integrated Distribution Networks

Usman Shafique, Marek Reformat, Syed Muhammad Ahsan, Petr Musı́lek
Sustainability
Optimal Power Flow Distribution
article

Machine Learning Surrogates for Congestion Prediction and Risk-Constrained Battery Energy Storage Scheduling in PV Integrated Distribution Networks

Usman Shafique, Marek Reformat, Syed Muhammad Ahsan, Petr Musı́lek
article en

Abstract

The increasing integration of distributed photovoltaics (PV) generation introduces bidirectional power flows and additional challenges for congestion management in distribution networks. Effective battery energy storage (BES) scheduling can help address these challenges, provided that network constraints and forecast uncertainty are adequately represented. This study develops a probabilistic congestion prediction framework integrated with risk-constrained day-ahead BES scheduling for an IEEE 33-bus feeder incorporating PV generation and five batteries. Linear support-vector regression (SVR) and elastic-net models predict signed branch powers and bus voltages, while branch currents are reconstructed using circuit relationships. Across 15 test days, current reconstruction reduces mean absolute current error from 3.57 to 0.42 A for SVR and from 3.56 to 0.23 A for elastic net. Congestion-detection recall improves from 0.773 to 0.917 and 0.934, respectively. The SVR model is incorporated into a convex BES scheduler using conditional value-at-risk constraints and leave-one-day-out forecast scenarios. Newton–Raphson verification of 50 daily schedules shows that congestion decreases from 24 to 17 days and from 29 to 19 h, whereas scheduling based on average forecasts leaves the batteries idle. Conducting an additional leave-one-day-out evaluation, which excludes the target day from both scaler fitting and surrogate training, maintains these overall congestion reductions. Nevertheless, two daily peak currents increase, and one new violation occurs. These results demonstrate the potential of network-informed BES scheduling to alleviate congestion and improve the utilization of storage flexibility in PV integrated distribution networks. By addressing operational constraints associated with distributed renewable generation, the proposed approach contributes to the reliable and sustainable operation of distribution networks, while highlighting the limitations imposed by forecast uncertainty.

SustainabilityVol. 18(22)
University of Alberta (CA), University of Hradec Králové (CZ)
National Science Foundation, Government of Alberta, Natural Sciences and Engineering Research Council of Canada
Affordable and clean energy, Industry, innovation and infrastructure
Optimal Power Flow Distribution
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon