A Safe and Explainable Bayesian Energy Management Framework with Probabilistic Safety Shielding for EV-Integrated Renewable Microgrids

The increasing integration of renewable energy sources and electric vehicle (EV) charging, which is central to the transition towards sustainable energy systems, creates operational challenges for microgrids, where generation uncertainty, demand variability, and charging congestion can lead to energy imbalances and system overloads. This study proposes a Bayesian energy management framework integrating probabilistic forecasting, explainable charging-decision modelling, and safety-constrained action filtering for renewable microgrids with EV charging. The framework combines Bayesian Ridge regression for demand forecasting, Gaussian Process regression for photovoltaic (PV) generation forecasting, Bayesian charging classification, SHapley Additive exPlanations (SHAP), and an independent Safety Shield to assess the operational feasibility of candidate charging decisions. On an independent test set of 150 observations, the demand forecasting model achieved an R2 of 0.8546 and 90.00% empirical coverage of nominal 90% prediction intervals. The charging classifier attained 98.00% agreement with reference decisions and a Brier score of 0.0132. However, the unshielded controller recorded 139 overload events, corresponding to an observed overload rate of 92.67% (95% confidence interval: 87.35–95.86%). With the Safety Shield enabled, no overload events were observed in the test set, yielding an observed rate of 0% (95% confidence interval: 0–2.50%). This reduction was accompanied by a decrease in decision agreement to 64.00% and reported EV charging curtailment of 1567.33 kWh, illustrating the operational trade-off between overload avoidance and charging availability. Both shielded and unshielded configurations reported a daily operating cost of $210.44, compared with $ 476.73 for the rule-based benchmark, equivalent to an approximately 55.86% reduction based on the reported costs. However, this economic comparison excludes the cost of curtailed EV charging, and the absence of observed overloads does not establish zero population-level risk. Overall, the results demonstrate that high charging-decision agreement alone is insufficient to ensure physical feasibility and support the integration of explicit safety filtering with uncertainty-aware prediction and interpretable decision-making for the safe and sustainable operation of EV-integrated renewable microgrids.

Authors

Institutions

Publication Details

Journal
Sustainability
Published
2026-11-16
DOI
https://doi.org/10.3390/su182211663
Primary Topic
Microgrid Control and Optimization
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Safe and Explainable Bayesian Energy Management Framework with Probabilistic Safety Shielding for EV-Integrated Renewable Microgrids

Abayomi A. Adebiyi, Mayowa Samuel Olokun
3 links
Sustainability
Microgrid Control and Optimization
article

A Safe and Explainable Bayesian Energy Management Framework with Probabilistic Safety Shielding for EV-Integrated Renewable Microgrids

Abayomi A. Adebiyi, Mayowa Samuel Olokun
article en

Abstract

The increasing integration of renewable energy sources and electric vehicle (EV) charging, which is central to the transition towards sustainable energy systems, creates operational challenges for microgrids, where generation uncertainty, demand variability, and charging congestion can lead to energy imbalances and system overloads. This study proposes a Bayesian energy management framework integrating probabilistic forecasting, explainable charging-decision modelling, and safety-constrained action filtering for renewable microgrids with EV charging. The framework combines Bayesian Ridge regression for demand forecasting, Gaussian Process regression for photovoltaic (PV) generation forecasting, Bayesian charging classification, SHapley Additive exPlanations (SHAP), and an independent Safety Shield to assess the operational feasibility of candidate charging decisions. On an independent test set of 150 observations, the demand forecasting model achieved an R2 of 0.8546 and 90.00% empirical coverage of nominal 90% prediction intervals. The charging classifier attained 98.00% agreement with reference decisions and a Brier score of 0.0132. However, the unshielded controller recorded 139 overload events, corresponding to an observed overload rate of 92.67% (95% confidence interval: 87.35–95.86%). With the Safety Shield enabled, no overload events were observed in the test set, yielding an observed rate of 0% (95% confidence interval: 0–2.50%). This reduction was accompanied by a decrease in decision agreement to 64.00% and reported EV charging curtailment of 1567.33 kWh, illustrating the operational trade-off between overload avoidance and charging availability. Both shielded and unshielded configurations reported a daily operating cost of $210.44, compared with $ 476.73 for the rule-based benchmark, equivalent to an approximately 55.86% reduction based on the reported costs. However, this economic comparison excludes the cost of curtailed EV charging, and the absence of observed overloads does not establish zero population-level risk. Overall, the results demonstrate that high charging-decision agreement alone is insufficient to ensure physical feasibility and support the integration of explicit safety filtering with uncertainty-aware prediction and interpretable decision-making for the safe and sustainable operation of EV-integrated renewable microgrids.

SustainabilityVol. 18(22)
Durban University of Technology (ZA)
Affordable and clean energy, Industry, innovation and infrastructure, Sustainable cities and communities, Climate action
Microgrid Control and Optimization
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