Zhu, Y. orcid.org/0009-0000-0909-5876, Li, K. orcid.org/0000-0001-6657-0522 and Zhang, L. orcid.org/0000-0002-4535-3200 (2026) Stacked-Ensemble Forecasting for Resilient Operation of Base Station Microgrids Under Communication Failures. IEEE Internet of Things Journal. ISSN: 2327-4662
Abstract
The resilient and economical operation of cellular base station microgrids (BSMGs) depends critically on accurate forecasting of renewable generation and load, particularly under communication link failures that disrupt real-time data exchange. However, forecasting model performance varies significantly with the underlying stability characteristics of historical datasets—a factor rarely considered in existing BSMG studies. This paper addresses this gap by proposing a stacked-ensemble forecasting framework that first assesses dataset stability and then guides model selection accordingly. Five complementary metrics, namely the augmented Dickey–Fuller (ADF) test, seasonal strength, approximate entropy (ApEn), the Hurst exponent, and trend-to-residual variance ratio (TRVR), are employed to characterize the stability, periodicity, and randomness of each dataset. These heterogeneous indicators are integrated into a unified stability score using grey relational analysis (GRA) enhanced by the entropy weight method (EWM), criteria importance through inter-criteria correlation (CRITIC), and the triangular fuzzy analytic hierarchy process (TF-AHP). Based on the resulting stability tier, a four-layer ensemble pool comprising three statistical models, three machine learning (ML) algorithms, two deep learning (DL) architectures, and two reinforcement learning (RL) schemes is selectively assembled. A random forest (RF) stacking model fuses their outputs to maximize forecast accuracy while maintaining low complexity. Simulation results demonstrate that the proposed stability-guided approach reduces weekday median root mean square error (RMSE) by 18.5% and improves median coefficient of determination (R2) by 16.9% compared to baseline methods. Under two-link and three-link communication failure scenarios, the RF-Stacking-enabled dispatch reduces fault-induced transaction cost surges by approximately 15% and up to 40%, respectively. These findings confirm that systematically linking data stability assessment to ensemble model selection enhances both forecasting accuracy and operational resilience in communication-constrained BSMG environments.
Metadata
| Item Type: | Article |
|---|---|
| Authors/Creators: | |
| Copyright, Publisher and Additional Information: | This is an author produced version of an article published in Transportation Research Part C: Emerging Technologies, made available via the University of Leeds Research Outputs Policy under the terms of the Creative Commons Attribution License (CC-BY), which permits unrestricted use, distribution and reproduction in any medium, provided the original work is properly cited. |
| Keywords: | Base station microgrids, stacked ensemble forecasting, machine learning, random forest, dataset stability assessment. |
| Dates: |
|
| Institution: | The University of Leeds |
| Academic Units: | The University of Leeds > Faculty of Engineering & Physical Sciences (Leeds) > School of Electronic & Electrical Engineering (Leeds) > Institute of Communication & Power Networks (Leeds) |
| Date Deposited: | 16 Nov 2026 14:41 |
| Last Modified: | 16 Nov 2026 14:41 |
| Status: | Published online |
| Publisher: | Institute of Electrical and Electronics Engineers (IEEE) |
| Identification Number: | 10.1109/jiot.2026.3752576 |
| Open Archives Initiative ID (OAI ID): | oai:eprints.whiterose.ac.uk:247942 |
Download
Filename: Yingqi's second journal paper_Final Version.pdf
Licence: CC-BY 4.0

CORE (COnnecting REpositories)
CORE (COnnecting REpositories)