Adaptive Hybrid Deep Learning for Nonstationary IoT Load Forecasting in Smart Grid Demand Response Operations
International Transactions on Electrical Energy Systems · 2026-01 · 2026 (1)
DOI: 10.1155/etep/4447735 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Transactions on Electrical Energy Systems, online erschienen am 15.11.2026. Verfasst von Muhammad Touqeer Hashmi Awan und 5 weiteren Autor:innen. Die Arbeit nennt 43 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Accurate load forecasting is critical for optimizing demand response (DR) programs in smart grid systems, yet existing methods struggle with the nonstationary patterns and concept drift inherent in IoT‐driven energy consumption data. This paper proposes the adaptive hybrid deep learning (AHDL) framework, which couples convolutional, recurrent, and attention pathways through a learnable attention‐based fusion layer that reweights the pathway representations for each input window, together with an ADWIN‐based drift detector for online adaptation to distribution shifts in streaming data. Across four benchmark datasets—UCI Household Power Consumption, GEFCom2014, UK‐DALE, and Pecan Street—AHDL attains the lowest RMSE among twelve models, reducing RMSE by 4.0%–7.4% relative to the strongest baseline (iTransformer) and by 33.1%–37.6% relative to a conventional LSTM; the improvements are significant under Diebold–Mariano tests with Benjamini–Hochberg false discovery rate (FDR) correction ( q < 0.05), although the margin over the three most recent Transformer baselines is narrow. An ablation on UCI Household shows that learned fusion lowers RMSE by 9.0% relative to static concatenation and that removing the Transformer branch causes the largest degradation (21.8%). In a single 30‐day closed‐loop DR dispatch and load frequency control (LFC) simulation on GEFCom2014, AHDL reduces the share of mistimed DR activations from 11.7% to 6.6% (8 versus 15 events; not statistically significant at this sample size) and average frequency deviation by 9.9% relative to iTransformer (28.1% relative to LSTM). Source code, preprocessing pipelines, and exact fold splits will be released in a public repository upon acceptance to facilitate reproducibility.
Zitieren
APA
Awan, M. T. H., Nawaz, A., Khan, M. Y. A., Hassan, M. I. U., Javid, Z., & Holderbaum, W. (2026). Adaptive Hybrid Deep Learning for Nonstationary IoT Load Forecasting in Smart Grid Demand Response Operations. International Transactions on Electrical Energy Systems, 2026(1). https://doi.org/10.1155/etep/4447735
Vancouver
Awan MTH, Nawaz A, Khan MYA, Hassan MIU, Javid Z, Holderbaum W. Adaptive Hybrid Deep Learning for Nonstationary IoT Load Forecasting in Smart Grid Demand Response Operations. International Transactions on Electrical Energy Systems. 2026;2026(1). doi:10.1155/etep/4447735
BibTeX
@article{Awan2026,
author = {Awan, Muhammad Touqeer Hashmi and Nawaz, Aamir and Khan, Muhammad Yousaf Ali and Hassan, Muhammad Ikram Ul and Javid, Zahid and Holderbaum, William},
title = {Adaptive Hybrid Deep Learning for Nonstationary IoT Load Forecasting in Smart Grid Demand Response Operations},
journal = {International Transactions on Electrical Energy Systems},
year = {2026},
volume = {2026},
number = {1},
doi = {10.1155/etep/4447735}
}
EUVIMED