Trend-Aware Synthetic Time-Series Generation for Microgrid Design
IEEE Access · 2026 · 14 · 86442-86457
DOI: 10.1109/access.2026.3699529 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Pooriya Motevakel und 2 weiteren Autor:innen. Die Arbeit nennt 57 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Universitat Politècnica de València under the Sustainability Fund for Research, Development, and Innovation.
Abstract (Originalsprache)
High-resolution time-series data are crucial for effective microgrid planning; however, such datasets are often incomplete or unavailable. This study presents a lightweight and transparent framework for generating realistic synthetic time series of energy variables. The method combines: 1) probabilistic modeling to fit statistical distributions to segmented hourly data and 2) a trend-informed correction mechanism that preserves hour-to-hour continuity based on observed transition patterns. The proposed framework is validated using high-resolution datasets from a rural site in Spain, first against historical records from 2019 to 2023 used for model construction, and then against an independent dataset from 2023 to 2024 to assess its generalization to unseen data. Results show that finer temporal segmentation improves local statistical accuracy, while the trend-informed correction significantly reduces abrupt transitions, leading to smoother, more realistic temporal behavior. Across the conducted evaluations, the refined framework achieves improvements of up to approximately 30% in selected metrics and variables. In addition, the proposed approach is benchmarked against representative classical and neural time-series models, demonstrating competitive point-wise accuracy under a consistent evaluation setting, with benchmark-case percentage prediction error reductions of up to 35% across the evaluated variables. Consistent improvements across validation horizons confirm the robustness of the method. Rather than relying on heavily tuned or fully data-driven deep learning approaches, the proposed framework combines classical probabilistic modeling with a trend-based correction layer, improving temporal continuity while maintaining relatively low computational cost. The resulting synthetic data provides temporally coherent scenarios suitable for microgrid sizing, storage optimization, and resilience-oriented analysis.
Förderung
- Universitat Politècnica de València under the Sustainability Fund for Research, Development, and Innovation
Zitieren
APA
Motevakel, P., Roldán-Blay, C., & Roldán-Porta, C. (2026). Trend-Aware Synthetic Time-Series Generation for Microgrid Design. IEEE Access, 14, 86442-86457. https://doi.org/10.1109/access.2026.3699529
Vancouver
Motevakel P, Roldán-Blay C, Roldán-Porta C. Trend-Aware Synthetic Time-Series Generation for Microgrid Design. IEEE Access. 2026;14:86442-86457. doi:10.1109/access.2026.3699529
BibTeX
@article{Motevakel2026,
author = {Motevakel, Pooriya and Roldán-Blay, Carlos and Roldán-Porta, Carlos},
title = {Trend-Aware Synthetic Time-Series Generation for Microgrid Design},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {86442--86457},
doi = {10.1109/access.2026.3699529}
}
EUVIMED