EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

VG-BiLSTM: Modeling Feature Variability for Robust PV Power Prediction

Are Sambasiva Rao, K. Dhana Sree Devi

Engineering, Technology & Applied Science Research · 04.04.2026 · 16 (2) · 33405-33410

DOI: 10.48084/etasr.16956 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Engineering, Technology & Applied Science Research, online erschienen am 04.04.2026. Verfasst von Are Sambasiva Rao und 1 weiteren Autor:innen. Die Arbeit nennt 22 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Photovoltaic (PV) power generation oscillates across multiple timescales due to changing weather and environmental conditions. These fluctuations have a huge impact on the stability and operational costs of the grid and reduce its effective use. Motivated by the need for reliable grid integration, this study emphasizes the importance of explicitly modeling variability in PV features to improve the robustness of solar power prediction models. Conventional approaches often neglect short-term fluctuations in key meteorological features, which can lead to degraded model performance and accuracy. This study presents the Variability Gated Bidirectional Long Short-Term Memory (VG-BiLSTM) framework, which explicitly models both features and variability using descriptors to predict short-term PV power output. The proposed method computes feature descriptors, the rolling standard deviation, ramp rate, and coefficient of variation for each feature, and employs a variability attention module to dynamically weight them. A gating network combines raw and variability representations into a single latent embedding, which is then input into a Bidirectional LSTM for temporal modeling. Experimental results show that VG-BiLSTM outperforms baseline RNN, GRU, LSTM, and their bidirectional versions, achieving much lower RMSE and nRMSE in different sky types. The model also achieved the best error levels across regions, with RMSE and NRMSE reduced to 21.6 W and 20.4%, respectively, outperforming other models.

Zitieren

APA

Rao, A. S., & Devi, K. D. S. (2026). VG-BiLSTM: Modeling Feature Variability for Robust PV Power Prediction. Engineering, Technology & Applied Science Research, 16(2), 33405-33410. https://doi.org/10.48084/etasr.16956

Vancouver

Rao AS, Devi KDS. VG-BiLSTM: Modeling Feature Variability for Robust PV Power Prediction. Engineering, Technology & Applied Science Research. 2026;16(2):33405-33410. doi:10.48084/etasr.16956

BibTeX
@article{Rao2026,
  author = {Rao, Are Sambasiva and Devi, K. Dhana Sree},
  title = {VG-BiLSTM: Modeling Feature Variability for Robust PV Power Prediction},
  journal = {Engineering, Technology & Applied Science Research},
  year = {2026},
  volume = {16},
  number = {2},
  pages = {33405--33410},
  doi = {10.48084/etasr.16956}
}
Kostenlos starten →