Physics-Guided GWO-SVR for Cross-Day Power Estimation of Crystalline PV Modules Under Limited Data
Jurnal JEETech · 11.11.2026 · 7 (2) · 282-294
DOI: 10.32492/jeetech.v7i2.7209 ↗Kurz gefasst
Fachartikel in der Zeitschrift Jurnal JEETech, online erschienen am 11.11.2026. Verfasst von Giovanni Dimas Prenata und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).
Abstract (Originalsprache)
Photovoltaic power estimation under limited field data is challenging because module output is influenced by rapidly changing irradiance, temperature, module technology, and electrical operating characteristics. Conventional random data splitting may also produce overly optimistic results when measurements from the same day are distributed across the training and testing sets. This study proposes a Physics-Guided Grey Wolf Optimizer–Support Vector Regression model, referred to as PG-GWO-SVR, for cross-day power estimation of monocrystalline and polycrystalline photovoltaic modules. The dataset comprised 63 field observations obtained from three nominally 100 W modules over four measurement days in Surabaya, Indonesia. The proposed feature set integrated irradiance, module temperature, measurement time, technology category, normalized open-circuit voltage and short-circuit current, datasheet-derived electrical ratios, standard-test-condition fill factor, and a normalized physics-based power proxy. GWO optimized the penalty coefficient, radial-basis-function parameter, and ε-insensitive margin. Five approaches were evaluated using nested leave-one-day-out validation and ten independent optimization runs: a Physics baseline, fixed-parameter environmental SVR, GWO-SVR-Env, PG-GWO-SVR, and PG-Residual-GWO-SVR. PG-GWO-SVR achieved the best mean performance, with an RMSE of 8.8600±0.9931 W, MAE of 7.0925±0.6482 W, R2 of 0.8205±0.0440, nRMSE of 12.5585±1.4077%, and WAPE of 21.8758±1.9991%. It reduced RMSE by 51.89% relative to the Physics baseline and by 24.96% relative to GWO-SVR-Env. Averaging the ten independent predictions further improved RMSE to 8.2841 W and R2 to 0.8449. These results demonstrate that combining module-specific physical knowledge with metaheuristic-optimized regression improves cross-day PV power estimation when only limited multi-module field measurements are available. Geographic and climatic transferability beyond these four field days remains unverified
Zitieren
APA
Prenata, G. D., & Ridho’i, A. (2026). Physics-Guided GWO-SVR for Cross-Day Power Estimation of Crystalline PV Modules Under Limited Data. Jurnal JEETech, 7(2), 282-294. https://doi.org/10.32492/jeetech.v7i2.7209
Vancouver
Prenata GD, Ridho’i A. Physics-Guided GWO-SVR for Cross-Day Power Estimation of Crystalline PV Modules Under Limited Data. Jurnal JEETech. 2026;7(2):282-294. doi:10.32492/jeetech.v7i2.7209
BibTeX
@article{Prenata2026,
author = {Prenata, Giovanni Dimas and Ridho’i, Ahmad},
title = {Physics-Guided GWO-SVR for Cross-Day Power Estimation of Crystalline PV Modules Under Limited Data},
journal = {Jurnal JEETech},
year = {2026},
volume = {7},
number = {2},
pages = {282--294},
doi = {10.32492/jeetech.v7i2.7209}
}
EUVIMED