Short-term photovoltaic power forecasting using time series models: comparative analysis
Journal of Logic and Computation · 02.11.2026 · 36 (8)
DOI: 10.1093/logcom/exag057 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Logic and Computation, online erschienen am 10.11.2026. Verfasst von Noel Freire-Mahía und 5 weiteren Autor:innen. Die Arbeit nennt 45 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
The growing integration of solar photovoltaic systems into electrical grids requires reliable forecasting of photovoltaic power generation. This study compares four short-term forecasting models: gradient-boosted trees, long short-term memory networks, BiLSTM and a transformer encoder-based model. These are applied to hourly photovoltaic power data from a site in Ferrol, northwestern Spain. Permutation-based feature importance is used to interpret all models. To assess whether performance differences are meaningful, each model is retrained over five random seeds and compared pairwise via the Diebold–Mariano test on matched seeds, and further evaluated under four-fold rolling-origin cross-validation. Although the Transformer attains the best average point estimate (RMSE $=103.22\pm 1.16$ W, $\text{R}^{2}=0.8049\pm 0.0044$), no pair of models differs significantly in a majority of seeds, and the between-model spread is smaller than the variation induced by seed or test-window choice. The four architectures—including gradient boosting over engineered lag features—are therefore statistically equivalent rather than rankable, with the residual gap to the Naïve baseline attributed to the stochastic, cloud-driven irradiance variability of this Atlantic climate.
Zitieren
APA
Freire-Mahía, N., Victoria-Cinza, R., Gómez-González, P., GarcÍa-Fischer, A., Michelena, Á., & Jove, E. (2026). Short-term photovoltaic power forecasting using time series models: comparative analysis. Journal of Logic and Computation, 36(8). https://doi.org/10.1093/logcom/exag057
Vancouver
Freire-Mahía N, Victoria-Cinza R, Gómez-González P, GarcÍa-Fischer A, Michelena Á, Jove E. Short-term photovoltaic power forecasting using time series models: comparative analysis. Journal of Logic and Computation. 2026;36(8). doi:10.1093/logcom/exag057
BibTeX
@article{FreireMaha2026,
author = {Freire-Mahía, Noel and Victoria-Cinza, Rafael-Juan and Gómez-González, Pedro-Pablo and GarcÍa-Fischer, AgustÍn and Michelena, Álvaro and Jove, Esteban},
title = {Short-term photovoltaic power forecasting using time series models: comparative analysis},
journal = {Journal of Logic and Computation},
year = {2026},
volume = {36},
number = {8},
doi = {10.1093/logcom/exag057}
}
EUVIMED