Generalized versus Panel-specific Machine Learning Models for Photovoltaic Power Estimation Under Limited Multi-Module Field Data

Giovanni Dimas Prenata; Ahmad Ridho'i

doi:10.25170/jurnalelektro.v19i2.8268

Authors

Giovanni Dimas Prenata Department of Electrical Engineering, Faculty of Intelligent Electrical and Informatics Technology, Universitas 17 Agustus 1945 Surabaya, INDONESIA
Ahmad Ridho'i Department of Electrical Engineering, Faculty of Intelligent Electrical and Informatics Technology, Universitas 17 Agustus 1945 Surabaya, INDONESIA

DOI:

https://doi.org/10.25170/jurnalelektro.v19i2.8268

Keywords:

Cross-day Validation, Generalized Modeling, Panel-specific Modeling, Photovoltaic Power Estimation, Random Forest, Ridge Regression

Abstract

Limited multi-module field data creates a trade-off between generalized models that pool observations from several photovoltaic modules and panel-specific models that preserve module dependent behavior but use fewer training samples. This study compares three modeling scopes for photovoltaic power estimation using 63 observations from three nominally 100 W modules measured over four days. Generalized-Shared, Generalized-Panel-Aware, and Panel-Specific scopes were evaluated using Ridge Regression, K-Nearest Neighbor Regression, Radial Basis Function Network Regression, and Random Forest Regression under nested leave-one-day-out validation. A datasheet physical power estimate was also evaluated as a non-machine-learning baseline. Panel-Specific Ridge achieved the best overall performance with an RMSE of 7.3236 W, MAE of 5.6750 W, and R2 of 0.8788, reducing RMSE by 60.23% relative to the physical baseline (18.4161 W). Generalized-Panel-Aware Random Forest achieved an RMSE of 7.7876 W overall and became the best configuration for P3 (6.3310 W) and for the Tuesday left-out day (4.9565 W). Generalized KNN and panel-aware RBF also benefited from pooled training data, whereas Panel-Specific Ridge remained the strongest regularized linear configuration. Wilcoxon signed-rank tests with Holm correction found no statistically significant scope differences across the 12 panel-day units; the largest scope effect was observed for panel-aware Random Forest (rank-biserial = -0.4872, adjusted p = 0.3027). The results show that the benefit of pooling or specialization is algorithm and operating condition dependent and should be interpreted within the limitations of the small single-site field dataset.

References

[1] L.-G. Hua, Q. A. Memon, M. F. Shaikh, S. A. Shaikh, R. A. Rahimoon, S. H. H. Shah, and A. Qadir, “Comparative analysis of power output, fill factor, and efficiency at fixed and variable tilt angles for polycrystalline and monocrystalline photovoltaic panels – The case of Sukkur IBA University,” Energies, vol. 15, no. 11, Art. no. 3917, 2022, doi: 10.3390/en15113917.

[2] A. Mellit, A. M. Pavan, E. Ogliari, S. Leva, and V. Lughi, “Advanced methods for photovoltaic output power forecasting: A review,” Applied Sciences, vol. 10, no. 2, Art. no. 487, 2020, doi: 10.3390/app10020487.

[3] S. Theocharides, M. Theristis, G. Makrides, M. Kynigos, C. Spanias, and G. E. Georghiou, “Comparative analysis of machine learning models for day-ahead photovoltaic power production forecasting,” Energies, vol. 14, no. 4, Art. no. 1081, 2021, doi: 10.3390/en14041081.

[4] E. Sauter, M. A. Mughal, and Z. Zhang, “Evaluation of machine learning methods on large-scale spatiotemporal data for photovoltaic power prediction,” Energies, vol. 16, no. 13, Art. no. 4908, 2023, doi: 10.3390/en16134908.

[5] M. Alrashidi and S. Rahman, “Short-term photovoltaic power production forecasting based on novel hybrid data-driven models,” Journal of Big Data, vol. 10, Art. no. 26, 2023, doi: 10.1186/s40537-023-00706-7.

[6] S. M. Radhi, S. D. Al-Majidi, M. F. Abbod, and H. S. Al-Raweshidy, “Machine learning approaches for short-term photovoltaic power forecasting,” Energies, vol. 17, no. 17, Art. no. 4301, 2024, doi: 10.3390/en17174301.

[7] Y. Nie, Q. Paletta, A. Scott, L. M. Pomares, G. Arbod, S. Sgouridis, J. Lasenby, and A. Brandt, “Sky image-based solar forecasting using deep learning with heterogeneous multi-location data: Dataset fusion versus transfer learning,” Applied Energy, vol. 369, Art. no. 123467, 2024, doi: 10.1016/j.apenergy.2024.123467.

[8] J. Schreiber and B. Sick, “Model selection, adaptation, and combination for transfer learning in wind and photovoltaic power forecasts,” Energy AI, vol. 14, Art. no. 100249, 2023, doi: 10.1016/j.egyai.2023.100249.

[9] A. Hooshmand and R. Sharma, “Energy predictive models with limited data using transfer learning,” in Proc. 10th ACM Int. Conf. Future Energy Systems, 2019, pp. 12–16, doi: 10.1145/3307772.3328284.

[10] A. E. Hoerl and R. W. Kennard, “Ridge regression: Biased estimation for nonorthogonal problems,” Technometrics, vol. 12, no. 1, pp. 55–67, 1970, doi: 10.1080/00401706.1970.10488634.

[11] T. M. Cover and P. E. Hart, “Nearest neighbor pattern classification,” IEEE Transactions on Information Theory, vol. 13, no. 1, pp. 21–27, 1967, doi: 10.1109/TIT.1967.1053964.

[12] D. S. Broomhead and D. Lowe, “Multivariable functional interpolation and adaptive networks,” Complex Systems, vol. 2, pp. 321–355, 1988.

[13] L. Breiman, “Random forests,” Machine Learning, vol. 45, pp. 5–32, 2001, doi: 10.1023/A:1010933404324.

[14] R. Banik and A. Biswas, “Improving solar PV prediction performance with RF-CatBoost ensemble: A robust and complementary approach,” Renewable Energy Focus, vol. 46, pp. 207–221, 2023, doi: 10.1016/j.ref.2023.06.009.

[15] S. Varma and R. Simon, “Bias in error estimation when using cross-validation for model selection,” BMC Bioinformatics, vol. 7, Art. no. 91, 2006, doi: 10.1186/1471-2105-7-91.

[16] D. R. Roberts, V. Bahn, S. Ciuti, M. S. Boyce, J. Elith, G. Guillera-Arroita, S. Hauenstein, J. J. Lahoz-Monfort, B. Schröder, W. Thuiller, D. I. Warton, B. A. Wintle, F. Hartig, and C. F. Dormann, “Cross-validation strategies for data with temporal, spatial, hierarchical, or phylogenetic structure,” Ecography, vol. 40, no. 8, pp. 913–929, 2017, doi: 10.1111/ecog.02881.

[17] J. Demšar, “Statistical comparisons of classifiers over multiple data sets,” Journal of Machine Learning Research, vol. 7, pp. 1–30, 2006.

[18] D. E. C. Barragán, B. A. A. Acurio, J. López, H. Morais, C. P. Guzman and L. Silva, "Day-Ahead Photovoltaic Power Forecasting with Limited Data," 2024 IEEE URUCON, Montevideo, Uruguay, 2024, pp. 1-5, doi: 10.1109/URUCON63440.2024.10850063.

[19] Y. Wang, W. Fu, J. Wang, Z. Zhen, and F. Wang, “Ultra-short-term distributed PV power forecasting for virtual power plant considering data-scarce scenarios,” Applied Energy, vol. 373, Art. no. 123890, 2024, doi: 10.1016/j.apenergy.2024.123890.

[20] W. Zheng, H. Xiao, and W. Pei, “Distributed-regional photovoltaic power generation prediction with limited data: A robust autoregressive transfer learning method,” Applied Energy, vol. 380, Art. no. 125058, 2025, doi: 10.1016/j.apenergy.2024.125058.

[21] G. D. Prenata and A. Ridhoi, “Analisis perbandingan kinerja KNN Regression dan Support Vector Regression dalam prediksi kehandalan sistem tenaga listrik berdasarkan indeks SAIDI-SAIFI,” Jurnal Teknik Elektro dan Komputasi (ELKOM), vol. 8, no. 1, pp. 139–150, 2026, doi: 10.32528/elkom.v8i1.5213.

[22] G. D. Prenata and A. Ridho'i, “Breakdown voltage prediction of transformer oil under copper/iron contamination using SVR and KNN regression: A baseline study on small experimental data,” El Sains: Jurnal Elektro, vol. 8, no. 1, pp. 81–87, 2026, doi: 10.30996/elsains.v8i1.133785.

[23] G. D. Prenata, “Evaluasi komparatif DFFNN, DFFNN yang dioptimasi GA, DFFNN yang dioptimasi GWO, dan LSTM untuk prediksi konsumsi energi listrik menggunakan sliding window cross-validation,” Techné: Jurnal Ilmiah Elektroteknika, vol. 25, no. 1, pp. 91–104, 2026, doi: 10.31358/techne.v25i1.656.

[24] G. D. Prenata, “Predicting breakdown voltage of transformer oil under copper/iron contamination: A comparative study of gradient vs metaheuristic training,” ELKHA: Jurnal Teknik Elektro, vol. 18, no. 1, pp. 65–73, 2026, doi: 10.26418/elkha.v18i1.101026.