Rethinking Solar Forecasting: Why Univariate Transformers Challenge Classical Statistical Models in Multi-Series Data

• Beatriz Ribeiro UNIFESP
• Didier Vega-Oliveros UNIFESP

DOI: https://doi.org/10.5753/eniac.2026.31491

Resumo

Electricity production must keep up with the growing demand, particularly from renewable sources. However, anticipating generation from renewable sources can be challenging, as they are highly variable and consist of multiple series, one for each plant. To prevent power shortages, prediction techniques can be used. This study compared the performance of statistical techniques that require a model for each series with neural network models that can generalize with a single model. The techniques tested were Seasonal Autoregressive Moving Average (SARIMA), Prophet, Long-Short Term Memory (LSTM), LSTM with Convolutional Neural Network (CNN), and Patch Time Series Transformer (PatchTST). The neural network models included only power generation models, as well as models using solar radiation and temperature. The best models were SARIMA and PatchTST, showing that even a single model can perform as well as neural networks.

Referências

Alameer, Z., Ye, H., Jianhua, Z., Elaziz, M. A., and Ewees, A. A. (2019). Forecasting gold price fluctuations using improved multilayer perceptron neural network and whale optimization algorithm. Resources Policy, 61:250–260.

Chen, X., Li, Y., Ye, X., Xiong, X., Zhang, Y., and Zhang, F. (2021). A novel hybrid model based on an improved seagull optimization algorithm for short-term wind speed forecasting. Processes, 9(2):121.

Gao, Z., Zhang, X., Xing, J., Deng, Z., Chen, T., and Wang, C. (2025). Research on building energy consumption prediction based on hybrid gru neural network. Arabian Journal for Science and Engineering.

Hewamalage, H., Bergmeir, C., and Bandara, K. (2021). Recurrent neural networks for time series forecasting: Current status and future directions. International Journal of Forecasting, 37(1):388–427.

Instituto Nacional de Meteorologia (2025). Dados históricos anuais das estações meteorológicas automáticas. [link]. Conjunto de dados.

Liu, M., He, J., Huang, Y., Tang, T., Hu, J., and Xiao, X. (2022). Algal bloom forecasting with time-frequency analysis: A hybrid deep learning approach. Water Research, 219.

Liu, Z., Zhu, Z., Gao, J., and Xu, C. (2021). Forecast methods for time series data: A survey. IEEE Access, 9:91896–91912.

Mirboluki, A., Mehraein, M., and Kisi, O. (2022). Improving accuracy of neuro fuzzy and support vector regression for drought modelling using grey wolf optimization. Hydrological Sciences Journal, 67(10):1582–1597.

Monjoly, S., André, M., Calif, R., and Soubdhan, T. (2017). Hourly forecasting of global solar radiation based on multiscale decomposition methods: A hybrid approach. Energy, 119:288–298.

Montieri, A., Bovenzi, G., Aceto, G., Ciuonzo, D., Persico, V., and Pescapé, A. (2021). Packet-level prediction of mobile-app traffic using multitask deep learning. Computer Networks, 200.

Nie, Y., Nguyen, N. H., Sinthong, P., and Kalagnanam, J. (2022). A time series is worth 64 words: Long-term forecasting with transformers. arXiv preprint.

O’Malley, T., Bursztein, E., Long, J., Chollet, F., Jin, H., Invernizzi, L., et al. (2019). Kerastuner. [link].

ONS (2025a). Fator de capacidade de geração eólica e solar. [link]. Conjunto de dados do Operador Nacional do Sistema Elétrico, BR.

ONS (2025b). Geração por usina em base horária. [link]. Conjunto de dados do Operador Nacional do Sistema Elétrico, BR.

Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., Vanderplas, J., Passos, A., Cournapeau, M., Brucher, M., Perrot, M., and Duchesnay, E. (2011). Scikit-learn: Machine learning in python. Journal of Machine Learning Research, 12:2825–2830.

Rasul, K. and et al. (2024). Patch time series transformer in hugging face. GitHub.

Rick, R. and Berton, L. (2022). Energy forecasting model based on cnn-lstm-ae for many time series with unequal lengths. Engineering Applications of Artificial Intelligence, 113:104998.

Saeed, N., Nguyen, S., Cullinane, K., Gekara, V., and Chhetri, P. (2023). Forecasting container freight rates using the prophet forecasting method. Transport Policy, 133:86–107.

Succetti, F., Rosato, A., Araneo, R., and Panella, M. (2020). Deep neural networks for multivariate prediction of photovoltaic power time series. IEEE Access, 8:211490–211505.

Walther, J. and Weigold, M. (2021). A systematic review on predicting and forecasting the electrical energy consumption in the manufacturing industry. Energies, 14(4):968.

Wu, N., Green, B., Ben, X., and O’Banion, S. (2020). Deep transformer models for time series forecasting: The influenza prevalence case. arXiv preprint. arXiv:2001.08317.

Zhang, Y.-L., Bai, Y.-T., Jin, X.-B., Su, T.-L., Kong, J.-L., and Zheng, W.-Z. (2023). Deep fusion prediction method for nonstationary time series based on feature augmentation and extraction. Applied Sciences, 13(8):5088.