Simple Models or Deep Learning Architectures? A Study on Intergranular Temperature Prediction in Vertical Grain Silos

• Gustavo Queiroz Silveira UTFPR
• Danilo Sipoli Sanches UTFPR
• Rafael Staiger Bressan UFPR
• Robero Ortoncelli UTFPR

DOI: https://doi.org/10.5753/eniac.2026.30822

Resumo

This study presents a comparative analysis of accurate intergranular temperature prediction using real data from three Brazilian vertical silos, comparing statistical methods, tabular machine learning, temporal deep learning, and spatiotemporal deep learning models under a unified evaluation protocol. To avoid overestimating performance, results are analyzed by aeration condition and compared with a persistence baseline. The results show that predictive performance saturates at the tabular level: Random Forest and XGBoost with aeration-related features consistently outperform persistence during aeration. At the same time, temporal and spatiotemporal deep learning models incur higher computational costs without improving accuracy.

Referências

Box, G. E. P. et al. (2015). Time Series Analysis: Forecasting and Control. John Wiley & Sons, 5th edition.

Demšar, J. (2006). Statistical comparisons of classifiers over multiple data sets. Journal of Machine Learning Research, 7:1–30.

Diebold, F. X. and Mariano, R. S. (1995). Comparing predictive accuracy. Journal of Business & Economic Statistics, 13(3):253–263.

Elsayed, S. et al. (2021). Do we really need deep learning models for time series forecasting? arXiv preprint arXiv:2101.02118.

Géron, A. (2022). Hands-on machine learning with Scikit-Learn, Keras, and TensorFlow. ” O’Reilly Media, Inc.”.

Grinsztajn, L., Oyallon, E., and Varoquaux, G. (2022). Why do tree-based models still outperform deep learning on typical tabular data? In Advances in Neural Information Processing Systems (NeurIPS), volume 35, pages 507–520.

Hyndman, R. J. and Athanasopoulos, G. (2018). Forecasting: Principles and Practice. OTexts, Melbourne, Australia, 2nd edition.

Ishida, K. et al. (2024). Use of one-dimensional cnn for input data size reduction in lstm for improved computational efficiency and accuracy in hourly rainfall-runoff modeling. Journal of Environmental Management, 359:120931.

Junior, F. et al. (2024). Estimating energy efficiency of the aeration process of stored grains through machine learning. Revista Brasileira de Engenharia Agrícola e Ambiental, 28(11):e281001.

Li, X. et al. (2025). Smart grain storage solution: Integrated deep learning framework for grain storage monitoring and risk alert. Foods, 14(6):1024.

Maier, D. E. and Plumier, B. M. (2022). Grain operations: Temperature cable placement. World Grain. Acesso em: abr. 2026.

Makridakis, S., Spiliotis, E., and Assimakopoulos, V. (2018). Statistical and machine learning forecasting methods: Concerns and ways forward. PLoS ONE, 13(3):e0194889.

Montross, M. D., Maier, D. E., and Haghighi, K. (2002). Validation of a finite-element stored grain ecosystem model. Transactions of the ASAE, 45(5):1465–1474.

Navarro, S. and Noyes, R. T., editors (2002). The Mechanics and Physics of Modern Grain Aeration Management. CRC Press, Boca Raton, FL.

Paraginski, R. T. et al. (2015). Qualidade de grãos de milho armazenados em diferentes temperaturas. Revista Brasileira de Engenharia Agrícola e Ambiental, 19(4):358–363.

Plumier, B. M. et al. (2021). Effect of temperature sensor numbers and placement on aeration cooling of a stored grain mass using a 3D finite element model. Agriculture, 11(3):231.

Reed, C. and Harner, J. (1998). Cooling of stored wheat in multiple or single cycles using automatic aeration controllers. Applied Engineering in Agriculture, 14(5):495–500.

Reginato, M. P. et al. (2014). Boas práticas de armazenagem de grãos. Anais do ENIC, (6).

Shi, X. et al. (2015). Convolutional LSTM network: A machine learning approach for precipitation nowcasting. In Advances in Neural Information Processing Systems (NeurIPS), volume 28, pages 802–810.