ZAMANSAL FÜZYON DÖNÜŞTÜRÜCÜ (TFT) YÖNTEMİYLE YAĞIŞ TAHMİNİ

Abstract

Geçmişten günümüze atmosferde gerçekleşen olaylar olayın gerçekleştiği çevrede tahmin yürütebilmek için araştırma konusu olmaktadır. Bu tahminler, toplanan verilerin sayısal formüllerden geçirilmesiyle oluşan sonuçların değerlendirilmesiyle gerçekleşmektedir. Teknolojinin yazılım ve donanım olarak hızla gelişmesi, literatüre yeni yöntem ve tekniklerin kazandırılmasının temelini oluşturmaktadır. Tahmin aşamasında kullanılan Yapay Sinir Ağları bu yöntemlerden biridir. Bu çalışmada geçmiş toplam küresel güneş radyasyonu, güneşlenme süresi ,rüzgar hızı, bulutluluk miktarı, nem, sıcaklık parametreleri kullanılarak Uzun Kısa Süreli Bellek (LSTM- Long Short-Term Memory) ile Zamansal Füzyon Dönüştürücü (TFT- Temporal Fusion Transformer) yöntemleri karşılaştırılmıştır. Elde edilen sonuçlar, TFT modelinin (RMSE = 0.039, MAE = 0.028, R² = 0.981) performansının, LSTM modeline (RMSE = 0.061, MAE = 0.045, R² = 0.954) göre belirgin biçimde daha yüksek tahmin doğruluğu ve genelleme yeteneği sunduğunu ortaya koymaktadır. Bu üstünlük, TFT’nin çok değişkenli meteorolojik veriler arasındaki zamansal ilişkileri gated residual network yapısı ve çok başlı dikkat (multi-head attention) mekanizması aracılığıyla etkili biçimde yakalayabilmesinden kaynaklanmaktadır; böylece kısa vadeli yağış tahmininde yaklaşık %35 düzeyinde hata azalımı elde edilmiştir.

Keywords

• TFT
• LSTM
• Yapay Sinir Ağları
• Yağış Tahmini

Ethical Statement

Yazarlar, bu çalışmada herhangi bir çıkar çatışması bulunmadığını beyan eder. Araştırma süreci boyunca etik ilkelere uyulmuş ve gerekli onaylar alınmıştır.

Thanks

Bu çalışmada kullanılan meteorolojik verilerin temininde desteklerinden dolayı Meteoroloji Genel Müdürlüğü’ne teşekkür ederiz.

RAINFALL PREDICTION USING THE TEMPORAL FUSION TRANSFORMER (TFT) METHOD

Abstract

Atmospheric events from the past to the present have been the subject of research to enable predictions in the environment in which they occur. These predictions are made by processing collected data through numerical formulas and evaluating the results. The rapid development of technology in software and hardware forms the basis for the introduction of new methods and techniques into the literature. Artificial Neural Networks, used in the prediction phase, are one of these methods. This study compares the Long Short-Term Memory (LSTM) and Temporal Fusion Transformer (TFT) methods using past total global solar radiation, sunshine duration, wind speed, cloud cover, humidity, and temperature parameters. The obtained results demonstrate that the performance of the TFT model (RMSE = 0.039, MAE = 0.028, R² = 0.981) is significantly higher in terms of prediction accuracy and generalization capability compared to the LSTM model (RMSE = 0.061, MAE = 0.045, R² = 0.954). This superiority arises from the ability of the TFT architecture to effectively capture temporal dependencies among multivariate meteorological variables through its gated residual network structure and multi-head attention mechanism, resulting in approximately a 35% reduction in forecasting error for short-term precipitation predictions.

Keywords

• TFT
• LSTM
• Artificial Neural Networks
• Precipitation Forecast

Ethical Statement

The authors declare that there is no conflict of interest in this study. Ethical principles were adhered to throughout the research process, and necessary approvals were obtained.

Thanks

We would like to thank the General Directorate of Meteorology for their support in providing the meteorological data used in this study.

References

1. Akbari, A.A., Yang, T., Hsu, K., Sorooshian, S., Lin, J. and Peng, Q., 2018. Short-term precipitation forecast based on the PERSIANN system and LSTM recurrent neural networks. Journal of Geophysical Research: Atmospheres, 123(22). doi: 10.1029/2018JD028375.
2. Ataseven, B., 2013. Yapay sinir ağları ile öngörü modellemesi. Öneri Dergisi, 10(39), 101–115. doi: 10.14783/od.v10i39.1012000311.
3. Bayram, N., 2015. Araştırmalarda istatistik ve etik. ISGUC The Journal of Industrial Relations and Human Resources, 17(3), 17–29.
4. Biswas, S. K., Sinha, N., Purkayastha, B. and Marbaniang, L., 2014. Weather prediction by recurrent neural network dynamics. International Journal of Intelligent Engineering Informatics, 2(2/3), 166. doi: 10.1504/IJIEI.2014.066208.
5. Burrichter, B., Koltermann da Silva, J., Niemann, A. and Quirmbach, M., 2024. A temporal fusion transformer model to forecast overflow from sewer manholes during pluvial flash flood events. Hydrology, 11(3), 41. doi: 10.3390/hydrology11030041.
6. Ceyhan, V., "Performans ölçümleri ve seçim kriterleri." [Online]. Available: https://avys.omu.edu.tr/storage/app/public/vceyhan
7. Çiğli havalimanı meteoroloji istasyonu. Accessed: Dec. 18, 2024. [Online]. Available: https://www.mgm.gov.tr/kurumsal/istasyonlarimiz.aspx?il=%C4%B0zmir
8. Civitarese, D. S., Szwarcman, D., Zadrozny, B. and Watson, C., 2021. Extreme precipitation seasonal forecast using a transformer neural network.

9. Erken, Ş. and Şenyay, L., 2023. Makine öğrenmesi ile eksik veri tamamlama yöntemlerinin sınıflandırma performansına etkileri. Kayseri Üniversitesi Sosyal Bilimler Dergisi, 5(1), 51–71. doi: 10.51177/kayusosder.1307226.
10. Gökgöz, F., "Çoklu regresyon modeli." Accessed : Dec. 26, 2025. [Online]. Available: https://acikders.ankara.edu.tr/pluginfile.php/117326/mod_resource/content/1/11-Coklu Regresyon 1.pdf
11. Gürbüz, F. B., "Uzun kısa dönemli bellek (Long / Short Term Memory) (LSTM)." Accessed: Dec. 25, 2024. [Online]. Available: https://medium.com/@batincangurbuz/uzun-k%C4%B1sa-d%C3%B6nemli-bellek-long-short-term-memory-lstm-c526980c28b1
12. Gürlek, C. and Bilgili, M., 2023. Aylık toplam güneş ışınımının uzun-kısa süreli bellek (LSTM) yöntemiyle tahmini: Sivas ili örneği. Sivas Cumhuriyet Üniversitesi Bilim ve Teknoloji Dergisi, 2(1), 24–30.
13. Habi, H. V. and Messer, H., 2019. RNN models for rain detection. 2019 IEEE International Workshop on Signal Processing Systems (SiPS), 184–188. doi: 10.1109/SiPS47522.2019.9020603.
14. Han, J. M., Ang, Y. Q., Malkawi, A. and Samuelson, H. W., 2021. Using recurrent neural networks for localized weather prediction with combined use of public airport data and on-site measurements. Build Environ, 192, 107601. doi: https://doi.org/10.1016/j.buildenv.2021.107601.
15. Islam, M., Shuvo, S. S., Ahmed Shohan, J. and Faruque, O., 2023. Forecasting of PV plant output using interpretable temporal fusion transformer model. 2023 North American Power Symposium (NAPS), 1–6. doi: 10.1109/NAPS58826.2023.10318698.
16. Kang, J., Wang, H., Yuan, F., Wang, Z., Huang, J. and Qiu, T., 2020. Prediction of precipitation based on recurrent neural networks in Jingdezhen, Jiangxi Province, China. Atmosphere (Basel), 11(3). doi: 10.3390/atmos11030246.
17. Karabulut, M. A. and Topçu, E., 2022. Derin öğrenme tekniği kullanılarak Kars ilinin hava sıcaklık tahmini. Mühendislik Bilimleri ve Tasarım Dergisi, 10(4), 1174–1181. doi: 10.21923/jesd.1067700.
18. Koya, S.R. and Roy, T., 2024. Temporal fusion transformers for streamflow prediction: Value of combining attention with recurrence. J Hydrol (Amst), 637, 131301. doi: https://doi.org/10.1016/j.jhydrol.2024.131301.
19. Köse, B., 2016. Saatlik ve günlük rüzgâr hızını tahmin için yeni bir analitik yöntem yaklaşımı ve yapay sinir ağları ile karşılaştırması. 10th International Clean Energy Symposium.
20. Kumar, D., Singh, A., Samui, P. and Jha, R. K., 2019. Forecasting monthly precipitation using sequential modelling. Hydrological Sciences Journal, 64(6), 690–700. doi: 10.1080/02626667.2019.1595624.
21. Kumar, D. P. and Picerno, S., 2023. TFT-powered PV production forecasting: A multivariate approach for long, medium, and short terms. 2023 7th International Conference on Computer Applications in Electrical Engineering-Recent Advances (CERA), 1–6. doi: 10.1109/CERA59325.2023.10455743.
22. Latifoğlu, L., 2022. Derin sinir ağları modeli ile standardize yağış indeksi tahmini. Niğde Ömer Halisdemir Üniversitesi Mühendislik Bilimleri Dergisi, 11(4), 1006–1024. doi: 10.28948/ngumuh.1145279.
23. Liao, X., 2024. TFT-transformer: An advanced temporal fusion transformer model for demand forecasting in supply chain management. Proceedings of the X International Scientific and Practical Conference, International Science Group.
24. Lim, B., Arik, S. O., Loeff, N. and Pfister, T., 2019. Temporal fusion transformers for interpretable multi-horizon time series forecasting.
25. López Santos, M., García-Santiago, X., Echevarría Camarero, F., Blázquez Gil, G. and Carrasco Ortega, P., 2022. Application of temporal fusion transformer for day-ahead PV power forecasting. Energies (Basel), 15(14). doi: 10.3390/en15145232.
26. MGM, "Hava Tahmini - Meteoroloji Genel Müdürlüğü." [Online]. Available: https://www.mgm.gov.tr/genel/meteorolojinedir.aspx
27. Mouloud, L.A., Kheldoun, A., Deboucha, A. and Mekhilef, S., 2023. Explainable forecasting of global horizontal irradiance over multiple time steps using temporal fusion transformer. Journal of Renewable and Sustainable Energy, 15(5). doi: 10.1063/5.0159899.
28. Nair, J. P. and Vijaya, M., 2023. TFT architecture and RNN variants for water quality prediction of Bharathapuzha River. International Journal on Recent and Innovation Trends in Computing and Communication, 11, 822.
29. Ouma, Y. O., Cheruyot, R. and Wachera, A. N., 2022. Rainfall and runoff time-series trend analysis using LSTM recurrent neural network and wavelet neural network with satellite-based meteorological data: case study of Nzoia hydrologic basin. Complex & Intelligent Systems, 8(1), 213–236. doi: 10.1007/s40747-021-00365-2.
30. Poornima, S. and Pushpalatha, M., 2019. Prediction of rainfall using intensified LSTM based recurrent neural network with weighted linear units. Atmosphere (Basel), 10(11). doi: 10.3390/atmos10110668.
31. Priatna, M. A. and Djamal, E. C., 2020. Precipitation prediction using recurrent neural networks and long short-term memory. TELKOMNIKA (Telecommunication Computing Electronics and Control), 18(5), 2525. doi: 10.12928/telkomnika.v18i5.14887.
32. Salman, A. G., Kanigoro, B. and Heryadi, Y., 2015. Weather forecasting using deep learning techniques. 2015 International Conference on Advanced Computer Science and Information Systems (ICACSIS), 281–285. doi: 10.1109/ICACSIS.2015.7415154.
33. Seyyarer, E., Ayata, F., Uçkan, T. and Karci, A., 2020. Derin öğrenmede kullanılan optimizasyon algoritmalarının uygulanması ve kıyaslanması. Computer Science, 5(2), 90–98.
34. Sharma, R., Mishra, D. K. and Bhuyan, S., 2024. Prospects of Science, Technology and Applications, 116–135. London: CRC Press. doi: 10.1201/9781003489443.
35. SHI, X., Chen, Z., Wang, H., Yeung, D.-Y., Wong, W. and WOO, W., 2015. Convolutional LSTM network: A machine learning approach for precipitation nowcasting. Advances in Neural Information Processing Systems. Available: https://proceedings.neurips.cc/paper_files/paper/2015/file/07563a3fe3bbe7e3ba84431ad9d055af-Paper.pdf
36. Singh, S., Kaushik, M., Gupta, A. and Malviya, A. K., 2019. Weather forecasting using machine learning techniques. SSRN Electronic Journal. doi: 10.2139/ssrn.3350281.
37. Suanpang, P. and Jamjuntr, P., 2024. Machine learning models for solar power generation forecasting in microgrid application implications for smart cities. Sustainability, 16(14). doi: 10.3390/su16146087.
38. Tang, T., Jiao, D., Chen, T. and Gui, G., 2022. Medium- and long-term precipitation forecasting method based on data augmentation and machine learning algorithms. IEEE J Sel Top Appl Earth Obs Remote Sens, 15, 1000–1011. doi: 10.1109/JSTARS.2022.3140442.
39. Waqas, M., Humphries, U. W., Hlaing, P. T., Wangwongchai, A. and Dechpichai, P., 2024. Advancements in daily precipitation forecasting: A deep dive into daily precipitation forecasting hybrid methods in the tropical climate of Thailand. MethodsX, 12, 102757. doi: https://doi.org/10.1016/j.mex.2024.102757.
40. Wu, B. and Wang, L., 2024. Two-stage decomposition and temporal fusion transformers for interpretable wind speed forecasting. Energy, 288, 129728. doi: https://doi.org/10.1016/j.energy.2023.129728.
41. Wu, B., Wang, L. and Zeng, Y.-R., 2022. Interpretable wind speed prediction with multivariate time series and temporal fusion transformers. Energy, 252, 123990. doi: https://doi.org/10.1016/j.energy.2022.123990.
42. Wu, T., Feng, F., Lin, Q. and Bai, H., 2019. Advanced method to capture the time-lag effects between annual NDVI and precipitation variation using RNN in the arid and semi-arid grasslands. Water (Basel), 11(9), 1789. doi: 10.3390/w11091789.
43. Yıldıran, A. and Kandemir, S. Y., 2018. Yağış miktarının yapay sinir ağları ile tahmini. Bilecik Şeyh Edebali Üniversitesi Fen Bilimleri Dergisi, 5(2), 97–104.
44. Yu, Z., Ming, W. E. I., Nan, L. I. and ILYAS, A. M., 2020. Improved radar heavy precipitation estimation based on RNN. China Sciencepaper, 15(5), 585–592. Available: https://research.ebsco.com/linkprocessor/plink?id=2bf53347-7ba0-35e4-847c-601c445c3da8
45. Zhu, H.-C., Ren, C., Wang, J., Feng, Z., Haghighat, F. and Cao, S.J., 2024. Fast prediction of spatial temperature distributions in urban areas with WRF and temporal fusion transformers. Sustain Cities Soc, 103, 105249. doi: https://doi.org/10.1016/j.scs.2024.105249.