Prediction of B-737-300 Climb Fuel Consumption Using a Multilayer Perceptron

Abstract

In this study, the fuel consumption (pounds) during the climb phase of Boeing B-737-300 aircraft is modeled using a Multilayer Perceptron (MLP) artificial neural network (ANN), with altitude (feet), time (minutes) and distance (nautical miles) as input variables. In addition to the 21-training 3-test data adopted by previous studies, a Leave-One-Out Cross-Validation (LOOCV) protocol is employed to assess the true generalization performance of the model. A 360-run architecture sweep (12 architectures × 3 activations × 10 seeds) over depth, width and activation function identifies an 8×8-neuron MLP with tanh activation as the most suitable configuration. When re-evaluated with 30 random seeds, this architecture yields a test Mean Square Error (MSE) of 134.88 ± 46.12 under the 21/3 split, which is statistically comparable to the previously reported Radial Basis Function Neural Network (RBFNN) (122.30) and Backpropagation Multilayer Neural Network (BP-MLP) (143.35) results on the same dataset. However, the LOOCV mean MSE is 1924.68 ± 182.52, indicating that the three-sample test set underestimates the true generalization error by approximately a factor of fourteen. The findings reveal that for small datasets, the use of LOOCV and multi-seed statistics is a methodological requirement.

Keywords

• Artificial neural networks
• Multilayer perceptron
• Fuel consumption prediction
• LOOCV
• Flight control systems

B-737-300 Tırmanış Yakıt Tüketiminin Çok Katmanlı Algılayıcı ile Tahmini

Abstract

Bu çalışmada, Boeing B-737-300 tipi uçakların tırmanış esnasında tüketmesi gereken yakıt miktarı (libre), uçuş yüksekliği (feet), zaman (dakika) ve mesafe (deniz mili) giriş değişkenleri kullanılarak Çok Katmanlı Algılayıcı (Multilayer Perceptron, MLP) yapay sinir ağı (Artificial Neural Network, YSA) ile modellenmiştir. Önceki çalışmalardaki 21 eğitim ve 3 test verisinin yanı sıra, Tekli Dışlama Çapraz Doğrulama (Leave-One-Out Cross-Validation, LOOCV) protokolü uygulanarak modelin gerçek genelleme performansı değerlendirilmiştir. Saklı katman sayısı, nöron sayısı ve aktivasyon fonksiyonu için yapılan 360 koşumluk mimari taraması (12 mimari × 3 aktivasyon × 10 seed) sonucunda en uygun yapı 8×8 nöronlu, tanh aktivasyonlu MLP olarak belirlenmiştir. Bu yapı 30 farklı random seed ile tekrarlandığında 21/3 protokolünde test Ortalama Karesel Hata (Mean Square Error, MSE) ortalaması 134,88 ± 46,12 olarak elde edilmiştir; bu değer, aynı veri seti üzerinde önceki çalışmalarda rapor edilen Radyal Tabanlı Yapay Sinir Ağı (Radial Basis Function Neural Network, RTYSA) (122,30) ve Çok Katmanlı Yapay Sinir Ağı – Geri Yayılım (Multilayer Artificial Neural Network – Backpropagation, ÇKYSA-BP) (143,35) sonuçlarıyla istatistiksel olarak örtüşmektedir. Ancak LOOCV ile elde edilen ortalama MSE değeri 1924,68 ± 182,52 olarak hesaplanmış; bu bulgu, üç örneklik test setinin gerçek genelleme hatasını yaklaşık on dört kat iyimser bir şekilde tahmin ettiğini ortaya koymaktadır. Bulgular, küçük veri setlerinde LOOCV ve çoklu seed istatistiklerinin kullanılmasının yöntemsel bir gereksinim olduğunu ortaya koymaktadır.

Keywords

• Yapay sinir ağları
• Çok katmanlı algılayıcı
• Yakıt tüketimi tahmini
• LOOCV
• Uçuş kontrol sistemleri

References

1. Jang, J.-S. R. (1993). ANFIS: Adaptive-network-based fuzzy inference system. IEEE Transactions on Systems, Man, and Cybernetics, 23(3), 665–685. https://doi.org/10.1109/21.256541
2. Haykin, S. (2009). Neural networks and learning machines (3rd ed.). Pearson Education. ISBN: 978-0131471399.
3. Konar, M. (2010). Bulanık mantık ve yapay sinir ağları kullanılarak uçuş kontrol sistemlerine ait bazı parametrelerin hesaplanması (Yüksek Lisans Tezi). Erciyes Üniversitesi Fen Bilimleri Enstitüsü, Kayseri.
4. Specht, D. F. (1991). A general regression neural network. IEEE Transactions on Neural Networks, 2(6), 568–576. https://doi.org/10.1109/72.97934
5. Broomhead, D. S., & Lowe, D. (1988). Multivariable functional interpolation and adaptive networks. Complex Systems, 2(3), 321–355.
6. Cawley, G. C., & Talbot, N. L. C. (2010). On over-fitting in model selection and subsequent selection bias in performance evaluation. Journal of Machine Learning Research, 11, 2079–2107. https://jmlr.org/papers/v11/cawley10a.html
7. Liu, D. C., & Nocedal, J. (1989). On the limited memory BFGS method for large scale optimization. Mathematical Programming, 45(1–3), 503–528. https://doi.org/10.1007/BF01589116
8. 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, D., Brucher, M., Perrot, M., & Duchesnay, É. (2011). Scikit-learn: Machine learning in Python. Journal of Machine Learning Research, 12, 2825–2830. https://jmlr.org/papers/v12/pedregosa11a.html

9. Kohavi, R. (1995). A study of cross-validation and bootstrap for accuracy estimation and model selection. Proceedings of the 14th International Joint Conference on Artificial Intelligence (IJCAI), 2, 1137–1143.
10. Boeing Commercial Airplane Company. (2005). 737-300 Operations Manual. Seattle, WA.
11. Macit, R., Ozkan, H., & Baklacioglu, T. (2025). Machine learning-based fuel flow rate prediction for Boeing 737-800 aircraft: A comprehensive approach across climb, cruise and descent flight phases. Energy. https://www.sciencedirect.com/science/article/pii/S0360544225042185
12. Jarry, G., Dalmau, R., Very, P., & Sun, J. (2024). On the generalization properties of deep learning for aircraft fuel flow estimation models. arXiv. https://doi.org/10.48550/arXiv.2410.07717
13. Zhao, Y., Wang, Z., Wang, X., Song, Y., & Han, Y. (2025). Data-driven fuel consumption prediction model for green aviation using radial basis function neural network. Scientific Reports, 15, 26275. https://doi.org/10.1038/s41598-025-11941-8
14. Tang, W., Dai, J., & Li, Y. (2025). Aircraft range fuel prediction study based on WPD with IAPO optimized BiLSTM–KAN model. Scientific Reports, 15, 12231. https://doi.org/10.1038/s41598-025-97264-0
15. Yıldırım Dalkıran, F., & Yıldırım, F. R. (2021). Uçağın trim parametresinin ANFIS kullanılarak tahmin edilmesi. Niğde Ömer Halisdemir Üniversitesi Mühendislik Bilimleri Dergisi, 10(1), 54–62. https://doi.org/10.28948/ngumuh.743036
16. Konar, M., & Bağış, A. (2016). Simultaneous computation of the speed and fuel parameters of flight control system by using ANFIS and artificial neural networks. In 2016 24th Signal Processing and Communication Application Conference (SIU) (pp. 333–336). IEEE. https://doi.org/10.1109/SIU.2016.7496008
17. Bağış, A., & Konar, M. (2018). ABC and DE algorithms based fuzzy modeling of flight data for speed and fuel computation. International Journal of Computational Intelligence Systems, 11(1), 790–802. https://doi.org/10.2991/ijcis.11.1.60
18. Konar, M. (2019). Redesign of morphing UAV's winglet using DS algorithm based ANFIS model. Aircraft Engineering and Aerospace Technology, 91(9), 1214–1222. https://doi.org/10.1108/AEAT-09-2018-0255