Efficient data-enhanced surrogate model for aircraft fuel consumption to support decision-making and policy analysis

Abstract

The rapid growth in commercial air transportation and the volatility of fuel price push for fuel reduction policy to be implemented. Some changes in technology (e.g., improved aircraft design or engine), operations (e.g., improved flight routes), or both have showed promising results on fuel reduction in air transportation. While there are some existing fuel burn evaluation models, some of them are computationally expensive or built based on data that might be outdated; and some others suffer from the lack of accuracy due to simplification assumptions and computations. This might impose limitations in the aforementioned policy analysis, in particular when we need to predict future projections for different scenarios. As such, I develop a fast, efficient, and yet accurate fuel burn evaluation model by combining low-fidelity physics-based model with BADA trajectory simulation results. In particular, I derive correction factors based on the simulation data and incorporate them in the modeling, to achieve a higher accuracy that would not have been possible with the low-fidelity physics-based models alone. In this thesis, a fuel burn database corresponding to 40 aircraft types is generated based on the Bureau of Transportation Statistic (BTS) flight missions database from the year of 2015. A sample-based surrogate model is then derived for each aircraft type. The verification and validation results show that the model can estimate the total aggregate fuel burn for each aircraft type with less than 1% prediction errors using flight missions data from 2016, and less than 6% prediction errors when compared with the actual fuel burn data corresponding to three commercial airliners in 2015 and 2016. The developed models are then used to investigate the two common simplifying assumptions in fuel burn evaluation, namely the cruise-only approximation and the similar aircraft type mapping. The results provide insight into the inaccuracies caused by these simplifications in fuel burn computation. The developed models would open doors to performing more computationally intensive analyses, such as sensitivity analyses, uncertainty analyses, and optimization.

Date of Award	2018
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology