Improving airline fuel efficiency via fuel burn prediction and uncertainty estimation

Abstract

Reducing fuel consumption is a unifying goal across the aviation industry. One fuel-saving opportunity for airlines is the possibility of reducing discretionary fuel loading by dispatchers. In this study, we propose a novel discretionary fuel estimation approach that can assist dispatchers with better discretionary fuel loading decisions. Based on the analysis on our study airline, our approach is found to substantially reduce unnecessary discretionary fuel loading while maintaining the same safety level compared to the current fuel loading practice. The idea is that by providing dispatchers with more accurate information and better recommendations derived from flight records, unnecessary fuel loading and corresponding cost-to-carry could both be reduced. We apply ensemble learning techniques to improve fuel burn prediction and construct prediction intervals (PIs) to capture the uncertainty of model predictions. The upper bound of a PI can then be used for discretionary fuel loading. The potential benefit of this approach is estimated to be $61.5 million in fuel savings and 428 million kg of CO2 reduction per year for our study airline. This study also builds a link between discretionary fuel estimation and aviation system predictability in which the proposed models can also be used to predict benefits from reduced fuel loading enabled by improved Air Traffic Management (ATM) targeting on improved system predictability.