A reliable predict-then-optimize approach for minimizing aircraft fuel consumption

Abstract

Achieving sustainability in aviation necessitates optimizing loaded fuel to reduce both financial costs and environmental impact, as loaded fuel directly affects aircraft weight, which in turn influences fuel consumption throughout the flight. This study develops a reliable predict-then-optimize approach for minimizing aircraft fuel consumption. First, artificial intelligence-based models are developed to predict fuel consumption rates using Quick Access Recorder data. Then, based on accurate fuel consumption predictions, a data-driven optimization model is further established to determine the minimum loaded fuel, assisting dispatchers in airlines with flight planning. We rigorously prove that under mild assumptions, the approach can return the minimum loaded fuel with given reliability within polynomial times. Experiments were conducted using the four most widely used aircraft models, i.e., A320, A321, B737, and B738. The results show that optimized loaded fuel can achieve an average fuel consumption reduction of 3.67% compared to actual consumption.