Data-driven models for full-flight fuel flow rate prediction in civil aviation: A case study on A320 and B787 aircraft

To achieve sustainable development in the aviation industry, assessing the environmental benefits of aviation emission-reduction strategies, such as route optimization, requires accurate fuel consumption prediction models. This study proposes a machine learning-based model for predicting aircraft fuel flow rate (FFR) using quick access recorder (QAR) data. We evaluate several machine learning architectures, e.g., the multilayer perceptron (MLP), convolutional neural network (CNN), and gradient-boosted decision tree (GBDT), on two representative aircraft types: the narrow-body Airbus A320 and the wide-body Boeing 787, respectively. Results demonstrate that the GBDT model achieves the highest predictive accuracy, with a coefficient of determination (R ²) exceeding 0.97 for the A320, followed by the CNN (R ²=0.96) and MLP (R ²=0.94). Similar performance trends are observed for the B787. Moreover, the GBDT also demonstrates superior predictive accuracy compared to two open-source models, OpenAP and Acropole. The trained GBDT model is further combined with interpretable analytical techniques to identify the key input features influencing fuel consumption during the cruise phase. As expected, altitude, gross weight, airspeed are the three most significant parameters affecting aircraft FFR. Finally, we present a proof-of-concept demonstration that integrates meteorological data from external sources to predict FFR, laying the groundwork for real-time trajectory optimization in future works.