Prediction of Aircraft Trip Fuel Deviations for Fuel Loading Decisions with a Deep Time Series Approach

Prediction of Aircraft Trip Fuel Deviations for Fuel Loading Decisions with a Deep Time Series Approach

Lampe, Reinoud (TU Delft Aerospace Engineering)

Li, L. (mentor)
Santos, Bruno F. (mentor)

Delft University of Technology

Aerospace Engineering | Air Transport and Operations

2022-12-19

Reducing fuel consumption is an increasingly important topic within aviation. One approach to accomplish this goal is reducing excess fuel weight being loaded on aircraft. Flight dispatchers and pilots load extra fuel to account for uncertainties present in trip fuel consumption, which is currently computed by the flight planning system (FPS). In this paper, a time-series-based model is proposed that predicts deviations in trip fuel consumption of commercial flights. The aim is to assist dispatchers with fuel loading decisions, using the proposed model. A 2-layered time series is proposed, able to capture temporal patterns present in deviations of trip fuel consumption. The first layer is a fixed time interval time series, grouping flights per time interval, to estimate average trip fuel deviation for the coming time intervals. The output of this layer is used for the second layer, which is a sequential time series model, modelling each flight individually, able to capture patterns present in the individual flight information. To estimate the trip fuel deviations for coming flights, input features from the operational flight flan, weather descriptive features from terminal area forecasts and historical flight descriptive input features from the flight data recorder are used. The new prediction model is able to reduce the root mean squared error (RMSE) of trip fuel predictions of the FPS by 26% and reduces the RMSE compared to the baseline gradient boosting model by 5.4%. Using a fixed-buffer loading strategy, 0.12 - 0.39% of fuel consumption could be reduced, depending on the desired safety key performance indicators, which leads to yearly savings of up to $1.5 million and 4,792 tonnes of CO₂.

Fuel
prediction
airline
time series
deep learning

http://resolver.tudelft.nl/uuid:d272e763-9026-464e-b12d-9902b5d7eaf8

Student theses

master thesis

© 2022 Reinoud Lampe