Développement de modèles dérivés de la théorie des graphes pour l’optimisation des trajectoires au sol des avions

Abstract

Ground trajectory optimization represents a promising solution for minimizing aircraft fuel consumption and reducing associated pollutant emissions. The primary objective of this study is to evaluate the potential reduction achievable through this strategy, which is still relatively unexplored, to meet the environmental goals mentioned by the Intergovernmental Panel on Climate Change (IPCC). The first part of the thesis focuses on developing a performance model to estimate aircraft fuel consumption during the taxiing phase. This model, inspired by in-flight performance models, was then adjusted for the Cessna Citation X business jet using a level D professional flight simulator, designed and manufactured by CAE. The model is subsequently adapted to conform to the formalism of the cost functions to be optimized. The first optimization method, Bidir-nacfA*, inspired by Dijkstra’s algorithm, adapts to the constraints imposed by the cost function and airport traffic restrictions by quickly exploring the most promising solutions first. This heuristic method is presented as an efficient solution for generating optimized trajectories in terms of computation time and costs. Its flexibility and ability to adapt to various cost functions are highlighted. Moreover, applying this method to practical cases allows for an estimated fuel saving ranging from 3.5. The conditional weighting shortest path (CWSP) method, on the other hand, modifies the problem’s structure using iterations of the problem’s line graph and relies on conditional weighting using local variations of the cost function to solve problems accurately. This method proves to be particularly robust and effective for local optimization variables, though it requires simplification of cost functions when they exhibit global variations. This method, tested on a constrained optimization problem in an airport, allowed us to solve 100% of the problems compared to 28% for Dijkstra’s algorithm. The cost functions developed in these methods are adapted using a deep learning model to classify the trajectory segments in the airport graph. A comparison of the two methods showed that the Bidir-nacfA* method offers results close to those obtained with CWSP, with an error ranging from 0.12% to 1.34%.

Date	11 Sept 2024
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Georges Ghazi (Supervisor) & Ruxandra Botez (Co-supervisor)