Archive ouverte de l'ENAC
Not a member yet
3458 research outputs found
Sort by
Contributions à la robustesse de l'apprentissage machine avec applications à la prédiction de trafic
No full textThe advent of deep learning brings many opportunities to the civil aviation community. Data-driven methods can automatize tedious tasks (e.g., speech-to-text for incident reporting) or introduce new functions to improve the safety, efficiency, or environmental impact of the air transportation system (e.g., reduction of the complexity of air traffic). However, the unique features of machine learning models (data-driven, stochastic, lack of transparency) prevent their introduction into safety-critical applications. Beyond accuracy, safety-critical applications require the model to be trustworthy by providing guarantees on robustness and explainability. In this work, we explore several aspects of neural networks robustness against adversarial noise using concepts borrowed from information geometry. In the first part, we introduce a regularization term to improve the robustness of neural networks against L2 attacks. This term pushes the model to be locally isometric with respect to the Euclidean metric (in input) and the Fisher information metric (in output). The method is evaluated on MNIST and CIFAR-10 datasets. In the second part, we introduce a method to study the robustness of recurrent neural networks using stochastic differential equations. Then, we conduct several experiments on synthetic data and on MNIST in order to study the interaction between adversarial vulnerability and the foliations induced by the kernel fisher information metric in the input space. In the third part, we apply recurrent neural networks for the prediction of airspace congestion. Finally, we derive a method for uncertainty quantification for Transformer models based on variational inference. This method is evaluated on road traffic data.L'émergence de l'apprentissage profond est porteur de nombreuses opportunités pour l'aviation civile. Les méthodes d'apprentissage peuvent automatiser des tâches pénibles (e.g., transcription de la parole vers le texte pour les rapports d'incidents) ou introduire de nouvelles fonctions visant à améliorer la sécurité, l'efficacité, ou l'impact environnemental du transport aérien (e.g., réduction de la complexité du trafic aérien). Cependant, les caractéristiques propres aux modèles d'apprentissage machine (fondés sur les données, stochastiques, manque de transparence) limitent leur introduction dans les applications où la sécurité est un enjeu central. Au-delà de la performance en termes de précision, ces applications exigent des modèles fiables, capables de fournir des garanties en termes de robustesse et d'explicabilité. Dans cette thèse, nous explorons divers aspects de la robustesse des réseaux de neurones contre les attaques par exemples contradictoires en s'appuyant sur des concepts empruntés à la géométrie de l'information. Dans une première partie, nous introduisons un terme de régularisation visant à améliorer la robustesse des réseaux de neurones contre les attaques L2. Ce terme pousse le modèle à être localement isométrique par rapport à la métrique euclidienne (en entrée) et à la métrique d'information de Fisher (en sortie). Cette méthode est évaluée sur les jeux de données MNIST et CIFAR-10. Dans la seconde partie, on introduit une méthode pour étudier la robustesse des réseaux de neurones récurrents basée sur les équations différentielles stochastiques. Plusieurs expériences sont ensuite menées sur des données synthétiques et sur MNIST afin d'étudier l'interaction entre la vulnérabilité aux exemples contradictoires et les foliations induites par le noyau de la métrique d'information de Fisher dans l'espace d'entrée. Dans la troisième partie, nous appliquons les réseaux de neurones récurrents à la prédiction de la congestion dans l'espace aérien. Finalement, nous développons une méthode fondée sur l'inférence variationnelle pour la quantification de l'incertitude dans les modèles de type Transformer. Cette méthode est évaluée sur des données de trafic routier
On Time-Delay Estimation Accuracy Limit Under Phase Uncertainty
No full textInternational audienceAccurately determining signal time-delay is crucial across various domains, such as localization and communication systems. Understanding the achievable optimal estimation performance of such technologies, especially during design phases, is essential for benchmarking purposes. One common approach is to derive bounds like the Cramér-Rao Bound (CRB), which directly reflects the minimum achievable estimation error for unbiased estimators. Different studies vary in their approach to deal with the degree of misalignment in the global phase originating from both the transmitter and the receiver in a single input, single output (SISO) link during time-delay estimation assessment. While some treat this phase term as unknown, others assume ideal calibration and compensation. As an alternative to these two opposing approaches, this study adopts a more balanced approach by considering that such a phase can be estimated with a defined uncertainty, a measure that could be implemented in many practical applications. The primary contribution provided lies in the derivation of a closed-form CRB expression for this alternative signal model, which, as observed, exhibits an asymptotic behavior transitioning between the results observed in previous studies, influenced by the uncertainty assumed for the mentioned phase term
C/N0 degradation in presence of chirp interference: theoretical model
No full textInternational audienceA growing threat for Global Navigation Satellite System (GNSS) service is Radio Frequency Interference (RFI). An important class of GNSS RFI signatures is time dependent frequency pattern signals, generically termed here as chirp signals radiated by Personal Privacy Devices which are jammers with a continuously growing (and illegal) use. The analysis of the impact of chirp signals on GNSS receivers is of the utmost importance in civil aviation. Civil aviation spectrum regulations characterize the Radio Frequency environment of the safety of life GNSS service at signal processing level by comparing the effective carrier-to-noise power density ratio (C/N0,eff), calculated from a degradation of the nominal C/N0, to a C/N0 threshold. Therefore, in this article the mathematical model of the theoretical C/N0 degradation of the received useful signal in presence of a chirp signal is derived from the traditional calculation of the Spectrum Separation Coefficient (SSC) and the theoretical RFI chirp signal power spectrum density also developed in this work. Moreover, the impact of the chirp signal characteristics on the SSC and C/N0 degradation are commented. Finally, the applicability of the proposed model based on the SSC is also analyzed from the GNSS signal, receiver local replica and RFI chirp signal characteristics
Free-space optics emulator for coherent detection of satellite-ground link
No full textInternational audienceThis paper presents a test bench for the emulation of a satellite-ground laser link, from the atmospheric propagation to its coherent detection. The scheme involves two projects. The EPLO project (Free-space Optical Propagation Emulator) aims at modeling and mitigating atmospheric impairments. The wave front deformation is emulated by time series with a DMD (Digital Micro-mirror Device), using Lee holographic methods and the Rytov approximation in the Kolmogorov theory. The signal then undergoes a correction in the TILBA-ATMO device, based on Cailabs’ Multi-Plane Light Conversion (MPLC). Alongside this approach, the CALICO project (Compensation of Atmospheric losses on a laser LInk by COherent detection) proposes a reception chain built on an Optical Phase-Locked Loop (OPLL) for homodyne demodulation and tracking of high Doppler rates due to low-Earth orbits. The end-to-end communication chain is here completed by implementing the OPLL-based receiver after the TILBA.Previous studies have explored the constraints of speed and stability requirements of an OPLL for the desired application, with theoretical and simulation results. Combining EPLO and CALICO will however make it possible to physically assess on a single test bench the robustness of phase-shift keying modulation schemes, both addressing atmospheric disturbances and data recovery.To achieve this coupling, both EPLO and CALICO parts need to be digitally simulated. The TILBA component is to be modeled and added to both simulations. The first results assessing the performance of the simulated OPLL are also expected, with a view to an FPGA implementation. This article presents the status of the modeling.<br /
Towards energy self-sufficiency for a regional airport thanks to local production of new energy carriers
No full textInternational audienceThe transition from fossil fuels to sustainable energy sources is an important challenge for current aviation industries. One of the critical questions in this transition is how much electricity is needed to accommodate future air transport demand and how to supply it. The answer depends on the type of energy carriers, their production process, and the source of the electricity. This paper provides a methodology to estimate the annual required electricity to produce energy carriers (particularly electrofuel and liquid hydrogen) with a given energy production process for an airport. The methodology also computes the surface of solar panels and/or the number of wind turbines needed to produce the annual electricity, which shows the possibility of energy self-sufficiency within the region around an airport. We apply the methodology to the Carcassonne Airport located in the Aude department, France, where sustainable energy sources are easily accessible. The result shows that 112 GWh and 138 GWh of electricity will be required when all the airplanes departing and landing at the airport in 2022 are replaced with electrofuel and liquid hydrogen-powered airplanes, respectively. 49 ha of solar panels or 21 onshore wind turbines are necessary to supply the required electricity in the case of electrofuel. 61 ha of solar panels or 25 onshore wind turbines are necessary to supply the required electricity in the case of liquid hydrogen. This corresponds to 1% to 3% of the electricity production of the Aude department in 2030, which implies that energy self-sufficiency of the airport is feasible. The methodology is also applied to all French airports in 2016. The total electricity required at the national level and per airport is assessed. This methodology can contribute to the elaboration of national and local energy strategies for aviation
Improving Flight Trajectory Predictions with Bayesian-Optimized ConvLSTM Models
No full textInternational audienceFlight trajectory prediction is a critical task in aviation, enabling efficient air traffic management and ensuring safe and seamless flight operations. Conventional methods for trajectory prediction struggle to capture complex spatiotemporal dependencies and uncertainties inherent in the aviation domain. Within this study, we introduce an innovative method to predict flight trajectories, named BayesCoordLSTM. It is a hybrid model that transforms coordinate and applies Bayesian theorem into the ConvLSTM models. The proposed model leverages the spatial features gleaned by the Convolutional Neural Network (CNN) architecture and the temporal dependencies captured by Long-Short Term Memory (LSTM) to enhance the accuracy of trajectory predictions. By incorporating Bayesian theorem, our model provides probabilistic trajectory forecasts and associated confidence levels while coordinate transformation enhances spatial awareness and predictive capabilities. The paper presents experimental results demonstrating the effectiveness of the proposed BayesCoordLSTM-based approach in improving flight trajectory prediction accuracy, with a focus on Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) values. The integration of the Bayesian theorem and Coordinate transformation into ConvLSTM models represents a substantial advancement in the field of flight trajectory prediction
Who is My Main Competitor? Measuring Airline Pairwise Competitive Intensity Based on Route Supply Strategies
No full textClassical approaches to the study of airline competition involve either assessing market concentration or studying rivalry dynamics, through adaptive responses to competition. However, these methods often lack precision at the individual level or reach the limits of computational complexity. To leverage the advantages of both methods, we discuss various measures of competitive intensity between two airlines using bivariate statistical analysis, based on their flight supply strategies across European routes. First, four pairwise symmetric indices are presented and compared. The weighted Jaccard index is identified as the best measure and used to examine the European domestic market in the first and third quarters of 2023. The results show the consistency of the index with real competitive situations. In addition, an asymmetric version of the weighted Jaccard index is introduced for further research and discussion.</div
Functional Ecological Inference
No full textIn this paper we consider the problem of ecological inference when one observes the conditional distributions of Y |W and Z|W from aggregate data and wants to infer the conditional distribution of Y |Z without observing Y and Z in the same sample. First, we show that this problem can be transformed into a linear equation involving operators for which, under suitable regularity assumptions, least squares solutions are available. Then we propose to use the least squares solution with the minimum Hilbert-Schmidt norm, which in our context can be structurally interpreted as the solution with minimum dependence between Y and Z. Interestingly, in the case where the conditioning variable W is discrete and belongs to a finite set, such as the labels of units/groups/cities, the solution of this minimal dependence has a closed form. In the more general case, we use a regularization scheme and show the convergence of our proposed estimator. A numerical evaluation of our procedure is proposed
Synchronisation des plannings train-avion : estimation de la demande passager, modèles d'optimisation mathématique, et application au cas de l'Europe de l'Ouest
No full textAir transportation network has been developed and optimised for years in order to provide passengers with a high level of service. However, the growing increase in environmental awareness and airport congestion make trains a relevant alternative to replace short-haul flights. If trains have to complement flights, collaboration between both operators will be required to maintain attractiveness and limit passengers' discomfort during transfers between modes. In this thesis, we propose synchronising air and rail timetables to enhance the quality of air-rail transfers for passengers at hub airports. As the transfer demand between rail and air is not publicly available, we first introduce a methodology to generate realistic air-rail transferring passenger demand using open-source data, using constraint programming. Then, we propose two mixed integer linear programmes to address this air-rail synchronisation issue, that modify flight and train initial schedules in order to offer passengers seamless connections between the two modes. The first model, well-suited for long-term planning, allows transportation schedulers to include seamless connections with other modes in their operations. The second formulation aims at rescheduling trains and flights at a tactical level to minimise the overall passenger delay across a multimodal transportation network. Both models take into account operational constraints, and are tested on the Western Europe transportation network. In a final step, we propose to re-think the scheduling process from scratch, and jointly develop an air-rail transportation network: an integrated service network design process between rail and air is proposed, considering passengers' travel option preferences and dollar co dollar emissions. A mixed-integer linear programming formulation of the problem is proposed. The methodology is tested on the Spanish transportation case study, using mobile phone network data as the passenger demand input.Le transport aérien a été développé et optimisé depuis de nombreuses années afin de proposer aux passagers un service de qualité. Cependant, la prise de conscience écologique et la congestion croissante des aéroports amènent à repenser le transport aérien. Aujourd'hui, les trains apparaissent comme une alternative pertinente aux vols de courte distance. Si les trains doivent remplacer les vols, une collaboration entre les acteurs du transport aérien et ferroviaire sera nécessaire afin de maintenir un niveau de service élevé pour les passagers. À travers cette thèse, des mécanismes de coordination entre le transport aérien et ferroviaire sont proposés afin d'améliorer la qualité des transferts entre les deux modes. La demande passager entre les avions et les trains étant aujourd'hui peu connue, et non accessible de manière publique, une première étape de simulation, basée sur de la programmation par contraintes, est tout d'abord proposée. Ensuite, deux modèles de synchronisation des horaires, utilisant les plannings existants, sont développés afin de proposer aux passagers des connexions train-vol plus fluides. Le premier, adapté pour une utilisation à l'échelle stratégique (plusieurs semaines avant les opérations), permet aux compagnies aériennes et ferroviaires de proposer aux passagers des connexions plus confortables, en termes de temps de transfert, entre les deux modes. Le second modèle a pour objectif de re-planifier les vols et les trains en temps réel, afin d'attendre les passagers impactés par un retard, minimisant ainsi le risque de connexion manquées. Les deux problèmes sont modélisés sous forme de problèmes linéaires mixtes en nombres entiers, et testés à l'échelle d'un réseau de transport européen. Enfin, nous proposons dans une dernière partie, un modèle de synchronisation plus en amont des deux précédentes méthodes: une estimation jointe des fréquences journalières des vols et des trains. Le point de vue passager est adopté, avec pour objectif principal de minimiser le temps de trajet porte-à-porte, incluant des trajets multimodaux. Le coût dollar co dollar du réseau de transport est également pris en compte. La méthodologie est testée sur un cas d'étude réel : le réseau de transport espagnol. Une demande réaliste est obtenue à partir de l'exploitation de données des opératures de téléphonie mobile
Balancing fuel efficiency and environmental impact: 4D trajectory optimization through Fast Marching Tree for transatlantic flights considering contrails
No full textThis paper proposes a method for computing an airliner trajectory in its cruise phase, minimizing environmental impact. In particular, the problem of contrails is addressed. The proposed method is based on an algorithm derived from robotics, the Fast Marching Tree algorithm. After having been tested on Unmanned Aerial Vehicles (UAVs) and on computing trajectories in a wind field for commercial aircraft, it is now adapted to the case of contrails. Several modifications and additions are made to enable it to deal with soft obstacles that evolve over time, as well as the operational constraints associated with the cruise phase. The method is thus able to propose flight level changes in line with operational expectations, adaptable to the strategy of the user (airline for example). Two experiments are proposed on the Paris-Miami and Paris-Denver flights, showing low computation times and satisfactory results