Archive ouverte de l'ENAC
Not a member yet
3458 research outputs found
Sort by
Applying Bayesian inference in a hybrid CNN-LSTM model for time series prediction
No full textInternational audienceConvolutional neural networks (CNN) and Long short-term memory (LSTM) provide stateof-the-art performance in various tasks. However, these models are faced with overfitting on small data and cannot measure uncertainty, which have a negative effect on their generalization abilities. In addition, the prediction task can face many challenges because of the complex long-term fluctuations, especially in time series datasets. Recently, applying Bayesian inference in deep learning to estimate the uncertainty in the model prediction was introduced. This approach can be highly robust to overfitting and allows to estimate uncertainty. In this paper, we propose a novel approach using Bayesian inference in a hybrid CNN-LSTM model called CNN-Bayes LSTM for time series prediction. The experiments have been conducted on two real time series datasets, namely sunspot and weather datasets. The experimental results show that the proposed CNN-Bayes LSTM model is more effective than other forecasting models in terms of Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) as well as for uncertainty quantification
Exact Computation of Maximal Invariant Sets for Safe Markov Chains—Lattice Theoretic Approach
No full textInternational audienceSet theoretic analysis plays a crucial part in understanding safety requirements for control systems. We apply the lattice theoretic insights of fixed point theorems to construct the maximal invariant set and verify safety constraint satisfaction. We present a Kleene iteration-based algorithm with a novel initialization—based on omega-limit sets—for exactly computing the maximal invariant set under a general safety constraint, and we analyze this algorithm for the special case of Markov chains under polyhedral safety constraints. We also establish sufficient conditions for finite termination of the for a general discrete-time system, which is applicable to Markov chains. In addition to proving that the new initialization of the Kleene iteration algorithm requires no more iterations than the previously considered initialization, we present an example where our method computes the exact maximal invariant set in finitely many iterations and the previous method does not
Modelling interactive computing systems: Do we have a good theory of what computers are?
No full textInternational audienceComputers are increasingly interactive. They are no more transformational systems producing a final output after a finite execution. Instead, they continuously react in time to external events that modify the course of computing execution. While philosophers have been interested in conceptualizing computers for a long time, they seem to have paid little attention to the specificities of interactive computing. We propose to tackle this issue by surveying the literature in theoretical computer science, where one can find explicit proposals for a model of interactive computing. In that field, the formal modelling of interactive computing systems has been brought down to whether the new interaction models are reducible to Turing Machines. There are three areas where interaction models are framed. The comparison between TMs and interactive system models is at stake in all of them. These areas are namely some works on concurrency by Milner, on Reactive Turing Machines, and on interaction as a new computing paradigm. For each of the three identified models, we present its motivation, sum up its account for interaction and its legacy, and point out issues regarding the understanding of computers. The survey shows difficulties for epistemologists. The reason is that these analyses focus on the formal equivalence between interactive models of computation and classic ones. Such a project is different from addressing how a computing machine can be interactive: in other words, which mechanisms allow it
Managing Future Challenges for Safety
No full textInternational audienceThis open access book addresses the future of work and industry by 2040—a core interest for many disciplines inspiring a strong momentum for employment and training within the industrial world. The future of industrial safety in terms of technological risk-management, although of obvious concern to international actors in various industries, has been quite sparsely addressed. This brief reflects the viewpoints of experts who come from different academic disciplines and various sectors such as oil and gas, energy, transportation, and the digital and even the military worlds, as expressed in debates and discussions during a two-day international seminar. The contributors address such questions as: What influence will ageing and lack of digital skills in the workforce of the occidental world have on safety culture? What are the likely impacts of big data, artificial intelligence and autonomous technologies on decision-making, and on the roles and responsibilities of individual actors and whole organizations? What role have human beings in a world of accelerating changes? What effects will societal concerns and the entrance of new players have on technological risk management and governance?Managing Future Challenges for Safety will interest and influence researchers considering the future effects of a number of currently developing technologies and their practitioner counterparts working in industry and regulation
Safety Management Systems and their Origins: Insights from the Aviation Industry
No full textInternational audienceSafety Management Systems and their Origins: Insights from the Aviation Industry presents different perspectives on SMS to better decode what it means as a safety approach and what it implicitly conveys beyond safety. The book uses the aviation industry as a basis for analyzing where the SMS stands in terms of safety enhancement. Through a socio-historical analysis of how SMSs emerged and spread across high-risk industries and countries, the book also explains the other stakes underpinning this new approach to safety management
Distribution Prediction of Strategic Flight Delays via Machine Learning Methods
No full textInternational audiencePredicting flight delays has been a major research topic in the past few decades. Various machine learning algorithms have been used to predict flight delays in short-range horizons (e.g., a few hours or days prior to operation). Airlines have to develop flight schedules several months in advance; thus, predicting flight delays at the strategic stage is critical for airport slot allocation and airlines' operation. However, less work has been dedicated to predicting flight delays at the strategic phase. This paper proposes machine learning methods to predict the distributions of delays. Three metrics are developed to evaluate the performance of the algorithms. Empirical data from Guangzhou Baiyun International Airport are used to validate the methods. Computational results show that the prediction accuracy of departure delay at the 0.65 confidence level and the arrival delay at the 0.50 confidence level can reach 0.80 without the input of ATFM delay. Our work provides an alternative tool for airports and airlines managers for estimating flight delays at the strategic phase
AEON: Toward a concept of operation and tools for supporting engine-off navigation for ground operations
No full textInternational audienceThe SESAR Advanced Engine-off Navigation project (AEON) aims at exploring the reduction of ground operations environmental impact based on the use of three classes of greener taxiing solutions: singleengine taxiing solutions, hybrid towing taxiing solutions and electric engine solutions. This approach requires a novel concept of operation and new support tools for sustainable airport ground operations to cope with the additional vehicles on the ground to tow aircraft, discrepancies between aircraft in terms of ground speed according to the taxiing technique and the management of the fleet of towing vehicles. In this paper we first describe the motivation and the context of the project. We then introduce the AEON concept of operation that includes a new role responsible for the supervision of towing vehicles. We also articulate the architecture of our solution and describe three inter-operating support tools: (1) an optimization tool to estimate the number of necessary tugs and assign them to the aircraft before the operations; (2) a multiagent path planning system providing real-time, conflict-free routes including speed profiles to maximize capacity as well as fuel efficiency. (3) Human-Machine Interfaces for supporting Air Traffic Controllers and the Tug Fleet Manager to control and supervise the whole traffic. Then, we present the validation activities including an evaluation with Air Traffic Controllers using realistic real-time simulations on the Paris Charles de Gaulle airport. Finally, we conclude with a discussion on how this concept could impact on a number of key performance indicators (KPA), such as human performance, safety, capacity, environmental impact and proactive liability allocation in case of an accident
Air-rail timetable synchronization for a seamless passenger journey
No full textInternational audienceThis paper proposes a method to generate an integrated air-rail timetable at a hub airport with direct access to a train station. A passenger-oriented metric is introduced to assess the connection time between trains and flights. In order not to impact severely initial flight and train schedules, only small perturbations on the initial timetable are authorized. An integer linear programming formulation is proposed based on this metric. An approached resolution method is implemented to solve the optimization problem. Solution quality and computational time are compared with an exact resolution method. Computational results on the case study of Paris-Charles de Gaulle airport are presented. Results show that a change of an average 11 minutes in schedules could increase passenger comfort by almost 10%
