https://enac.hal.science/hal-02388282Mori, RyotaRyotaMoriDelahaye, DanielDanielDelahayeENAC - Ecole Nationale de l'Aviation CivileSimulation-Free Runway Balancing Optimization Under Uncertainty Using Neural NetworkHAL CCSD2019simulated annealingarrival managerconvolutional neural network[MATH.MATH-OC] Mathematics [math]/Optimization and Control [math.OC]Porte, Laurence2019-12-01 16:22:142021-11-03 04:52:442019-12-03 13:47:34enConference papersapplication/pdf1This paper proposes a new optimization scheme using neural network for runway balancing to minimize departure and arrival aircraft delay. While other researchers have proposed solutions to the runway balancing problem using a simulation-based technique to calculate aircraft delay, the proposed method replaces the simulation by a neural network model-based estimation using the actual operational data, thus providing the following two advantages. First, accurate estimation of aircraft delay can improve the solution of the runway balancing problem. Second, the simulation process is not required in the optimization. Although it is difficult to develop an accurate simulation model especially under uncertain environment, the neural network model can estimate the average delay without explicitly modeling uncertainty. In this paper, as a first step, the effectiveness of the proposed method is validated through simulations. First, simulations considering uncertainty are used to generate the data, which are then used to train the neural network. The neural network predicts the delay under the current traffic and only this predicted delay is used for the runway balancing optimization with simulated annealing. The simulation result shows that the result by neural network outperforms the one by the simulation-based method under uncertainty. This means that the neural network can accurately estimate the delay under uncertainty environment, and is applicable in the optimization process.