基于轨迹的操作（TBO）中的概率约束预测与不确定性量化.pdf

1、?Probabilistic Constraints Prediction with Uncertainty Quantification in Trajectory Based Operations(TBO)Paolino DE FALCOMehtap KARAARSLANThis project has received funding from the SESAR Joint Undertaking(JU)under grant agreement No 101114701.The JU receives supportfrom the European Unions Horizon E

2、urope researchand innovation programmeand the SESAR JU members other than the Union.The Challenge?Probabilistic Machine Learning Approach Data from the entire 2024 Pre-processing agreed and flown flight trajectories to label single way points with possible constraints Only trajectory input data Bina

3、ry classifier CatBoost algorithm Main output:probability of a constraint to be applied(positive class)and not(negative class)?Uncertainty Quantification in Machine Learning We can improve the users decision making by providing more insights into the quality of individual predictions Predictions can

4、be affected by data uncertainty and knowledge uncertainty Data uncertainty occurs due to noise and class overlap in the data Knowledge uncertainty happens due to mismatch between training and testing dataset.It can be reduced by increasing the training data Different techniques to quantify uncertain

5、ties:single model vs ensembles(*),(*),(*)The entropy of a discrete probability distribution is a measure of uncertainty Data and total uncertainties can be quantified?Our approach:Bootstrapping with 50 models(*)Malinin,Andrey.Uncertainty estimation in deep learning with application to spoken languag

6、e assessment.Diss.2019.(*)De Falco,P.,&Karaarslan,M.Probabilistic Constraints Prediction with Uncertainty Quantification in Trajectory Based Operations(TBO).SESAR Innovation Days 2024(*)Roadmap,E.A.I.(2021).EASA Concept Paper:First usable guidance for Level 1