A Probabilistic Framework for Real-Time Processing of Aircraft Surface Movement Data in Digital Twins

This study introduces a probabilistic framework for real-time processing of aircraft surface movement data within an airport digital twin. A key challenge in creating accurate digital representation of airport operations is the noise and potential missing data in streaming surveillance systems (e.g., A-SMGCS). To address this, we model aircraft ground movement as a Markov Decision Process (MDP) and incorporate a pre-trained, contextaware speed model, learned via imitation learning, that provides a prior probability distribution over feasible aircraft speeds [1]. This speed model, trained on historical data, captures realistic taxiing behavior and acts as a constraint within an optimization framework. This framework minimizes a objective that balances the fit to the current (noisy) surveillance observations with the likelihood of the estimated speeds given the prior distribution from the speed model. By incorporating this prior knowledge, the system mitigates the impact of noise and missing data. Experiments using simulated data (generated to reflect the noise characteristics of real A-SMGCS data from Singapore Changi Airport) demonstrate that the proposed approach significantly outperforms baseline methods, including a simple projectionbased method and a Kalman filter, achieving nearly a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{5 0 \%}$</tex> reduction in the 95th percentile, 99th percentile, and maximum distance errors between estimated and ground truth aircraft locations. This improved accuracy in real-time tracking is essential for enabling core digital twin functionalities, such as enhanced situational awareness, operational optimization, and predictive capabilities for airport ground operations. The approach represents a crucial step towards more robust and reliable airport digital twins.