Abstract: as insufficient data diversity, dificulty in acquiringcritical data, high model complexity and poor generalization, this paper proposes a light gradient boosting ma-chine (LightGBM)-based phased multimodapn method (LightGBM-based PMTPM). Themethod can intelligently identify the flight phabased on data from the aircraft's own sensors,predict the 4D trajectory and real-time qualig an onboard computer. Experimental resultsshow that, during all flight phases, the LightGoutperforms the back propagation neural net-work-based phased multimodal trajectory predN-based PMTPM) in predictive performance,with root mean square error (RMSE) reductions of 64.86%, 13.15%, 80.88%, 77.46%, 86.45%, 3.46% and19.22%, respectively. The average evaluation time of the LighiGBM-based PMTPM is 59.890 ms, meeting theaccuracy and real-time requirements for 4D trajectory prediction of aircraft.

Key words: four-dimensional (4D) trajectory prediction, trajectory-based operation (TBO), light gradient boosting machine(LightGBM), multimodal trajectory prediction