A LightGBM Framework for Rapid Flight Time Prediction with High-Dynamic Validation

Abstract

To meet the urgent need for real-time and accurate flight time prediction in intelligent aeronautical systems, this paper proposes a rapid prediction framework for aerial vehicle flight time based on the Light Gradient Boosting Machine (LightGBM). To validate the method's effectiveness, a high-dynamic autonomous flight scenario with a well-defined dynamic model was selected. Experimental results demonstrate that, compared to traditional physics-based numerical integration, the trained LightGBM model maintains prediction accuracy while reducing the time per prediction by approximately two orders of magnitude to the millisecond level. Furthermore, the model's lightweight nature helps reduce the energy consumption of computational tasks, aligning with sustainable computing principles. The proposed framework is generalizable, with its technical pathway also applicable to other aeronautical fields requiring rapid time prediction, such as estimating time of arrival in civil aviation.