Dynamic airspace sectorization with machine learning enhanced workload prediction and clustering

Abstract

Addressing the complexities of modern Air Traffic Management (ATM), this paper introduces a novel framework for dynamic airspace sectorization, tailored to enhance efficiency and safety in congested airspaces. Central to this framework is the WP-ConvLSTM model, an innovative deep learning approach equipped with attention mechanisms. This model excels in accurately predicting workload dynamics, a critical factor in managing air traffic flow. To implement sectorization, we adopt a constrained K-means clustering technique for spatial division, followed by a refinement process involving Support Vector Machine (SVM) algorithms for precise boundary generation. Further optimization of sector boundaries is achieved through an evolutionary algorithm, ensuring both flexibility and stability in airspace divisions. Our methodology was thoroughly evaluated using real-world data from one of the busiest airspaces, demonstrating significant improvements in workload prediction accuracy and airspace sector management. The findings highlight the model’s robustness in practical scenarios, offering a scalable solution for ATM challenges. We conclude with a recognition of the study’s limitations and propose avenues for future research to build upon our findings, particularly in enhancing real-time data integration and adapting to evolving air traffic patterns.