A digital twin-driven computation and analysis framework for low-altitude airspace

This study aims to address critical challenges in low-altitude airspace management, including high dynamic complexity, substantial safety hazards, and information fragmentation, and proposes a digital twin-enabled computation and analysis framework. The proposed framework adopts a four-layer digital twin architecture, enabling high-fidelity spatial mapping through multi-source heterogeneous data fusion, unified spatiotemporal representation, dynamic evolution modeling, and bidirectional physical-virtual closed-loop interaction. The proposed framework also incorporates intelligent components: a bidirectional GRU-Seq2Seq trajectory prediction model and a Kalman filter-based error compensation mechanism, supporting quantitative resource assessment, real-time trajectory analysis, and conflict warning. The proposed framework is verified experimentally in three operational scenarios: conventional single-target flight, multi-target cooperative flight, and extreme weather interference. Experimental results demonstrate that the proposed method has an average trajectory prediction error of 1.52 m under standard conditions, achieving a reduction of 47.9% compared to the Transformer baseline (2.36 m), 38.7% compared to ST-GCN (2.48 m), and 42.2% compared to the traditional geometric twin (2.63 m). Furthermore, the proposed method consistently maintains a low error across varying prediction horizons (5-15 s), with a maximum error of only 3.9 m at 15 s, significantly outperforming Transformer (4.3 m), ST-GCN (4.6 m), and the traditional geometric twin (6.3 m) models. In challenging scenarios involving GPS denial and communication interruption, the proposed method maintains superior robustness, with an average error of 2.15 and 2.38 m, respectively, which are markedly lower than those of competing methods. The results confirm that the proposed method outperforms both traditional geometric twin and deep learning-based approaches in key performance metrics, including position accuracy, prediction stability, and robustness, which validates its efficacy and practical suitability for low-altitude airspace management tasks.