Knowledge-data-model-driven Trajectory Fusion Prediction Method for Low-altitude Moving Target

Aiming at the moving target trajectory prediction problem in low-altitude environments, a knowledge-data-model-driven trajectory fusion prediction framework for moving target is proposed. A flight knowledge mixture-of-experts model is constructed based on the kinematic characteristics of low-altitude aerial vehicles. Multi-source sensor data are fed into various flight knowledge expert modules to achieve precise identification of target maneuver modes, while spatiotemporal correlation features are extracted by using the Mamba model. A weight adaptive adjustment mechanism is designed to dynamically fuse multi-source perception data by using an attention mechanism, thereby addressing the spatiotemporal asynchrony issue of sensors. Long-term temporal dependencies are modeled by using gated recurrent unit to produce preliminary trajectory predictions based on historical flight data of target. A physics-informed neural network is constructed based on the kinematic equations of low-altitude targets. By dynamically balancing data-driven loss and physical constraint loss, the network corrects data-driven biases, ensures predicted trajectories satisfy kinematic constraints, and effectively suppresses error accumulation in multi-step prediction. Numerical simulations and experimental validation results demonstrate that the proposed knowledge-data-model-driven trajectory fusion prediction method can effectively forecast low-altitude moving target flight trajectories.