Inertial Projectile Trajectory Estimation Method Based on Conv-LSTM and Particle Filtering Fusion

The development of trajectory parameter estimation methods only based on inertial sensors under GNSS-denied scenarios plays a crucial role in enhancing the survivability of projectile correction ammunition. To alleviate the accumulation of inertial errors during flight, we propose an integrated framework that combines neural networks and particle filters. Specifically, a Conv-LSTM network is employed to learn the dynamic model of the projectile using inertial sensor data and initial trajectory parameters. The predictions generated by the network are then utilized in the particle filter for measurement correction. Furthermore, a simplified strapdown inertial navigation solution model and error model are constructed specifically for close-range correction ammunition, while the latter is employed for state update. The impact of inertial sensor noise and untrained launch angles on the prediction performance is evaluated. The simulation results demonstrate that the dual correction mechanism effectively overcomes the limitations of the traditional Dead-Reckoning algorithm. and maintains a higher level of measurement accuracy over an extended time period, which highlights the promising potential for projectile correction applications.