GPNet: A Deep Learning Guided Projectiles Navigation Framework

High-precision navigation of guided projectiles is the basis of accurate guidance and control for projectiles. Microinertial navigation system (INS) of projectiles is faced with high complexity such as satellite interference and random wind disturbance, and high dynamics such as high spinning (≥20 r/s) and high overload (≥10 000 g). Thus, we propose a GPNet, a deep learning guided projectiles navigation framework, which is suitable for autonomous navigation and positioning of guided projectiles in the whole process. This framework mainly includes three parts: data enhancement, deep learning network, and fusion filtering. According to projectile flight characteristics, we propose a trajectory data enhancement method based on geometric constraints. Deep learning network is used to establish a trajectory increment prediction model. Then, a trajectory increment error filtering model is established to complete data fusion of deep learning prediction results and INS calculation results, so as to realize high-precision autonomous navigation and positioning for guided projectiles in the whole process. Simulation and experiment results show that compared with traditional methods, the positioning accuracy of GPNet is improved by more than 16%.