LSTM-Driven Predictive Mobility Management with Dynamic UAV Deployment for Air–Ground Integrated IoV Networks

Ultra-dense networking (UDN) in air–ground integrated Internet of Vehicles (IoV) networks promises substantial gains in network capacity. However, the mobility of both unmanned aerial vehicle (UAV) base stations and ground vehicles makes handovers increasingly frequent and drives up signaling overhead, which compromises the low-latency and high-reliability requirements of IoV applications. To address this issue, this paper proposes an active mobility management scheme based on virtual air–ground cells (VAGC) for air–ground integrated IoV networks. We construct a long short-term memory (LSTM) neural network to predict vehicle trajectories, and we enhance the prediction accuracy by incorporating distance and azimuth information. Within the software-defined networking (SDN) controller, we design a multifunctional coordination module that creates VAGCs according to the prediction results and optimizes UAV deployment, thereby enabling timely updates of vehicle-specific UAV activation sets. Simulation results show that the proposed scheme reduces handover latency and lowers signaling overhead by approximately 35% compared with conventional passive mobility management schemes. The handover failure rate (HFR) decreases as the capacity-tolerance parameter δ increases. The performance metric reaches its maximum value of 0.53 when the prediction period is 3 s.