The attractive feature of conveniently deployed as an aerial access point (AP) to provide coverage and improve network performance, has made the Unmanned Aerial Vehicle (UAV) networks a research hotspot. In this paper, an UAV handover mechanism in the three-dimensional space is proposed. We build and train user trajectory prediction model based on neural network. Then we described the handover mechanism under the condition of predicted trajectory and the constructed signal transmission model in UAV networks. Simulation results show that compared to the traditional handover algorithm, our scheme nearly 10% higher in handover success rate.