无线电工程

5G网络是车联网(Internet of Vehicles, IoV)发展的关键一步,其低时延的特点可以实现自动驾驶车辆对前方实时交通信息感知的需求,为预先制动、提前绕行等驾驶行为提供参考。针对混合交通场景中手动驾驶车辆可能会出现的危险驾驶行为,基于长短时记忆(Long Short-Term Memory, LSTM)网络和无线集群学习框架,提出了一种基于车联网的无线集群智能轨迹预测(Swarm Learning-based Trajectory Prediction, SLTP)算法。SLTP算法以智能网联汽车为研究对象,在使用去中心化的无线集群学习保护用户隐私数据的同时,使用车车无线通信系统感知周边手动驾驶车辆的历史轨迹信息并给出轨迹预测。通过使用美国高速公路行车NGSIM(Next Generation Simulation)的真实交通数据集评估SLTP的轨迹预测准确性能,评估结果表明,与现有的基于LSTM网络轨迹预测方法相比,SLTP算法在同样的无线通信开销下对于周边手动驾驶车辆的轨迹预测误差降低了43.1%,表明SLTP算法在车联网场景下的轨迹预测应用中具有良好的鲁棒性和准确性。

5 G network is critical in developing the Internet of Vehicles(IoV).Its low delay characteristics can make autonomous vehicles realize the real-time traffic information perception and select driving behaviors ahead, such as pre braking and early detour.To address the risk caused by manually driving vehicles in mixed traffic scenarios, a wireless Swarm Learning Trajectory Prediction(SLTP) algorithm based on the Internet of vehicles is proposed, which is based on Long Short-Term Memory(LSTM) and wireless swarm learning.The SLTP algorithm takes the Automatic CV as the investigated subject.While using the decentralized Wireless Swarm Learning framework to protect the private data of users, SLTP predicts future trajectories of surrounding vehicles by using the vehicle-vehicle wireless communication system to aware the history information of the surrounding manual driving vehicles.Multiple experiments are done under the Next Generation Simulation(NGSIM),the actual traffic dataset from America is used to evaluate the accuracy of SLTP prediction results.The comparison results show that SLTP can improve the trajectory prediction accuracy by 43.1% compared with the existing methods based on LSTM,indicating that SLTP has better robustness and performance in trajectory prediction in IoV scenario.