车辆运动轨迹的预测在车辆的自动驾驶与车联网技术中有着重要意义，通过预测轨迹可以判断车辆未来运动状态，避免发生碰撞。针对车辆换道轨迹的预测问题，提出了基于生成对抗网络的换道轨迹预测模型。通过实车实验，以城市道路中换道行为为实例，采用高精度 GPS 仪器采集车辆换道轨迹数据。在此基础上，建立基于生成对抗网络的轨迹预测模型，其中生成模型采用了 LSTM 的编码器－解码器结构，通过输入给定的历史换道轨迹，经解码器生成预测时段换道轨迹。判别模型通过搭建基于 MLP 的神经网络，将生成的预测轨迹与目标轨迹进行多重判别，并通过联合训练生成模型和判别模型，实现对车辆未来时段内的换道轨迹进行预测。同时通过交叉验证与模型对比，分析了不同长度的历史轨迹与预测轨迹对预测精度的影响，并验证了模型的有效性和准确性。结果表明轨迹生成对抗模型与传统模型相比，可实现对换道轨迹长时段的预测，且预测精度有明显的提高。

The prediction of vehicle trajectory has great significance in the autonomous vehicles and internet of ve-hicles systems． Vehicle trajectory prediction can help to judge the future motion state of vehicles and to avoid colli-sion． Therefore，a vehicle lane-change trajectory prediction model based on generative adversarial networks was suggested． Vehicle lane-changing data was collected with High-precision GPS instruments through complete vehicle test in urban highways． On this basis，a trajectory prediction model based on the generative adversarial networks was established． The generator of GAN adopts the LSTM encoder-decoder structure，and the future lane-changing trajectory is generated through the decoder by inputting the given observed lane-changing trajectories． By construc-ting neural network based on the MLP，the discriminative model can distinguish the generated trajectory and the target trajectory through multiple discriminating methods． By jointly training generative model and discriminative model，the future trajectory of single vehicle in real time can be predicted． Through cross-validation and model
comparison，the effects of historical trajectories and prediction trajectories of different lengths on prediction accura-cy were analyzed，and the validity and accuracy of the model was verified． The results show that，compared with the traditional model，our model can predict the lane-change trajectory over a long period of time with an obviously improved accuracy．