基于改进双向长短时记忆网络的自动驾驶车辆驾驶意图识别

Vehicle Driving Intent Recognition Based on Enhanced Bidirectional Long Short-Term Memory Network

针对高速混行多车交互环境下车辆驾驶意图识别模型大多忽视驾驶风格和车-车交互信息等问题,提出一种基于改进双向长短时记忆(Bi LSTM)网络的驾驶意图识别模型,以目标车辆轨迹序列、驾驶风格、周围车辆的交互特征作为模型的输入进行训练学习,实现对考虑驾驶风格的驾驶意图特征数据集的分类识别,同时使用鲸鱼优化算法对隐含层节点数和学习率等超参数进行寻优,以规避人工调参的负面影响。最后,使用NGSIM数据集对该模型的有效性进行验证,结果表明,模型的识别准确率达到97.5%,证明其在识别车辆驾驶意图方面具有较高的准确性。

In the context of high-speed mixed traffic and intricate multi-vehicle interaction,existing driving intention recognition models for research vehicles mostly neglect driving styles and vehicle-vehicle information interaction,this paper introduced a novel driving intention recognition model based on an enhanced Bidirectional Long Short-Term Memory(Bi LSTM)network,with the driving trajectory sequence of the target vehicle,driving style,and interaction features of surrounding vehicles as inputs for effective training and learning,to facilitate the classification and recognition of the driving intention feature dataset,specifically considering diverse driving styles.Additionally,the Whale Optimization Algorithm(WOA)was employed to optimize hyperparameters,encompassing the number of hidden layer nodes and learning rate,effectively mitigating the adverse impacts of manual parameter adjustment.The model’s efficacy was validated using the NGSIM dataset,exhibiting an impressive recognition accuracy of 97.5%in precisely identifying vehicle driving intentions.