鲸鱼算法优化CNN-BiGRU-ATTENTION的车辆换道意图识别模型
Research on vehicle lane change intention recognition model CNN-BIGRU-ATTENTION optimized by whale algorithm
朱孙科 1严健容 1熊开洋 1熊钊 1安邦1
作者信息
- 1. 重庆交通大学 机电与车辆工程学院,重庆 400074
- 折叠
摘要
针对时空多样特征、超参数敏感性影响车辆行为意图识别精度的问题,提出一种改进的CNN-BiGRU-ATTEN-TION混合换道意图识别模型.采用目标车辆轨迹序列和与周围车辆互动特征作为模型输入进行训练,实现考虑车辆动态变化状态的意图识别预测;使用鲸鱼优化算法对模型调整参数进行多目标寻优,降低模型调优难度;利用NG-SIM数据集对模型进行评估校验.结果表明:所提出的WOA-CNN-BiGRU-ATTENTION模型与CNN-BiGRU-ATTEN-TION模型、Transformer模型相比,准确率分别提升了4.53％、0.97％,达到97.64％;WOA-CNN-BiGRU-ATTENTION模型在不同预判时间下的意图识别准确率最高,在换道前2.5 s的识别精度均能达到91％以上,证明模型具有较强的车辆换道意图识别性能.
Abstract
To address the problem of spatio-temporal diverse features and hyperparameter sensitivity affecting the accuracy of vehicle behavioral intention recognition,this paper proposes an improved CNN-BiGRU-ATTENTION hybrid lane-changing intention recognition model.First,target vehicle trajectory sequences and interaction features with surrounding vehicles are used as model inputs for training to achieve intention recognition prediction considering the dynamic change state of vehicles.Then,a whale optimization algorithm is used to perform multi-objective optimization of model tuning parameters to reduce the difficulty of model tuning.Finally,the model is evaluated and calibrated using the NGSIM dataset. Our results show the accuracy of the proposed WOA-CNN-BiGRU-ATTENTION model is improved by 4.53％ and 0.97％ to 97.64％ compared with the CNN-BiGRU-ATTENTION model and the Transformer model.The WOA-CNN-BiGRU-ATTENTION model with different prediction times achieves the highest accuracy in the intention recognition with the recognition accuracy in the 2.5 s before lane changing exceeding 91％,demonstrating the model performs well in intention recognition of vehicles changing lanes.
关键词
自动驾驶/换道意图识别/鲸鱼算法/双向门控循环单元/注意力机制Key words
autonomous driving/lane change intent recognition/whale algorithm/bidirectional gated circulation unit/attention mechanism引用本文复制引用
基金项目
重庆市高校创新研究群体项目(CXQT20019)
出版年
2024
国家科技期刊平台
NSTL
万方数据