| 26 | 1 | 117 |
| 下载次数 | 被引频次 | 阅读次数 |
针对传统模型在网联环境下识别能力不足与换道场景复杂多变的问题，提出了一种网联环境下基于图卷积网络(Graph Convolutional Network, GCN)与双向长短期记忆网络(Bidirectional Long Short-Term Memory Network, Bi-LSTM)的半监督学习(Semi-Supervised Learning, SSL)模型SSL-GCN-LSTM。该模型以驾驶模拟器与眼动仪采集的多源行为数据为基础，采用GCN实现空间特征建模，并引入多头注意力机制以提取不同子空间中的图结构特征与节点属性；采用Bi-LSTM进行时间特征建模，以捕捉驾驶行为的时序动态特征；将空间特征与时间特征进行融合，构建基于有标签数据的换道意图识别模型。基于该模型计算无标签数据的置信度，获得伪标签损失和一致性正则化损失，通过反向传播更新模型权重。实验结果表明，所提SSL-GCN-LSTM模型使用50%标注数据的准确率为90.7%,使用100%标注数据的准确率为91.5%,优于GCN(86.3%)与LSTM(90.0%)等基线模型。代码地址为https://github.com/wuchenghao12/SSL-GCN-LSTM。
Abstract：Aiming at the limitations of traditional models in connected environments, particularly in recognizing complex and variable lane-changing scenarios, this paper proposes a semi-supervised learning model termed SSL-GCN-LSTM, which integrates Graph Convolutional Network(GCN) and Bidirectional Long Short-Term Memory Network(Bi-LSTM) for connected environments. The model is built on multi-source behavioral data collected from driving simulators and eye trackers. It employs GCN to model spatial features and incorporates a multi-head attention mechanism to extract graph structural features and node attributes across different representation subspaces. Bi-LSTM is utilized to model temporal features, capturing the sequential dynamics of driving behaviors. The spatial and temporal features are then fused to construct a lane-changing intention recognition model based on labeled data. Using this model, confidence scores for unlabeled data are computed, yielding pseudo-label loss and consistency regularization loss, which are used to update model weights via backpropagation. Experimental results demonstrate that the proposed SSL-GCN-LSTM model achieves an accuracy of 90.7% with 50% labeled data and 91.5% with 100% labeled data, outperforming baseline models such as GCN(86.3%) and LSTM(90.0%). Code is available at https://github.com/wuchenghao12/SSL-GCN-LSTM.
[1]Shangguan Q Q,Fu T,Wang J H,et al.A proactive lane-changing risk prediction framework considering driving intention recognition and different lane-changing patterns[J].Accident Analysis and Prevention,2022,164:106500.
[2]杨达，刘家威，郑斌，等.基于时域卷积网络与注意力机制的车辆换道轨迹预测模型[J].交通运输系统工程与信息，2024,24(2):114-126.
[3]Gao K,Li X H,Chen B,et al.Dual transformer based prediction for lane change intentions and trajectories in mixed traffic environment[J].IEEE Transactions on Intelligent Transportation Systems,2023,24(6):6203-6216.
[4]Shi Q,Zhang H.An improved learning-based LSTM approach for lane change intention prediction subject to imbalanced data[J].Transportation Research:Part C Emerging Technologies,2021,133:103414.
[5]Yuan R T,Abdel-Aty M,Gu X,et al.A unified modeling framework for lane change intention recognition and vehicle status prediction[J].Physica A:Statistical Mechanics and its Applications,2023,632:129332.
[6]Lin L,Li W Z,Bi H K,et al.Vehicle trajectory prediction using LSTMs with spatial-temporal attention mechanisms[J].IEEE Intelligent Transportation Systems Magazine,2021,14(2):197-208.
[7]Pan Y X,Zhang Q Y,Zhang Y F,et al.Lane-change intention prediction using eye-tracking technology:a systematic review[J].Applied Ergonomics,2022,103:103775.
[8]Li Y,Liu F,Xing L,et al.A deep learning framework to explore influences of data noises on lane-changing intention prediction[J].IEEE Transactions on Intelligent Transportation Systems,2024,25(7):6514-6526.
[9]Peng M X,Guo X S,Chen X D,et al.Lc-llm:explainable lane-change intention and trajectory predictions with large language models[J].Communications in Transportation Research,2025,5:100170.
[10]Khelfa B,Ba I,Tordeux A.Predicting highway lane-changing maneuvers:a benchmark analysis of machine and ensemble learning algorithms[J].Physica A:Statistical Mechanics and its Applications,2023,612:128471.
[11]Huang T,Fu R,Sun Q Y,et al.Driver lane change intention prediction based on topological graph constructed by driver behaviors and traffic context for human-machine co-driving system[J].Transportation Research:Part C Emerging Technologies,2024,160:104497.
[12]黄海峰，黄德启，黄德意，等.高速公路场景下的车辆换道意图预测研究[J].计算机应用研究，2025,42(12):3574-3581。
[13]孙秦豫，周航，付锐，等.基于驾驶人认知决策空间的换道意图预测[J].汽车工程，2025,47(8):1468-1478,1572.
[14]高凯，李勋豪，胡林，等.基于多头注意力的CNN-LSTM的换道意图预测[J].机械工程学报，2022,58(22):369-378.
[15]Jain A,Singh A,Koppula H S,et al.Recurrent neural networks for driver activity anticipation via sensory-fusion architecture[C]//2016 IEEE International Conference on Robotics and Automation (ICRA).Stockholm:IEEE,2016:3118-3125.
[16]Zhang S,Zheng D Q,Hu X C,et al.Bidirectional long short-term memory networks for relation classification[C]//Proceedings of the 29th Pacific Asia Conference on Language,Information and Computation.Shanghai:Association for Computation Linguistics,2015:73-78.
[17]Yan S,Liang Y,Wang B L.Multi-level deep learning kalman filter[C]//2023 International Conference on Advanced Robotics and Mechatronics (ICARM).Stockholm:IEEE,2023:1113-1118.
基本信息:
DOI：10.13367/j.cnki.sdgc.2026.06.007
中图分类号:U495;TP18
引用信息:
[1]吴承昊,刘家珲,陈建钊,等.基于SSL-GCN-LSTM的换道意图识别方法[J].山东理工大学学报(自然科学版),2026,40(06):17-23.DOI:10.13367/j.cnki.sdgc.2026.06.007.
基金信息:
山东省重点研发计划项目(2024CXGC010301); 山东省自然科学基金项目(ZR2023QF028); 山东省交通科技项目(2025BAI26)
2025-10-30
2025
2025-11-13
2026-05-18
2026
1
2026-07-21
2026-07-21