计算机工程与应用 ›› 2022, Vol. 58 ›› Issue (13): 280-287.DOI: 10.3778/j.issn.1002-8331.2111-0052

• 工程与应用 • 上一篇    下一篇

考虑换道意图的LSTM-AdaBoost车辆轨迹预测模型

孟宪伟,唐进君,王喆   

  1. 中南大学 交通运输工程学院,长沙 410075
  • 出版日期:2022-07-01 发布日期:2022-07-01

Trajectory Prediction of Vehicles Based on LSTM-AdaBoost Model Considering Lane-Changing Intention

MENG Xianwei, TANG Jinjun, WANG Zhe   

  1. School of Traffic and Transportation Engineering, Central South University, Changsha 410075, China
  • Online:2022-07-01 Published:2022-07-01

摘要: 不合理的车辆的换道行为是导致交通事故发生的主要原因之一,提前预知换道车辆的轨迹并及时做出相应调整有助于减少事故的发生。针对换道车辆轨迹预测问题,采用将深度学习和集成学习相结合的轨迹预测方法,并考虑了换道意图的影响。建立连续隐马尔可夫模型对车辆进行换道意图检测,提前判别车辆的换道状态,并输入至相应的轨迹预测模型中;将LSTM(long short term memory)作为AdaBoost算法(adaptive boosting)的基预测器,建立LSTM-AdaBoost模型,在多个基预测器同时进行轨迹预测的基础上,通过训练调整各个基预测器的权重并将结果加权集成,提升预测模型的精度和稳定性;通过NSGIM(next generation simulation)数据集对模型进行训练和测试,结果显示意图预测模型在变道前一秒的准确率在90%以上,LSTM-AdaBoost集成轨迹预测模型与单一的LSTM模型相比精度和稳定性显著提升,且预测结果中异常数据更少,具有较好的稳定性;同时预测对比结果也表明增加意图预测模块有助于提升换道轨迹预测的精度。

关键词: 车辆换道轨迹预测, 换道意图识别, 隐马尔可夫模型, 长短期记忆网络, AdaBoost集成算法

Abstract: Improper vehicle lane-changing behavior is one of the main causes of traffic accidents. Predicting the trajectory of lane-changing vehicles in advance and making corresponding adjustments in time will help to reduce the occurrence of accidents. A trajectory prediction method combining deep learning and ensemble learning is proposed for the lane-changing vehicles, the influence of lane-chang intention on trajectory prediction is considered at the same time. Firstly, establish continuous hidden Markov model(CHMM) to detect the lane-changing intention of the vehicles through their past trajectory data in real time. The driving state of the vehicle is predicted in advance through the past trajectory data of the vehicle, and the trajectory sequence predicted as lane-change is input into the corresponding trajectory prediction model. Secondly, taking LSTM(long short-term memory) as the base predictor of adaptive boosting algorithm, the LSTM-AdaBoost model is established. Based on the trajectory prediction of multiple base predictors at the same time, the weights of each base predictor are adjusted through training and the results are weighted and integrated to improve the accuracy and stability of the prediction model. Finally, the model is trained and tested through the NGSIM(next generation simulation) data set collected in the real driving environment. The results show that the accuracy of the intention prediction model is more than 90% one second before lane change, the accuracy and stability of the LSTM-AdaBoost integrated trajectory prediction model are significantly improved compared with the single LSTM prediction model. And there are fewer abnormal data samples in the prediction results, indicate that the model has good stability. At the same time, the prediction comparison results also show that adding the intention prediction module is helpful to improve the accuracy of lane-changing trajectory prediction.

Key words: vehicle lane-changing trajectory prediction, lane-changing intention identification, hidden Markov model(HMM), long short-term memory network(LSTM), AdaBoost ensemble algorithm