基于高斯混合隐马尔科夫模型与人工神经网络的紧急换道行为预测方法  

Vehicle Emergency Lane-changing Behavior Prediction Method Based on GMM-HMM and ANN

作　　者：于扬 梁军[1,2] 陈龙 陈小波[1] 朱宁[2] 华国栋 YU Yang;LIANG Jun;CHEN Long;CHEN Xiaobo;ZHU Ning;HUA Guodong(Automotive Engineering Research Institute,Jiangsu University,Zhenjiang,Jiangsu,212013;Department of Mechanical,Shizuoka Institute of Science and Technology,Fukuroi,Shizuoka,437-0032,Japan;Jiangsu Smart Travel Future Automobile Research Institute,Nanjing,210000)

机构地区：[1]江苏大学汽车工程研究院,镇江212013 [2]静冈理工科大学机械系,静冈袋井437-0032 [3]江苏智行未来汽车研究院,南京210000

出　　处：《中国机械工程》2020年第23期2874-2882,2890,共10页China Mechanical Engineering

基　　金：国家重点研发计划资助项目(2018YFB1600500);国家自然科学基金资助项目(U1664258,51875255,61601203);江苏省六大人才高峰项目(2015-DZXX-048);江苏省高等学校自然科学研究重大项目(18KJA580002)。

摘　　要：为了有效降低因驾驶员紧急换道行为而诱发的交通事故,提高道路交通事故链阻断效率,提出一种基于高斯混合隐马尔科夫模型(GMM-HMM)和人工神经网络(ANN)的紧急换道行为预测方法。首先利用GMM-HMM对车辆行驶状态以及驾驶行为连续观察序列进行换道意图辨识,采用ANN预测下一时段的驾驶行为,再预测换道过程中的横向加速度变化率,从而判断紧急换道的危险程度。驾驶员在环仿真实验及实车实验结果表明,该方法预测避险成功率达92.83%,实验避险成功率达90.32%。该方法能有效地对紧急换道行为进行提前警告与干预。In order to effectively reduce the traffic incidents caused by drivers’emergency lane-changing behaviors and improve the blocking efficiency of chains of road traffic incidents,a method of predicting emergency lane-changing behaviors was proposed based on GMM-HMM and ANN.First,GM-HMM was utilized to identify the intention of lane-changing based on the continuous observation sequences of vehicle states and the driving behaviors.ANN was utilized to predict the driving behaviors in the next interval,and then lateral acceleration rate during lane-changing was predicted to judge the dangerous degree of emergency lane-changing.The results of driver in-the-loop simulations and real vehicle experiments show that the predictive risk avoidance success rate is as 92.83%and the experiment risk avoidance success rate is as 90.32%.The method may effectively warn and intervene the emergency lane-changing behaviors.

关 键 词：换道行为预测 高斯混合隐马尔可夫模型 人工神经网络 道路交通事故链阻断 

分 类 号：U580.70[交通运输工程]