Please wait a minute...
浙江大学学报(工学版)
浙江大学学报(工学版)  2019, Vol. 53 Issue (6): 1171-1181    DOI: 10.3785/j.issn.1008-973X.2019.06.017
计算机与自动化技术     
采用梯度提升决策树的车辆换道融合决策模型
徐兵(),刘潇,汪子扬,刘飞虎,梁军*()
浙江大学 控制科学与工程学院，浙江 杭州 310058
Fusion decision model for vehicle lane change with gradient boosting decision tree
Bing XU(),Xiao LIU,Zi-yang WANG,Fei-hu LIU,Jun LIANG*()
College of Control Science and Engineering, Zhejiang University, Hangzhou 310058, China
 全文: PDF(1152 KB)   HTML
摘要：

车辆在执行换道行为时，由于受到较多环境因素影响，难以准确进行换道识别和预测. 为解决这一问题，提出一种基于梯度提升决策树（GBDT）进行特征变换的融合换道决策模型，以仿真驾驶员在高速公路上自由换道时的决策行为. 采用主体车辆与目标车道后车的碰撞时间 tlag 及车辆周围交通状态变量进行车辆换道行为的建模分析，在NGSIM数据集上对建立的融合换道决策模型进行参数标定和模型测试. 实验结果表明：融合换道决策模型以95.45%的预测准确率超越支持向量机、随机森林和GBDT等单一的换道决策模型，获得了最突出的表现. 变量分析结果表明：新引入的换道决策变量 tlag 对车辆换道行为具有重要影响. 提出的融合换道决策模型能够进一步减少因换道决策误判而导致的交通事故.
关键词： 梯度提升决策树（GBDT）;  自由换道行为;  NGSIM数据集;  换道决策模型;  碰撞时间    
Abstract:

When the vehicle performs lane changing behavior, it is challenging to accurately identify and predict the vehicle’s behavior due to the influence of various environmental factors. In order to solve this problem, a fusion lane changing decision model was proposed; the gradient boosting decision tree (GBDT) was applied for feature transformation. This fusion model was applied to simulate the driver’s decision-making behavior while freely changing lanes on the expressway. The collision time tlag of the main vehicle and lag vehicle on the target lane and other vehicle traffic state variables were introduced into the model to analyze the lane changing behavior. The parameter calibration and the test of the established fusion lane changing decision model were carried out on the NGSIM (Next Generation Simulation) dataset. The experimental results show that the proposed fusion lane changing decision model surpasses the single lane changing decision model, like support vector machine, random forest and GBDT, with prediction accuracy of 95.45%, giving the most outstanding performance. The variable analysis results show that the newly introduced lane change decision variable tlag plays a positive role in the vehicle’s lane changing behavior. The proposed fusion lane changing decision model is able to further reduce the traffic accidents caused by misjudgment of lane changing decisions.
Key words: gradient boosting decision tree (GBDT)    free lane changing behavior    NGSIM dataset    lane changing decision model    collision time
收稿日期: 2018-12-27 出版日期: 2019-05-22
CLC:  TP 181  
通讯作者: 梁军     E-mail: bingxu@zju.edu.cn;jliang@zju.edu.cn
作者简介: 徐兵（1995—），男，硕士生，从事无人驾驶和机器学习研究. orcid.org/0000-0003-3795-7263. E-mail: bingxu@zju.edu.cn
服务　　
把本文推荐给朋友
加入引用管理器
E-mail Alert
作者相关文章　　
徐兵
刘潇
汪子扬
刘飞虎
梁军

引用本文:

徐兵,刘潇,汪子扬,刘飞虎,梁军. 采用梯度提升决策树的车辆换道融合决策模型[J]. 浙江大学学报(工学版), 2019, 53(6): 1171-1181.

Bing XU,Xiao LIU,Zi-yang WANG,Fei-hu LIU,Jun LIANG. Fusion decision model for vehicle lane change with gradient boosting decision tree. Journal of ZheJiang University (Engineering Science), 2019, 53(6): 1171-1181.

链接本文:

http://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2019.06.017    或    http://www.zjujournals.com/eng/CN/Y2019/V53/I6/1171

图 1  基于梯度提升决策树（GBDT）的融合决策模型结构图
图 2  NGSIM数据集研究区域
图 3  No. 260车辆数据滤波结果
图 4  主体车辆自由换道行为示意图
特征编号 符号 特征含义
1 ${v_{\rm{s}}}$ 主体车辆速度
2 ${a_{\rm{s}}}$ 主体车辆加速度
3 ${d_{{\rm{lead}}}}$ 目标车道前车与主体车辆的距离
4 $\Delta {v_{{\rm{lead}}}}$ 目标车道前车与主体车辆速度差
5 ${d_{{\rm{lag}}}}$ 目标车道后车与主体车辆的距离
6 $\Delta {v_{{\rm{lag}}}}$ 目标车道后车与主体车辆速度差
7 ${d_{{\rm{lc}}}}$ 当前车道前车与主体车辆的距离
8 $\Delta {v_{{\rm{lc}}}}$ 当前车道前车与主体车辆速度差
9 ${d_{{\rm{fc}}}}$ 当前车道后车与主体车辆的距离
10 $\Delta {v_{{\rm{fc}}}}$ 当前车道后车与主体车辆速度差
11 $\Delta {a_{{\rm{lead}}}}$ 目标车道前车与主体车辆加速度差
12 $\Delta {a_{{\rm{lag}}}}$ 目标车道后车与主体车辆加速度差
13 $t_{\rm lc}$ 主体车辆与当前车道前车碰撞时间
14 ${t_{{\rm{lead}}}}$ 主体车辆与目标车道前车碰撞时间
15 ${t_{{\rm{lag}}}}$ 主体车辆与目标车道后车碰撞时间
16 $\Delta {a_{{\rm{lc}}}}$ 当前车道前车与主体车辆加速度差
17 $\Delta {a_{{\rm{fc}}}}$ 当前车道后车与主体车辆加速度差
表 1  自由换道决策行为特征变量
图 5  特征变量在不同评价指标下的重要程度
换道决策模型 参数寻优范围 参数说明
SVM “C”:（0.001,60） C：误差惩罚项参数
“gamma”:（0.000 1,2） Gamma：高斯核函数的参数
RF “n_estimators”:（100，1 000） n_estimators：树的数量
“max_depth”:（3,10） max_depth：树的最大深度
“max_features”:（0.1，0.999） max_features：寻找最佳分裂点时属性数量占比
“min_samples_split”:（2,10） min_samples_split：分裂节点所需最少样本数
GBDT “learning_rate”:（0.001，0.2） learning_rate：算法学习率
“n_estimators”:（100,2 000） n_estimators：树的数量
“subsample”:（0.1,1） subsample：训练样本下采样比例
“max_depth”:（5,10） max_depth：树的最大深度
“min_samples_leaf”:（1,12） min_samples_leaf：分裂节点所需最少样本数
表 2  不同换道决策模型的参数寻优范围
图 6  不同换道决策模型的参数优化过程
决策模型 ${R_{{\rm{acc}}}}$ ${R_{{\rm{TP}}}}$ ${R_{{\rm{TN}}}}$
SVM 81.82 72.79 85.30
RF 92.99 93.20 92.91
GBDT 93.56 91.84 94.23
融合决策模型 95.45 95.24 95.54
表 3  不同决策模型在测试集上的表现
特征变量组 ${R_{{\rm{acc}}}}$ ${R_{{\rm{TP}}}}$ ${R_{{\rm{TN}}}}$
[1，3，4，5，6，11，12] 85.42 89.12 83.99
[1，3，4，5，6，11，12，14] 88.64 89.12 88.45
[1，2，···，17] 95.45 95.24 95.54
表 4  不同特征变量组对比实验
图 7  基于GBDT的融合决策模型特征选择过程
1 DOU Y, YAN F, FENG D. Lane changing prediction at highway lane drops using support vector machine and artificial neural network classifiers [C] // 2016 IEEE International Conference on Advanced Intelligent Mechatronics (AIM). Banff: IEEE, 2016: 901–906.
2 张元良 基于Stackelberg博弈理论的自主性车道变换模型[J]. 交通运输系统工程与信息, 2014, 14 (5): 67- 73
ZHANG Yuan-liang Discretionary lane changing modeling based on Stackelberg game theory[J]. Journal of Transportation Systems Engineering and Information Technology, 2014, 14 (5): 67- 73
doi: 10.3969/j.issn.1009-6744.2014.05.010
3 GIPPS P G A model for the structure of lane-changing decisions[J]. Transportation Research, Part B (Methodological), 1986, 20 (5): 403- 414
doi: 10.1016/0191-2615(86)90012-3
4 YANG Q, KOUTSOPOULOS H N A microscopic traffic simulator for evaluation of dynamic traffic management systems[J]. Transportation Research-Part C Emerging Technologies, 1996, 4 (3): 113- 130
doi: 10.1016/S0968-090X(96)00006-X
5 AHMED K I. Modeling drivers' acceleration and lane changing behavior [D]. Cambridge: Massachusetts Institute of Technology, 1999.
6 TOLEDO T, KOUTSOPOULOS H N, BEN-AKIVA M Integrated driving behavior modeling[J]. Transportation Research Part C Emerging Technologies, 2007, 15 (2): 96- 112
doi: 10.1016/j.trc.2007.02.002
7 SCHUBERT R, SCHULZE K, WANIELIK G Situation assessment for automatic lane-change maneuvers[J]. IEEE Transactions on Intelligent Transportation Systems, 2010, 11 (3): 607- 616
doi: 10.1109/TITS.2010.2049353
8 WANG E, SUN J, JIANG S, et al Modeling the various merging behaviors at expressway on-ramp bottlenecks using support vector machine models[J]. Transportation Research Procedia, 2017, 25: 1327- 1341
doi: 10.1016/j.trpro.2017.05.157
9 MOTAMEDIDEHKORDI N, AMINI S, HOFFMANN S, et al. Modeling tactical lane-change behavior for automated vehicles: a supervised machine learning approach [C] // 2017 IEEE International Conference on Models and Technologies for Intelligent Transportation Systems (MT-ITS). Naples: IEEE, 2017: 268–273.
10 DíAZ-áLVAREZ A, CLAVIJO M, JIMéNEZ F, et al Modelling the human lane-change execution behaviour through Multilayer Perceptrons and Convolutional Neural Networks[J]. Transportation research part F: traffic psychology and behaviour, 2018, 56: 134- 148
doi: 10.1016/j.trf.2018.04.004
11 FRIEDMAN J H Greedy function approximation: a gradient boosting machine[J]. Annals of Statistics, 2001, 29 (5): 1189- 1232
12 HE X, PAN J, JIN O, et al. Practical lessons from predicting clicks on ads at facebook [C] // Proceedings of the Eighth International Workshop on Data Mining for Online Advertising. New York: ACM, 2014: 1–9.
13 PIETRA S A D, PIETRA S A D, PIETRA S A D. A maximum entropy approach to natural language processing[M]. Cambridge: MIT Press, 1996.
14 HOU Y, EDARA P, SUN C Modeling Mandatory Lane Changing Using Bayes Classifier and Decision Trees[J]. IEEE Transactions on Intelligent Transportation Systems, 2014, 15 (2): 647- 655
doi: 10.1109/TITS.2013.2285337
15 COIFMAN B, LI L A critical evaluation of the Next Generation Simulation (NGSIM) vehicle trajectory dataset[J]. Transportation Research Part B: Methodological, 2017, 105: 362- 377
doi: 10.1016/j.trb.2017.09.018
16 OSSEN S, HOOGENDOORN S Validity of trajectory-based calibration approach of car-following models in presence of measurement errors[J]. Transportation Research Record: Journal of the Transportation Research Board, 2008, (2088): 117- 125
17 PUNZO V, FORMISANO D J, TORRIERI V Nonstationary kalman filter for estimation of accurate and consistent car-following data[J]. Transportation research record, 2005, 1934 (1): 2- 12
doi: 10.1177/0361198105193400101
18 WANG Q, LI Z L L Investigation of Discretionary Lane-Change Characteristics Using Next-Generation Simulation Data Sets[J]. Journal of Intelligent Transportation Systems, 2014, 18 (3): 246- 253
doi: 10.1080/15472450.2013.810994
19 MENG Q, WENG J Classification and regression tree approach for predicting drivers’ merging behavior in short-term work zone merging areas[J]. Journal of Transportation Engineering, 2012, 138 (8): 1062- 1070
doi: 10.1061/(ASCE)TE.1943-5436.0000412
20 YANG D, ZHU L, RAN B, et al Modeling and analysis of the lane-changing execution in longitudinal direction[J]. IEEE Transactions on Intelligent Transportation Systems, 2016, 1- 9
21 ESTABROOKS A, JO T, JAPKOWICZ N A multiple resampling method for learning from imbalanced data sets[J]. Computational intelligence, 2004, 20 (1): 18- 36
doi: 10.1111/coin.2004.20.issue-1
22 李根 基于梯度提升决策树的高速公路交织区汇入模型[J]. 东南大学学报: 自然科学版, 2018, 48 (3): 563- 567
LI Gen Merging model in freeway weaving section based on gradient boosting decision tree[J]. Journal of Southeast University: Natural Science Edition, 2018, 48 (3): 563- 567
23 RESHEF D N, RESHEF Y A, FINUCANE H K, et al Detecting novel associations in large data sets[J]. Science, 2011, 334 (6062): 1518- 1524
doi: 10.1126/science.1205438
24 FRIEDMAN J H, MEULMAN J J Multiple additive regression trees with application in epidemiology[J]. Statistics in Medicine, 2003, 22 (9): 1365- 1381
doi: 10.1002/(ISSN)1097-0258
25 CHICCO, DAVIDE Ten quick tips for machine learning in computational biology[J]. BioData Mining, 2017, 10 (1): 35
doi: 10.1186/s13040-017-0155-3
26 BERGSTRA J, BENGIO Y Random Search for Hyper-Parameter Optimization[J]. Journal of Machine Learning Research, 2012, 13 (1): 281- 305
27 BERGSTRA J, BARDENET R, BENGIO Y, et al. Algorithms for hyper-parameter optimization [C] // International Conference on Neural Information Processing Systems. Granada: Curran Associates Inc. 2011.
[1] 刘晓伟,陈赟,张思,陈康. FDM型增材制造中送丝机构动态监测与识别[J]. 浙江大学学报(工学版), 2021, 55(3): 548-554.
[2] 陈巧红,陈翊,李文书,贾宇波. 多尺度SE-Xception服装图像分类[J]. 浙江大学学报(工学版), 2020, 54(9): 1727-1735.
[3] 李文书,邹涛涛,王洪雁,黄海. 基于双尺度长短期记忆网络的交通事故量预测模型[J]. 浙江大学学报(工学版), 2020, 54(8): 1613-1619.
[4] 胡天磊,王皓波,尹文栋. 基于深度双向分类器链的多标签新闻分类算法[J]. 浙江大学学报(工学版), 2019, 53(11): 2110-2117.
[5] 王硕朋,杨鹏,孙昊,刘迈. 两级参考点匹配位置指纹声源定位方法[J]. 浙江大学学报(工学版), 2019, 53(6): 1198-1204.
[6] 朱东阳, 沈静逸, 黄炜平, 梁军. 基于主动学习和加权支持向量机的工业故障识别[J]. 浙江大学学报(工学版), 2017, 51(4): 697-705.
[7] 丰小月, 梁艳春, 林希珣, 管仁初. 永恒语言学习研究与发展[J]. 浙江大学学报(工学版), 2017, 51(1): 82-88.
[8] 裘日辉, 刘康玲, 谭海龙, 梁军. 基于极限学习机的分类算法及在故障识别中的应用[J]. 浙江大学学报(工学版), 2016, 50(10): 1965-1972.
[9] 居斌, 钱沄涛, 叶敏超. 基于结构投影非负矩阵分解的协同过滤算法[J]. 浙江大学学报(工学版), 2015, 49(7): 1319-1325.
[10] 谭海龙, 刘康玲, 金鑫, 石向荣, 梁军. 基于μσ-DWC特征和树结构M-SVM的多维时间序列分类[J]. 浙江大学学报(工学版), 2015, 49(6): 1061-1069.
[11] 林亦宁, 韦巍, 戴渊明. 半监督Hough Forest跟踪算法[J]. J4, 2013, 47(6): 977-983.
[12] 李侃,黄文雄,黄忠华. 基于支持向量机的多传感器探测目标分类方法[J]. J4, 2013, 47(1): 15-22.
[13] 姚伏天, 钱沄涛, 李吉明. 空间约束半监督高斯过程下的高光谱图像分类[J]. J4, 2012, 46(7): 1295-1300.
[14] 戴兴虎,钱沄涛,唐凤仙,居斌. 基于图表标题信息的在线生物文献MRI图像检测[J]. J4, 2012, 46(7): 1307-1313.
[15] 王洪波, 赵光宙, 齐冬莲, 卢达. 一类支持向量机的快速增量学习方法[J]. J4, 2012, 46(7): 1327-1332.