Comparison of Data-Driven Human-Like Driving Models Under Different Behavior Model Frameworks

doi:10.12141/j.issn.1000-565X.210783

Abstract

Abstract:

The traditional driving behavior model framework divides the driving behaviors into car-following and lane-changing, which are modeled separately. While the integrated driving behavior model framework believes that car-following and lane-changing are inseparable, so all driving behaviors are modeled as a whole. Based on these two behavioral model frameworks, this paper analyzed the performance of the data-driven human-like driving models. Firstly, it established integrated driving behavior model framework and car-following lane-changing combined model framework and then determined the input and output of the models according to the influencing factors in driving. Secondly, two combinations of car-following, lane-changing and intention recognition modules were proposed: discriminative combination and probability combination. Subsequently, the processing of the original data were carried out to build integrated driving behavior, car-following, lane-changing, and intention recognition datasets, which were used to train and calibrate the corresponding modules. Finally, the study compared the performance of the two combination models with the integrated driving behavior model in various aspects, including model accuracy, safety, robustness and migration. The results show that, when the model input and output, the parameter calibration process and the dataset are the same, the accuracy of the human-like driving model based on long short-term memory neural network (LSTM) is better than the model based on FNN. The mean square error of the model based on LSTM can reach 0.227 m², and the mean square error of the model based on FNN is 0.470 m². Within the LSTM-based model, the model using the car-following lane-changing combined model framework has better robustness and transferability than the model using the integrated driving behavior model framework. For the car-following lane-changing combined model, the mean square error of ±10% noise robustness can reach 1.383 m², and the mean square error of transferability can reach 0.462 m². For the integrated driving behavior model, the mean square error of ±10% noise robustness is 2.314 m², and the mean square error of transferability is 0.484 m².

Key words: data-driven, human-like driving, integrated driving, lane-changing, car-following, long short-term memory neural network

CLC Number:

U491

HUANG Ling, HUANG Zixu, WU Zerong, et al. Comparison of Data-Driven Human-Like Driving Models Under Different Behavior Model Frameworks[J]. Journal of South China University of Technology(Natural Science Edition), 2022, 50(10): 1-10.

Figures/Tables 16

Fig.1

Fig.2

Fig.3

Fig.4

Fig.5

Table 1

Input and output parameters for different models or modules"

模型或模块		输入参数	输出参数
LSTM-ID		$G I, C i, v i, C j, Δ s j i, ? j = C L, ? L L, ? L F, ? R L, ? R F$	$Δ s ? i t \| θ$
LSTM-C和LSTM-P	跟驰模块	$G I, C i, v i, C j, Δ s j i, ? j = C L$	$Δ s ? i t \| θ$
	换道模块	$G I, C i, v i, C j, Δ s j i, ? j = C L, ? L L, ? L F ? 或 ? C L, ? R L, ? R F$	$Δ s ? i t \| θ$
	意图识别模块	$G I, C i, v i, C j, Δ s j i, ? j = C L, ? L L, ? L F, ? R L, ? R F$	$Ω$

Table 1

Table 2

Decision parameters for LSTM human-like driving models"

决定参数	参数描述	取值范围
历史时间序列 $N$	模型输入中历史数据的序列长度， $N$ 越大说明模型使用的历史信息越多	2， 3， 4， …， 10
隐藏层数量 $H$	LSTM神经网络模型的隐藏层的数量	1， 2， 3， 4
神经元数量 $n H$	隐藏层 $H$ 中神经元的数量	20， 50， 100， 150， 200
随机失活（Dropout）率 $d r$	被失活的神经元的比例	0.1， 0.5， 0.9
训练次数（Epoch）	训练中的迭代次数	1， 2， 3， …， 1 000

Table 2

Table 3

Table 4

Table 5

Fig.6

Table 6

Statistical table of relative error of space headway with leading vehicle of each model"

$R E h s$	相对误差分布百分比/%
$R E h s$	LSTM-ID	LSTM-P	LSTM-C	FNN-ID（R）
［-0.01， 0.01］	47.21	46.02	44.44	31.58
［-0.02， 0.02］	72.32	69.83	68.37	55.49
［-0.03， 0.03］	84.63	82.39	81.07	70.81
［-0.04， 0.04］	90.88	89.13	88.02	80.65
［-0.05， 0.05］	94.34	92.93	92.00	86.95

Table 6

Fig.7

Table 7

Table 8

Table 9

References 24

1	黄玲，郭亨聪，张荣辉，等．人机混驾环境下基于LSTM的无人驾驶车辆换道行为模型［J］．中国公路学报，2020，33（7）：156-166.
	HUANG Ling， GUO Heng-cong， ZHANG Rong-hui，et al ．LSTM-based lane-changing behavior model for unmanned vehicle under environment of heterogeneous human-driven and autonomous vehicles ［J］．China Journal of Highway and Transport，2020，33（7）：156-166.
2	GIPPS P G ．A behavioural car-following model for computer simulation ［J］．Transportation Research Part B：Methodological，1981，15（2）：105-111.
3	WU J， BRACKSTONE M， MCDONALD M ．The validation of a microscopic simulation model：a methodological case study ［J］．Transportation Research Part C：Emerging Technologies，2003，11（6）：463-479.
4	HE Z， ZHENG L， GUAN W ．A simple nonparametric car-following model driven by field data ［J］．Transportation Research Part B：Methodological，2015，80：185-201.
5	HUANG X， SUN J， SUN J ．A car-following model considering asymmetric driving behavior based on long short-term memory neural networks ［J］．Transportation Research Part C：Emerging Technologies，2018，95：346-362.
6	HUANG L， GUO H， ZHANG R，et al ．Capturing drivers’lane changing behaviors on operational level by data driven methods ［J］．IEEE Access，2018，6：57497-57506.
7	JIA B， YANG D， ZHANG X，et al ．Car-following model considering the lane-changing prevention effect and its stability analysis ［J］．The European Physical Journal B，2020，93（8）：1-9.
8	TOLEDO T， KOUTSOPOULOS H N， BEN-AKIVA M. Integrated driving behavior modeling ［J］．Transportation Research Part C：Emerging Technologies，2007，15（2）：96-112.
9	HUANG L， GUO H， ZHANG R，et al ．A data-driven operational integrated driving behavioral model on highways ［J］．Neural Computing and Applications，2020，32（16）：13017-13033.
10	季学武，费聪，何祥坤，等．基于LSTM网络的驾驶意图识别及车辆轨迹预测［J］．中国公路学报，2019，32（6）：34-42.
	JI Xue-wu， FEI Cong， HE Xiang-kun，et al ．Intention recognition and trajectory prediction for vehicles using LSTM network ［J］．China Journal of Highway and Transport，2019，32（6）：34-42.
11	HOCHREITER S， SCHMIDHUBER J ．Long short-term memory ［J］．Neural Computation，1997，9（8）：1735-1780.
12	ADELI H ．Neural networks in civil engineering：1989-2000 ［J］．Computer‐Aided Civil and Infrastructure Engineering，2001，16（2）：126-142.
13	LECUN Y， BENGIO Y， HINTON G. Deep learning ［J］．Nature，2015，521（7553）：436-444.
14	van BRUMMELEN J， O’BRIEN M， GRUYER D，et al ．Autonomous vehicle perception：The technology of today and tomorrow ［J］．Transportation Research Part C：Emerging Technologies，2018，89：384-406.
15	FAN P， GUO J， ZHAO H，et al ．Car-following modeling incorporating driving memory based on autoencoder and long short-term memory neural networks ［J］．Sustainability，2019，11（23）：6755.
16	JIA S， HUI F， LI S，et al ．Long short-term memory and convolutional neural network for abnormal driving behaviour recognition ［J］．IET Intelligent Transport Systems，2020，14（5）：306-312.
17	TANG L， WANG H， ZHANG W，et al ．Driver lane change intention recognition of intelligent vehicle based on long short-term memory network ［J］．IEEE Access，2020，8：136898-136905.
18	CHEN X， SUN J， MA Z，et al ．Investigating the long-and short-term driving characteristics and incorporating them into car-following models ［J］．Transportation Research Part C：Emerging Technologies，2020，117：102698.
19	NAGAHAMA A， YANAGISAWA D， NISHINARI K ．Car-following characteristics of various vehicle types in respective driving phases ［J］．Transportmetrica B：Transport Dynamics，2020，8（1）：22-48.
20	WANG X， JIANG R， LI L，et al ．Capturing car-following behaviors by deep learning ［J］．IEEE Transactions on Intelligent Transportation Systems，2017，19（3）：910-920.
21	SUN L， JAFARIPOURNIMCHAHI A， KORNHAUSER A，et al ．A new higher-order viscous continuum traffic flow model considering driver memory in the era of autonomous and connected vehicles ［J］．Physica A：Statistical Mechanics and Its Applications，2020，547：123829.
22	Federal Highway Administration ．Next Generation Simulation （NGSIM）［EB/OL］．（2020-11-02）［2022-09-09］．.
23	Federal Highway Administration ．Revised monograph on traffic flow theory ［EB/OL］．（2017-01-31）［2022-09-09］ .
24	HUBER P J. Robust statistics ［M］．New York：John Wiley & Sons，2004.

模型	R_MSE/m²			r_MAE/m
模型	纵向	横向	总体	纵向	横向	总体
LSTM-ID	0.197	0.030	0.227	0.305	0.110	0.415
LSTM-P	0.225	0.032	0.257	0.323	0.113	0.436
LSTM-C	0.253	0.033	0.286	0.341	0.115	0.456
FNN-ID（R）^1）	0.447	0.023	0.470	0.466	0.098	0.564

意图	精确率/%	召回率/%	F1分数/%	准确率/%
换道	84.97	86.74	85.85	87.61
直行	89.71	88.28	88.99	87.61

干扰	模型	R_MSE/m²
干扰	模型	纵向	横向	总体
±10%噪声	LSTM-ID	2.282	0.032	2.314
	LSTM-P	1.350	0.033	1.383
	LSTM-C	1.388	0.033	1.421
	FNN-ID （R）	3.197	0.032	3.229
±20%噪声	LSTM-ID	5.655	0.033	5.689
	LSTM-P	3.461	0.033	3.494
	LSTM-C	3.489	0.033	3.523
	FNN-ID （R）	11.537	0.056	11.593
±30%噪声	LSTM-ID	9.253	0.034	9.287
	LSTM-P	5.868	0.033	5.901
	LSTM-C	5.948	0.034	5.982
	FNN-ID （R）	25.693	0.097	25.790

模型	R_MSE/m²			r_MAE/m
模型	纵向	横向	总体	纵向	横向	总体
LSTM-ID	0.451	0.033	0.484	0.430	0.107	0.537
LSTM-P	0.429	0.033	0.462	0.427	0.110	0.537
LSTM-C	0.543	0.035	0.578	0.494	0.119	0.613
FNN-I（R）	1.832	0.035	1.867	0.908	0.131	1.039

误差区间/m	测试误差分布/%
误差区间/m	LSTM-ID	LSTM-P	LSTM-C	FNN-ID（R）
［-0.3， 0.3］	52.75	52.36	46.04	27.03
［-0.5， 0.5］	75.76	75.37	70.30	44.63
［-1.0， 1.0］	89.47	89.17	85.60	63.80
［-1.5， 1.5］	97.28	96.33	94.23	79.89