基于高速公路事故数据的自适应事故持续时间预测模型研究

doi:10.3976/j.issn.1002-4026.20240094

摘要/Abstract

摘要：

高速公路交通事故严重影响道路的安全性与畅通性，精准预测事故持续时间是提升事故应急响应效率、减轻交通拥堵及降低二次事故风险的关键。提出了一种基于深度置信网络（DBN）与遗传算法（GA）的自适应参数优化模型，用于预测交通事故持续时间。收集了2020—2022年间山东省高速公路的交通事故数据，涵盖了包括道路属性、时间属性和环境属性在内的16个影响变量；通过引入斯皮尔曼相关性系数和箱型图，分析各影响变量与事故持续时间的相关性，确保所选变量的有效性与显著性。在此基础上，实验结合各类交通事故数据，构建了基于GADBN的自适应参数优化预测模型，该模型融合遗传算法的全局搜索与优化能力，显著提升了DBN模型的预测精度。为验证模型的有效性，实验选取SVR（支持向量回归）、RBF（径向基函数）、XGBoost（极限梯度提升树）和DBN等算法模型进行结果对比，并选取平均绝对百分比误差（δ_MAPE）和均方根误差（δ_RMSE）作为评价参数。实验结果表明，GADBN模型在δ_MAPE和δ_RMSE指标上的表现分别为16.49%和9.12，优于其他对比模型，验证了该模型的有效性与实用性。

关键词: 交通安全, 交通事故, 事故持续时间, 自适应参数优化, 高速公路

Abstract:

Freeway traffic accidents seriously affect road safety and accessibility. Accurately predicting the durations of accidents is key to improving emergency response efficiency，alleviating traffic congestion，and reducing the risk of secondary accidents. This paper proposes an adaptive parameter-optimization model based on a deep belief network （DBN） and genetic algorithm （GA） for predicting traffic accident durations. Traffic accident data from freeways in Shandong province were collected from 2020 to 2022，including 16 variables such as road，temporal attributes，and environmental attributes. The Spearman correlation coefficient and box plots were used to analyze the correlation between each variable and the accident duration，ensuring the validity and significance of the selected variables. Based on this analysis，we developed an adaptive parameter optimization-based prediction model，GADBN，using numerous traffic accident data. This model integrates the global search and optimization capabilities of the GA to notably improve the predictive accuracy of the DBN. To validate the model effectiveness，experimental comparisons were conducted with other algorithms such as support vector regression，radial basis functions，XGBoost，and DBNs，with mean absolute percentage error (δ_MAPE) and root-mean-square error (δ_RMSE) being used as evaluation metrics. The experimental results showed that the GADBN model achieved δ_MAPE and δ_RMSE values of 16.49% and 9.12，respectively，outperforming the other comparison models，thereby demonstrating its effectiveness and practicality.

Key words: traffic safety, traffic accidents, accident duration, adaptive parameter optimization, freeway

中图分类号:

U491.31

姜晓庆, 万青松, 郝文邦, 李立, 程卫平. 基于高速公路事故数据的自适应事故持续时间预测模型研究[J]. 山东科学, 2025, 38(5): 104-114.

JIANG Xiaoqing, WAN Qingsong, HAO Wenbang, LI Li, CHENG Weiping. Research on an adaptive accident duration-prediction model based on freeway accident data[J]. Shandong Science, 2025, 38(5): 104-114.

图/表 9

表1

图1

图2

图3

图4

图5

图6

表2

表3

参考文献 25

[1]	交通运输部. 我国高速公路总里程达17.7万公里[EB/OL].[2024-07-15]. https://baijiahao.baidu.com/s?id=1758612780478503239&wfr=spider&for=pc,2023-02-23.
[2]	HU L, BAO X Q, et al. A study on correlation of traffic accident tendency with driver characters using In-depth traffic accident data[J]. Journal of Advanced Transportation, 2020, 2020: 9084245. DOI: 10.1155/2020/9084245.
[3]	MOHAMMED A Z, ABDULLAHN M, AL-HUSSAINI H I. Review of incident duration prediction methods[J]. International Journal of Science and Research, 2020, 9(1):292-298.
[4]	康国祥, 方守恩. 基于风险分析的交通事件持续时间预测[J]. 同济大学学报(自然科学版), 2012, 40(2):241-245.
[5]	陈程, 张兰芳, 汪尚天. 高速公路危化品事件处置持续时间预测模型[J]. 交通信息与安全, 2017, 35(1): 55-61. DOI: 10.3963/j.issn.1674-4861.2017.01.007.
[6]	许宏科, 赵威, 杨孟, 等. 基于改进BPNN的高速公路交通事故持续时间预测[J]. 华东交通大学学报, 2020, 37(5): 60-65.
[7]	魏丹. 基于机器学习的交通状态判别与预测方法[D]. 长春: 吉林大学, 2020.
[8]	吕路, 李杰, 郭忠印, 等. 高速公路交通事故持续时间计算方法研究[J]. 公路交通科技, 2022, 39(12): 155-162. DOI: 10.3969/j.issn.1002-0268.2022.12.019.
[9]	SHENG X, et al. A deep fusion model based on restricted Boltzmann machines for traffic accident duration prediction[J]. Engineering Applications of Artificial Intelligence, 2020, 93: 103686. DOI: 10.1016/j.engappai.2020.103686.
[10]	柴涛. 基于机器学习的交通事故持续时间预测算法研究[D]. 哈尔滨: 哈尔滨工业大学, 2018.
[11]	陈娇娜, 靳引利, 陶伟俊, 等. 基于文本信息的高速公路事故持续时间中介效应研究[J]. 中国安全科学学报, 2023, 33(4):155-162.DOI: 10.16265/j.cnki.issn1003-3033.2023.04.0826.
[12]	BAI S M, DAI B Y, et al. An integrated model for the geohazard accident duration on a regional mountain road network using text data[J]. Sustainability, 2022, 14(19): 12429. DOI: 10.3390/su141912429.
[13]	何珂, 杨顺新, 郜勇刚. 基于PCA-RF组合模型的隧道交通事故持续时间预测[J]. 交通信息与安全, 2019, 37(5): 26-32.
[14]	KUANG L, YAN H, ZHU Y J, et al. Predicting duration of traffic accidents based on cost-sensitive Bayesian network and weighted K-nearest neighbor[J]. Journal of Intelligent Transportation Systems, 2019, 23(2): 161-174. DOI: 10.1080/15472450.2018.1536978.
[15]	CHEN J N, TAO W J. Traffic accident duration prediction using text mining and ensemble learning on expressways[J]. Scientific Reports, 2022, 12(1): 21478. DOI: 10.1038/s41598-022-25988-4. pmid: 36509866
[16]	ZHANG Y, WANG X H. Intrusion detection for IoT based on improved genetic algorithm and deep belief network[J]. IEEE Access, 2019, 7: 31711-31722. DOI: 10.1109/ACCESS.2019.2903723.
[17]	HINTON G E, OSINDERO S, TEH Y W. A fast learning algorithm for deep belief nets[J]. Neural Computation, 2006, 18(7): 1527-1554. DOI: 10.1162/neco.2006.18.7.1527. pmid: 16764513
[18]	TAO C X, WANG X, GAO F Y, et al. Fault diagnosis of photovoltaic array based on deep belief network optimized by genetic algorithm[J]. Chinese Journal of Electrical Engineering, 2020, 6(3): 106-114. DOI: 10.23919/CJEE.2020.000024.
[19]	UPADHYA V, SASTRY P S. Learning gaussian-bernoulli RBMs using difference of convex functions optimization[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 33(10): 5728-5738. DOI: 10.1109/TNNLS.2021.3071358.
[20]	CHEN S Y, ZHENG C J. Traffic incident duration prediction based on support vector regression[C]// ICCTP 2011. Nanjing: American Society of Civil Engineers, 2011: 2412-2421. DOI: 10.1061/41186(421)241.
[21]	易富君, 韩直, 邓卫. 遗传算法优化的RBF神经网络模型在公路隧道群交通事故微观预测中的应用[J]. 交通运输工程与信息学报, 2012, 10(1): 64-72. DOI: 10.3969/j.issn.1672-4747.2012.01.011.
[22]	TANG J J, ZHENG L L, HAN C Y, et al. Traffic incident clearance time prediction and influencing factor analysis using extreme gradient boosting model[J]. Journal of Advanced Transportation, 2020, 2020: 6401082. DOI: 10.1155/2020/6401082.
[23]	WANG P, HAO W B, JIN Y L. Fine-grainedtraffic flow prediction of various vehicle types via fusion of multisource data and deep learning approaches[J]. IEEE Transactions on Intelligent Transportation Systems, 2021, 22(11): 6921-6930. DOI: 10.1109/TITS.2020.2997412.
[24]	马德明, 梁宏斌. 基于时空混合残差网络的事故黑点预测研究[J]. 交通运输工程与信息学报, 2020, 18(4):68-75.
[25]	CHO M, PARK J, KIM S, et al. Estimation of driving direction of traffic accident vehicles for improving traffic safety[J]. Applied Sciences, 2023, 13(13): 7710. DOI: 10.3390/app13137710.

变量	变量值
事故类型	0=侧翻，1=自燃，2=追尾，3=撞护栏，4=货物洒落
天气	0=晴，1=（雨、雾、雪）
交通流状态	0=畅通，1=拥堵
是否涉及多车	0=否，1=是
交叉口/弯道	0=否，1=是
是否有抛洒物	0=否，1=是
是否工作日	0=否，1=是
是否需要倒货	0=否，1=是
是否需要拖车	0=否，1=是
发生时间	0=白天，1=夜间（20：00—06：00）
是否养护施工	0=否，1=是
小车（小客车）	0~10内取整数值
大车（大客车、货车、罐装车）	0~6内取整数值
人员伤亡数	0~5内取整数值
车道数	2、4、6、8
拥堵长度	拥堵长度取值范围为0~10 km

隐含节点	预训练迭代次数	微调迭代次数	δ_MAPE/%	δ_RMSE	耗时/s
[128，64]	100	200	19.53	11.8	585
[128，64，16]	150	300	16.49	9.12	632
[256，128，64，16]	100	200	18.19	10.39	768
[256，128，64，64，16]	200	300	20.37	12.73	920

评价指标	验证组别	预测时长/min	模型预测误差
评价指标	验证组别	预测时长/min	SVR	XGBoost	RBF	DBN	GADBN
δ_MAPE/%	组1	60	25.64	19.22	16.77	15.18	13.16
δ_MAPE/%	组2		27.33	18.12	15.88	14.78	13.36
δ_RMSE	组1		18.24	12.44	9.38	8.25	7.07
δ_RMSE	组2		20.91	13.25	10.16	7.65	6.88
δ_MAPE/%	组1	120	32.54	28.32	22.45	20.78	16.49
δ_MAPE/%	组2		35.55	29.24	23.66	21.39	17.65
${{\delta }_{R}}_{MSE}$	组1		25.87	20.38	14.33	12.39	9.12
${{\delta }_{R}}_{MSE}$	组2		28.31	19.85	14.98	11.47	10.03