基于 <b>GBDT</b> 的铁路事故类型预测及成因分析

钟敏慧; 张婉露; 李有儒; 朱振峰; 赵耀

doi:10.16383/j.aas.c190630

基于 GBDT 的铁路事故类型预测及成因分析

doi: 10.16383/j.aas.c190630

钟敏慧^{1, 2,},
张婉露^{1, 2,},
李有儒^{1, 2,},
朱振峰^{1, 2,},
赵耀^{1, 2,}

1.
北京交通大学信息科学研究所北京 100044
2.
北京市现代信息科学与网络技术重点实验室北京 100044

基金项目: 科技创新 2030- “新一代人工智能”重大项目(2018AAA0102101), 中央高校基本科研业务费(2018JBZ001), 国家自然科学基金(61976018, 61532005)资助

详细信息

作者简介:
钟敏慧：北京交通大学信息科学研究所硕士研究生. 主要研究方向为计算机视觉, 机器学习. E-mail: mhzhong@bjtu.edu.cn

张婉露：北京交通大学信息科学研究所硕士研究生. 主要研究方向为计算机视觉, 深度学习. E-mail: wlzhang@bjtu.edu.cn

李有儒：北京交通大学信息科学研究所硕士研究生. 主要研究方向为数据挖掘, 机器学习. E-mail: liyouru@bjtu.edu.cn

朱振峰：北京交通大学信息科学研究所教授. 2005年获中国科学院自动化研究所模式识别国家重点实验室工学博士学位. 主要研究方向为图像视频分析与理解, 计算机视觉, 机器学习. 本文通信作者. E-mail: zhfzhu@bjtu.edu.cn

赵耀：北京交通大学信息科学研究所教授, 所长. 1996年获北京交通大学工学博士学位. 主要研究方向为图像与视频编码, 数字水印与取证, 视频分析及理解, 人工智能. E-mail: yzhao@bjtu.edu.cn

计量
- 文章访问数: 1951
- HTML全文浏览量: 811
- PDF下载量: 242
- 被引次数: 0
出版历程
- 收稿日期: 2019-09-11
- 录用日期: 2020-01-17
- 网络出版日期: 2021-09-26
- 刊出日期: 2022-02-18

GBDT Based Railway Accident Type Prediction and Cause Analysis

ZHONG Min-Hui^{1, 2
,},
ZHANG Wan-Lu^{1, 2
,},
LI You-Ru^{1, 2
,},
ZHU Zhen-Feng^{1, 2
,},
ZHAO Yao^{1, 2
,}

1.
Institute of Information Science, Beijing Jiaotong University, Beijing 100044
2.
Beijing Key Laboratory of Advanced Information Science and Network Technology, Beijing 100044

Funds: Supported by Science and Technology Innovation 2030 Major Program: New Generation Artificial Intelligence (2018AAA0102101), the Fundamental Research Funds for the Central Universities (2018JBZ001), National Natural Science Foundation of China (61976018, 61532005)

More Information

Author Bio:
ZHONG Min-Hui Master student at the Institute of Information Science, Beijing Jiaotong University. Her research interest covers computer vision and machine learning

ZHANG Wan-Lu　Master student at the Institute of Information Science, Beijing Jiaotong University. Her research interest covers computer vision and deep learning

LI You-Ru　Master student at the Institute of Information Science, Beijing Jiaotong University. His research interest covers data mining and machine learning

ZHU Zhen-Feng　Professor at the Institute of Information Science, Beijing Jiaotong University. He received his Ph. D. degree in pattern recognition and intelligence system from the Institute of Automation, Chinese Academy of Sciences in 2005. His research interest covers image and video understanding, computer vision and machine learning. Corresponding author of this paper

ZHAO Yao　Professor and director at the Institute of Information Science, Beijing Jiaotong University. He received his Ph. D. degree from Beijing Jiaotong University in 1996. His research interest covers image/video coding, digital watermarking and forensics, video analysis and understanding and artificial intelligence

摘要

摘要: 运用数据挖掘技术进行铁路事故类型预测及成因分析, 对于建立铁路事故预警机制具有重要意义. 为此, 本文提出一种基于梯度提升决策树(Grandient boosting decision tree, GBDT)的铁路事故类型预测及成因分析算法. 针对铁路事故记录数据缺失的问题, 提出一种基于属性分布概率的补全算法, 最大程度保持原有数据分布, 从而降低数据缺失对事故类型预测造成的影响. 针对铁路事故记录数据类别失衡的问题, 提出一种集成的GBDT模型, 完成对事故类型的鲁棒性预测. 在此基础上, 根据GBDT预测模型中特征重要度排序, 实现事故成因分析. 通过在开放数据库上进行实验, 验证了本文模型的有效性.
- 事故类型预测 /
- 缺失补全 /
- GBDT /
- 集成学习 /
- 成因分析
Abstract: The application of data mining technology in railway accident type prediction and cause analysis is of great significance to establish railway accident early warning mechanism. This paper proposes a gradient boosting decision tree (GBDT) based algorithm for railway accident type prediction and cause analysis. In order to solve the problem of data missing in railway accident record dataset, we propose a new data complement algorithm based on the attribute distribution probability, which can keep the distribution of original data as much as possible, thus reducing the impact of data missing on predicting railway accident type. To reduce the impact of unbalanced categories of data in railway accident dataset, an ensemble GBDT model is proposed to predict the types of accidents effectively and robustly. On these bases, according to the importance of features in GBDT prediction model, we complete the cause analysis of railway accidents. Experimental results on an open database show that our proposed method can predict the types and causes of railway accidents effectively.
- Prediction of railway accident type /
- missing data completion /
- GBDT /
- ensemble learning /
- cause analysis

HTML全文

图 1 基于GBDT的铁路事故类型预测及成因分析框架

Fig. 1 The framework of GBDT-based railroad accident type prediction and cause analysis

下载: 全尺寸图片幻灯片

图 2 三种补全方法结果对比

Fig. 2 Comparison of three methods results

下载: 全尺寸图片幻灯片

图 3 不同GBDT集成个数下分类准确率

Fig. 3 Accuracy of classifiers with different number of GBDT

下载: 全尺寸图片幻灯片

图 4 混淆矩阵

Fig. 4 Confusion matrix

下载: 全尺寸图片幻灯片

图 5 不同特征数量下预测结果

Fig. 5 Prediction results of classifier with different features

下载: 全尺寸图片幻灯片

图 6 两类事故致因中不同因素的比例

Fig. 6 Proportion of different factors in causes of two types of railroad accidents

下载: 全尺寸图片幻灯片

表 1 原始数据描述

Table 1 Description of original data

	Record	Accident type	Attribute
Number	5 434	11	144

下载: 导出CSV

表 2 事故类型描述

Table 2 Description of accident types

Type	Description
1	Derailment
2	Head on collision
3	Rearend collision
4	Side collision
5	Raking collision
6	Broken train collision
7	Hwy-rail crossing
8	RR grade crossing
9	Obstruction
10	Fire
11	Other impacts

下载: 导出CSV

表 3 数据集部分示例

Table 3 Examples of the dataset

Name	Description	Number	Type
RAILROAD	Railroad code	5 434	Object
CARS	Num. of cars carrying hazmat	5 434	Int64
TYPSPD	Train speed type	5 086	Object
TRNDIR	Train direction	5 161	Float64
TONS	Gross tonnage, excluding power units	5 434	Int64
TYPEQ	Type of consist	5 081	Object
EQATT	Equipment attended	5 074	Object
CDTRHR	Num. of hours conductors on duty	3 628	Int64
ENGHR	Num. of hours engineers on duty	4 201	Int64
TRKNAME	Track identification	5 434	Object

下载: 导出CSV

表 4 预处理后数据描述

Table 4 Description of preprocessed data

	Record	Accident type	Attribute dimension
Number	5 434	11	119

下载: 导出CSV

表 5 三种方法补全前后特征TRNDIR取值分布

Table 5 Distribution of the attribute TRNDIR values before and after three completion methods

Algorithm	$a_j=1$	$a_j=2$	$a_j=3$	$a_j=4$
Before completion	0.22	0.20	0.31	0.27
Interpolation	0.21	0.19	0.30	0.30
Mode	0.21	0.19	0.34	0.26
Our algorithm	0.22	0.20	0.31	0.27

下载: 导出CSV

表 6 不同采样率下集成GBDT分类准确率

Table 6 Accuracy of classifiers with different sampling rates

$\alpha$	0.6	0.7	0.8	0.9	1.0
Accuracy	0.841	0.846	0.845	0.852	0.848

下载: 导出CSV

表 7 各分类器性能对比

Table 7 Performance comparison of classifiers

Classifier	Accuracy	Precision	Recall	F1
DT	0.728	0.73	0.73	0.73
RF	0.773	0.74	0.77	0.75
ET	0.734	0.70	0.73	0.71
GBDT	0.841	0.84	0.84	0.84
Ensemble GBDT	0.852	0.85	0.85	0.85

下载: 导出CSV

表 8 重要度排名前15的特征

Table 8 Features of top 15 in importance

No.	Name	Description
1	Latitude	Latitude in decimal degrees
2	Longitude	Longitude in decimal degrees
3	CNTYCD	FIPS county code
4	HIGHSPD	Maximum speed
5	TRKNAME	Track identification
6	RRCAR1	Car initials (fist involved)
7	TEMP	Temperature in degrees fahrenheit
8	MILEPOST	Milepost
9	STATION	Nearest city and town
10	TRNSPD	Speed of train in miles per hour
11	RRCAR2	Car initials (causing)
12	SUBDIV	Railroad subdivision
13	ENGHR	Num. of hours engineers on duty
14	CDTRHR	Num. of hours conductors on duty
15	TONS	Gross tonnage

下载: 导出CSV

参考文献(22)

[1]	Ming L. Data mining: concepts, models, methods, and algorithms. IIE Transaction, 2004, 36(5): 495-496
[2]	冯士雍. 回归分析方法. 北京: 科学出版社, 1974 Feng Shi-Yong. Regression Analysis Method. Beijing: Science Press, 1974
[3]	Rutkowski L, Jaworski M, Pietruczuk L, Duda P. Decision trees for mining data streams based on the gaussian approximation. IEEE Transactions on Knowledge and Data Engineering, 2014, 26(1): 108−119 doi: 10.1109/TKDE.2013.34
[4]	李定启, 程远平, 王海峰, 王亮, 周红星, 孙建华. 基于决策树ID3改进算法的煤与瓦斯突出预测. 煤炭学报, 2011, 36(4): 619−622 Li Ding-Qi, Cheng Yuan-Ping, Wang Hai-Feng, Wang Liang, Zhou Hong-Xing, Sun Jian-Hua. . Coal and gas outburst prediction based on improved decision tree ID3 algorithm. Journal of China Coal Society, 2011, 36(4): 619−622
[5]	Breiman L. Random forest. Machine Learning, 2001, 45(1): 5−32 doi: 10.1023/A:1010933404324
[6]	Friedman J H. Greedy function approximation: a gradient boosting machine. The Annals of Statistics, 2001, 29(5): 1189−1232
[7]	Friedman J H. Stochastic gradient boosting. Computational Statistics and Data Analysis, 2002, 38(4): 367−378 doi: 10.1016/S0167-9473(01)00065-2
[8]	周志华. 机器学习. 北京: 清华大学出版社, 2016. Zhou Zhi-Hua. Machine Learning. Beijing: Tsinghua University Press, 2016.
[9]	Schonlau M. Boosted regression (boosting): an introductory tutorial and a stata plugin. The Stata Journal, 2005, 5(3): 330−354. doi: 10.1177/1536867X0500500304
[10]	翁小雄, 吕攀龙. 基于 GBDT 算法的地铁 IC 卡通勤人群识别. 重庆交通大学学报 (自然科学版), 2019, 38(5): 8−12 Weng Xiao-Xiong, Lv Pan-Long. Subway IC card commuter crowd identification based on GBDT algorithm. Journal of Chongqing Jiaotong University(Natural Science), 2019, 38(5): 8−12
[11]	Mursalin M, Zhang Yuan, Chen Yue-Hui, Chawla N V. Automated epileptic seizure detection using improved correlation-based feature selection with random forest classifier. Neurocomputing, 2017, 241: 204−214 doi: 10.1016/j.neucom.2017.02.053
[12]	Cheng J, Li G, Chen X H. Research on travel time prediction model of freeway based on gradient boosting decision tree. IEEE Access, 2018, 7: 7466−7480
[13]	Ma X, Ding C, Luan S, Wang Y, Wang Y P. Prioritizing influential factors for freeway incident clearance time prediction using the gradient boosting decision trees method. IEEE Transactions on Intelligent Transportation Systems, 2017, 18(9): 2303−2310 doi: 10.1109/TITS.2016.2635719
[14]	Su H W, Zhang W J, Li Z H. Analysis and prediction of water traffic accidents in jingtang port based on improved GM(1, 1) model. In: Proceedings of the 37th Chinese Control Conference (CCC). New York, USA: IEEE, 2018.2212−2217
[15]	Das S, Sun X D. Investigating the pattern of traffic crashes under rainy weather by association rules in data mining. In: Proceedings of the 93rd Transportation Research Board (TRB) Annual Meeting. Washington D.C., USA: Nation Academy of Sciences, 2014
[16]	金勇进. 缺失数据的统计处理, 北京: 中国统计出版社, 2009. Jin Yong-Jin. Statistical Processing of Missing Data. Beijing: China Statistics Press, 2009.
[17]	金勇进. 调查中的数据缺失及处理 (I)-缺失数据及其影响. 数理统计与管理, 2001, 20(4): 58−60 doi: 10.3969/j.issn.1002-1566.2001.04.012 Jin Yong-Jin. Data loss and processing in survey(I)) data missing and impact. Journal of Applied Statistics and Management, 2001, 20(4): 58−60 doi: 10.3969/j.issn.1002-1566.2001.04.012
[18]	Collell G, Prelec D, Patil K R. A simple plug-in bagging ensemble based on threshold-moving for classifying binary and multiclass imbalanced data. Neurocomputing, 2018, 275: 330−340 doi: 10.1016/j.neucom.2017.08.035
[19]	Galar M, Fernandez A, Barrenechea E, Bustince H, Herrera F. A review on ensembles for the class imbalance problem: bagging-, boosting-, and hybrid-based approaches. IEEE Transactions on Systems, Man and Cybernetics, Part C (Applications and Reviews), 2012, 42(4): 463−484 doi: 10.1109/TSMCC.2011.2161285
[20]	朱振峰, 汤静远, 常冬霞, 赵耀. 基于 GBDT 的商品分配层次化预测模型. 北京交通大学学报, 2018, 42(2): 9−13+45 doi: 10.11860/j.issn.1673-0291.2018.02.002 Zhu Zhen-Feng, Tang Jing-Yuan, Chang Dong-Xia, Zhao Yao. GBDT based hierarchical model for commodity distribution prediction. Journal of Beijing Jiaotong University, 2018, 42(2): 9−13+45. doi: 10.11860/j.issn.1673-0291.2018.02.002
[21]	杨连报, 李平, 薛蕊, 马小宁, 吴艳华, 邹丹. 基于不平衡文本数据挖掘的铁路信号设备故障智能分类. 铁道学报, 2018, 40(2): 59−66 doi: 10.3969/j.issn.1001-8360.2018.02.009 Yang Lian-Bao, Li Ping, Xue Rui, Ma Xiao-Ning, Wu YanHua, Zou Dan. Intelligent classification of faults of railway signal equipment based on imbalancd text data mining. Journal of the China Railway Society, 2018, 40(2): 59−66 doi: 10.3969/j.issn.1001-8360.2018.02.009
[22]	Federal Railroad Administration Office of Safety Analysis [Online], available: https://safetydata.fra.dot.gov/OfficeofSafety/Default.aspx, June 1, 2019