中国安全科学学报 ›› 2019, Vol. 29 ›› Issue (10): 31-37.doi: 10.16265/j.cnki.issn1003-3033.2019.10.006

• 安全系统学 • 上一篇    下一篇

基于改进BN的集装箱船舶碰撞事故致因分析

司东森1, 张英俊1 教授, 郎坤**2 讲师   

  1. 1 大连海事大学 航海学院,辽宁 大连 116026;
    2 大连海事大学 航运经济与管理学院,辽宁 大连 116026
  • 收稿日期:2019-07-11 修回日期:2019-09-12 出版日期:2019-10-28 发布日期:2020-10-27
  • 通讯作者: ** 郎 坤(1988—),女,陕西汉中人,博士,讲师,硕士生导师,主要从事管理科学与工程方面的研究。E-mail:kun.lang@dlmu.edu.cn。
  • 作者简介:司东森 (1994—),男,河南商丘人,硕士研究生,研究方向为交通信息工程及控制。E-mail:18840821056@163.com。
  • 基金资助:
    国家重点研发计划(2018YFC0309600\03);中央高校基本科研业务费专项资金资助(3132019313)。

Causation analysis of container ship collision accidents based on improved BN

SI Dongsen1, ZHANG Yingjun1, LANG Kun2   

  1. 1 Navigation College, Dalian Maritime University, Dalian Liaoning 116026, China;
    2 Maritime Economics and Management College, Dalian Maritime University, Dalian Liaoning 116026, China
  • Received:2019-07-11 Revised:2019-09-12 Online:2019-10-28 Published:2020-10-27

摘要: 为解决利用贝叶斯网络(BN)模型分析集装箱船舶碰撞致因过程中,由于样本数据不足导致BN结构学习困难的问题,提出一种基于核密度估计和模型加权平均策略相结合的BN结构学习算法。首先利用核密度估计方法扩充小数据集,使数据规模达到BN结构学习的最低限度要求;然后利用模型平均策略加权融合不同学习算法得到的网络结构,来提高算法在小样本数据下的学习效果;最后基于少量集装箱船舶碰撞样本数据,利用所提算法构建BN模型分析事故致因。结果表明:用所提算法能在小样本数据下定量分析碰撞事故致因,并得出集装箱船舶碰撞事故致因链,有助于提高集装箱船舶运输的安全性。

关键词: 小样本数据, 贝叶斯网络(BN), 核密度估计, 模型平均法, 致因分析

Abstract: In order to address BN structure learning difficulties caused by insufficient data samples when BN is used to analyze causation of container ship collisions, a BN structure learning algorithm based on kernel density estimation and model weighted averaging was proposed. Firstly, kernel density estimation was used to expand small data set so as to meet minimum data size of BN structure learning. Then, model averaging strategy was utilized to integrate various learning algorithms by allocating different weights, which improved learning effect of the algorithm with small sample data. Finally, a BN model was established to analyze causation of container ship collision accidents based on a small number of samples. The results show that this proposed algorithm can quantitatively analyze causation of collision accidents on the basis of small sample data, and obtain accident causation chain. It is helpful to improve safety of container shipping.

Key words: small sample data, Bayesian network (BN), kernel density estimation, model averaging, causation analysis

中图分类号: