【作者】 刘丽；

【导师】 王艳辉；

【作者基本信息】 北京交通大学 ， 安全科学与工程， 2021， 硕士

【摘要】 安全是城市轨道交通运营的前提和核心竞争力。随着城市轨道交通高密度、强耦合性网络化程度加深,运营事故和安全问题引发社会各界高度重视,国家和行业相继出台了系列政策及管理办法规范城市轨道交通运营风险管理工作,系统风险管控成为热点话题。然而现如今的系统风险管控工作大多是在运营事故发生后凭借经验主义展开事故致因分析,这种“被动安全”模式已无法满足城轨运营现状的管控需求;科学预判潜事故风险并采取相应管控措施能够达到有效降低事故发生频率的目的,进而实现系统运营的“主动安全”的目的。鉴于此现实背景,本文以城轨运营事故为研究基础,提出一种规范化、信息化的事故风险知识提取方法,结合本体理论完成风险知识的结构化存储,最后采用贝叶斯理论完成对不确定性风险知识的推理研究。主要研究内容如下:(1)构建了表征城市轨道交通运营事故特征的系统专业词库深入城市轨道交通系统内部展开实地调研,对实际运营过程中的安全事故管理管控现状进行归纳总结;在此基础上分析了城轨系统运营事故统计特性、文本特征以及导致事故发生的原因;并考虑“人机环管”四个方面的致因因素完成了城市轨道交通运营事故专业词库的建立,同时根据获取的事故数据比例不同将运营事故专业词库划分为人员类子词库、物理结构类子词库、环境类子词库以及安全管理类子词库并设置不同的权重;子词库的划分可为下一步风险知识的有效提取做铺垫。(2)形成了城市轨道交通运营风险知识提取方法结合运营事故文本的非结构化存储现状,首先利用中文文本分词技术完成对事故文本的分词处理,并以隐形马尔科夫与Viterbi算法相结合的方式对分词效果进行优化;深入挖掘事故信息蕴含的风险知识,利用已构建的专业词库与TF-IDF算法综合考虑分词权重的方式完成事故文本风险概念词的有效抽取;为了进一步明确运营事故的发生机理,在提取了风险概念词的基础上,应用置信度和基于改进的K-means算法围绕顶层风险概念词展开概念间层次及非层次关系的抽取。(3)构建了城市轨道交通运营风险本体模型在完成城市轨道交通运营事故文本蕴含风险知识有效抽取的基础上,以半自动化的方式实现了运营事故风险本体的初步构建;针对风险知识的存储引入本体理论,明确了风险概念之间的层次、非层次关系以及风险概念的相关属性描述后,构建了表征风险知识的城市轨道交通运营事故风险本体并实现可视化展示;在其基础上对风险本体内部存在的逻辑关系进行检验评估并消除冗余错误连接关系,从而保证了风险本体结构向贝叶斯推理结构转换的可行性。(4)形成了基于改进贝叶斯结构的不确定风险知识推理方法已构建的城市轨道交通运营风险本体模型可以实现对确定性风险知识的有效存储,但是实际运营过程中仍存在不确定的风险知识。以风险本体结构向贝叶斯结构转换的方式完成贝叶斯网络拓扑结构的确定;在此基础上对风险本体进行概率扩展进一步明确贝叶斯推理结构;结合似然估计法和推理算法完成在不同风险知识状态有所变化的情况下相关运营事故发生的概率;同时针对系统主动安全提出了规范化、明确化的事故管理要求,以便于从事故中更准确抽取风险知识。(5)搭建了城市轨道交通系统运营风险管控平台基于上述研究成果,本文搭建了面向主动安全要求的城市轨道交通系统运营风险管控平台,并对系统的逻辑架构、主要功能及核心界面进行展示。更多还原

【Abstract】 Safety is the prerequisite and core competitiveness of the urban rail transit operation.With the deepening of high-density and strong coupling network of the urban rail transit,the operation safety problems have aroused great attention from all walks of life.The state and industry have issued a series of policies and management methods to standardize the operation risk management of the urban rail transit so as to the research about system risk management and control has become a hot topic problem.However,the work on system risk management and control are mostly based on empirical analysis of accident causes after the occurrence of which at present.This "passive safety" mode can’t meet the risk management and control needs of the urban rail transit operation status any more.Scientific prediction of the potential accident risk and taking corresponding measures can effectively reduce the frequency of accidents,so as to realize the "active safety" purpose of system operation safety.In view of this practical background,this paper proposes a method of the risk knowledge extraction based on the text research of urban rail operation accidents and combined with the ontology theory to complete the structured storage of the risk knowledge.Using Bayesian theory to complete the reasoning research on uncertain risk knowledge at last.The main research contents are as follows:(1)Constructed a systematic lexicon which is to represent the characteristics of urban rail transit operation accidentsBased on the deeply investigation inside the urban rail transit system,this paper summarizes the current situation about the safety accidents management and control in the actual operation process.And also analyzes the statistical characteristics,text characteristics and the causes of the urban rail transit system operation accidents.Based on these researches,we establish a systematic lexicon which considers the causative factors from the person,the machine,the environment and the management in four aspects.Meanwhile according to the different proportion of the accident data,the systematic lexicon is divided into personnel sub lexicon,physical component sub lexicon,environment sub lexicon and safety management sub lexicon with setting different weights.The division of sub lexicon can pave the way for the effective extraction of risk knowledge in the next step.(2)Formed the method of extracting operational risk knowledge of the urban rail transitCombined with the reality of unstructured storage about operation accident texts,the Chinese text segmentation technology is used to complete the word segmentation of the text firstly.And also using the hidden Markov and Viterbi algorithm to complete segmentation optimization.In order to further clarify the mechanism of operation accidents,the concept words of the risk are extracted with considering comprehensive weights by using the constructed professional lexicon and TF-IDF algorithm.To clarify the mechanism of the operation accidents furtherly,the confidence degree and improved K-means algorithm are used to develop the top-level risk concept words based on the risk concept words so as to complete the extraction of hierarchical and non-hierarchical relations among concepts.(3)Constructed an ontology model of urban rail transit operation riskAfter extracting the effective risk concepts and the connection relationship between the concepts from the urban rail transit operation accidents,the ontology concept model is introduced for the storage of risk knowledge,and a semi-structured urban rail operation risk ontology construction method is proposed.After clarifying the hierarchical and non-hierarchical relationship between the risk concepts and the description of the related attributes of the risk concepts,the construction of the urban rail transit operation risk ontology model is realized,and the editing software is used to complete the visual display of the risk ontology model.On this basis,the logical relationship existing in the risk knowledge is checked and evaluated,redundancy and wrong connection are eliminated,and the directed ring connection is avoided,which lays the foundation for the subsequent determination of the directed acyclic structure based on Bayesian inference.(4)Formed an uncertain risk knowledge reasoning method based on improved Bayesian structureThe constructed urban rail transit operation risk ontology model can realize the effective storage of deterministic risk knowledge,but there is still uncertain risk knowledge in the actual operation process.Combined with the assessed risk ontology structure,the probability extension is carried out to complete the determination of the Bayesian inference structure in this paper.The maximum likelihood estimation algorithm is also used to quantitatively analyze the relationship between the risk concept nodes,and infer the probability of related accidents in the case of uncertain risks.Under the results,we put forward the standardized accident text management ways to cater to the active safety requirements of urban rail transit operations,so as to more accurately extract risk knowledge from accidents.(5)An urban rail transit system operation risk management and control platform is developedBased on the above research,the paper develops an urban rail transit system operation risk management and control platform which is for active safety requirements.And demonstrate the logical architecture,main functions and core interfaces of the system.更多还原