数据驱动情景下嵌入组织因素的海事事故风险演化机理研究

Shipping is the main part of global trade transportation. However, the risk of maritime disaster remains high, which has forced to be an emergency for maritime transportation.With the the insight of data-driven and the method of data analysis, this project intends to study information extraction of organization factors, coupling law of disaster-causing factors and risk evolution mechanism of maritime accident, in order to provide a new perspective and a new method for the analysis of maritime accident research. More specifically, it includes as follows. .1) For the hidden feather of organization factors, the features of the fine-grained organization factors will be obtained from the maritime accident case text with using the ontology method and the vector space modal (VSM) method. 2）For the coupling feathers of high-dimensional variables, with the heuristic algorithm and the modified Expectation-maximization(EM) algorithm, the maritime risk probabilistic graphical model will be constructed from multi-source heterogeneous data; and then with the mutual information and the Gibbs sampling method,the key disaster-causing factors and their coupling relations will be analyzed, as well as the influencing mechanism of key disaster-causing factors on accidents. 3）Events scenarios and organization learning agent will be constructed, and then with the help of the dynamic Bayesian Network model, the dynamic risk evolution of accident risk under the control of organizational factors will be studied by numerical calculation. .This project elaborates the theoretical knowledge of computer science and management science, which expand the theoretical and methodological research of risk analysis, and support relevant government departments to make emergency decision.

海上运输是全球贸易运输的主要形式，但因组织因素导致的海事灾害风险一直居高不下，成为海上交通运输中亟待解决的难题。本项目拟基于数据驱动范式，利用数据分析处理的方法，研究海事事故中组织因素的信息提取、因素间耦合关系及灾害风险的演化机理，为事故风险分析提供新的视角和研究方法，具体包括：1）依托海事事故本体，针对组织因素的潜隐性，利用向量空间模型分析案例文本，获取组织因素的细粒度特征。2）聚焦高维变量的耦合性，利用启发式算法、改进的EM算法，构建多源异构数据下海事事故概率图模型；进而结合互信息、Gibbs采样法，研究关键的致灾因素及关键致灾因素间的耦合关系对事故的影响机理。3）构建关键事件情景和组织智能体，利用动态贝叶斯网络，结合数值分析方法模拟组织因素干预下海事事故风险的演化规律。项目集成了计算机科学和管理科学等多领域知识，拓展了风险分析的理论和方法研究，可为政府有关部门的应急决策提供学术支持。

船舶碰撞、溢油等事故往往会造成严重的经济损失、人身伤亡和环境污染，而研究表明人因和组织因是海事事故发生的主要原因。课题开展了以下研究：(1)利用自然语言学习，深入析取海事事故中组织因的空间向量模型；（2）融合文本、船舶动态轨迹及气象等多源异构数据，利用深度学习算法挖掘海事事故中高维致灾因子间耦合关系；（3）构建海事事故的多重情景，利用贝叶斯网络模型，建立不同情景下数据驱动的海事事故组织因风险评估机制。研究结果为制定有效的海事事故控制措施提供理论和技术支持。