基于贝叶斯网络的并发型突发事件链建模方法

现实情形中，由于致灾因子和作用对象的相似性，初始突发事件的发生易引发多个次生事件并发及耦合，致使事件的演化发展及可能造成的损失具有更大的不确定性。然而，现有的突发事件链式演化分析多运用串发型事件链，对于并发型突发事件存在适用性较低的问题。针对上述问题，本文将突发事件抽象描述为以输入、状态和输出属性为组成要素，通过属性要素间的作用关系构成的复杂系统，进而从属性层面分析事件间的关联关系；以贝叶斯网络为建模工具，识别并发型突发事件间具有的因果关系和耦合关系，给出事件贝叶斯网络关联方法，构建并发型突发事件链模型；基于历史数据获取网络节点间的先验概率信息，运用贝叶斯网络推理算法实现并发型突发事件的演化分析；最后，通过实例验证本文方法在实际应用中的科学性及可行性，并通过对比分析阐明其在提高灾害损失预测精度方面具有一定的优势。

Modeling Method of Concurrent Emergency Chain Based on Bayesian Network

In realistic circumstances, because of the similarities between the hazard factors and their affected objects in different emergencies, the occurrence of initial emergencies is likely to trigger the concurrence and coupling of multiple secondary emergencies, which makes the evolution of emergencies more uncertain and causes more serious losses. However, the existing emergency chain evolution analyses mostly used serial emergency chains, which is less applicable to concurrent emergencies. In view of the above problems, this paper presents a modeling method of concurrent emergency chain based on Bayesian Network to model the parallel evolution of concurrent emergencies. Firstly, emergency is described as a complex system composed of input, state, output attributes and the mutual influence relationships between them. And the causality and coupling relationships between emergencies are analyzed and defined on attribute level. Secondly, Bayesian network is applied to represent a single emergency formally. Based on the defined causality and coupling relationships between emergencies, the identification method of the causality and coupling relationships between single emergency Bayesian networks, the association method of concurrent emergency Bayesian networks, and the concrete construction method of the Bayesian network on concurrent emergency chain are put forward. Thirdly, the reasoning algorithm and its complexity and feasibility of the constructed concurrent emergencies Bayesian network are discussed. Through the Bayesian network reasoning process, the evolution analysis of concurrency emergencies can be realized in case of the prior probabilities between network nodes are obtained based on historical data analysis. Finally, to demonstrate the feasibility and validity of the proposed methods, the historical data of rainstorm disasters and the subsequent secondary disasters, such as mudslides, landslides, and floods and so on, in Sichuan Province from 2008 to 2015 is collected. Then the K2 network learning method is used to study the historical data to generate the Bayesian network of each disaster. Based on the identification of the interrelationships between the single Bayesian networks corresponding to each disaster, the integrated Bayesian network is constructed by correlating the interrelated single Bayesian networks. Taking the rainstorm disaster happened in August 16, 2015 in Yibin City as an example to instantiate the initial evidence information of the constructed integrated Bayesian network, the joint tree reasoning method is applied to predict the possible losses caused by the rainstorm and its derivative disasters. The analyses of the prediction result and the comparison with related serial emergency chains verify the scientificalness and effectiveness of the proposed methods while being used in evolution analysis of concurrent emergencies.