Expert elicitation and Bayesian Network modeling for shipping accidents: A literature review
Expert elicitation and Bayesian Network modeling for shipping accidents: A literature review
复制标题
DOI：
10.1016/j.ssci.2016.03.019
复制
发表时间：
2016-08
期刊：
影响因子：
6.1
通讯作者：
Guizhen Zhang;V. Thai
中科院分区：
文献类型：
--
作者：
Guizhen Zhang;V. Thai
贝叶斯网络（BBN）已成为一种流行的风险评估方法，特别是对罕见事故的建模。当历史数据不足以支持使用其他统计方法时，它可以利用专家的领域知识。在海事领域，贝叶斯网络通过对涉及大量人为因素和组织因素的船舶事故的因果关系进行建模，已被广泛用于风险预测。大多数模型依赖于专家的启发来构建模型和参数化。专家判断的介入带来了不确定性和偏见。相比之下，数据驱动的BBN被认为更客观，因为它是从经验数据中学习的。然而，尽管近年来研究人员开始探索数据驱动BBN的应用，但由于海上事故的罕见发生和事故数据库的不兼容，其应用仍然受到限制。因此，专家的知识仍然是建模的重要来源。减少启发工作量和促进个体条件概率的启发是利用专家知识进行BBN建模的两个最重要的任务。本文综述了促进专家启发过程的不同技术。其中一些方法已经应用于海事风险模型，但还需要开发和应用新的技术来解决船舶事故建模的不确定性，提高建模的准确性。
The Bayesian Network (BBN) has been a popular method for risk assessment especially for the modeling of rare accidents. It could make use of experts’ domain knowledge when historical data were not enough to support the use of other statistical methods. In the maritime domain, the Bayesian Network has been widely used for risk prediction by modeling the causal relationship of shipping accidents where a lot of human and organizational factors are involved. Most of the models depend on experts’ elicitation for model construction and parameterization. The involvement of experts’ judgment brings uncertainty and biases. In contrast, data-driven BBN is considered more objective since it is learnt from empirical data. However, even though researchers started to explore the application of data-driven BBN in recent years, its application is still constrained due to the rare occurrence of maritime accidents and the incompatibility of accident databases. As a result, experts’ knowledge continues to be an important source for modeling. Reducing the elicitation workload and facilitating the elicitation of individual conditional probability are the two most important tasks for BBN modeling with experts’ knowledge. Different techniques that facilitate experts’ elicitation process were reviewed in this paper. Some of these methods have been applied in the maritime risk model while new techniques should be developed and applied as well to address the uncertainty and improve accuracy of modeling shipping accidents.