基于贝叶斯网络的朝鲜海峡大气能见度推理模型及实验

Inference model and experiment for visibility in the Korean Strait based on Bayesian Network

针对当前海上贸易航道通航风险评估工作中存在的能见度数据缺失等问题,提出基于贝叶斯网络的能见度数据推理技术。通过研究海域的确定、节点因子的选取、样本数据集的生成、推理模型的构建及参数学习和推理计算等流程,构建了基于贝叶斯网络技术的能见度数据推理模型,并以朝鲜海峡海域为例展开试验分析。结果表明:能见度具有年变化和年际变化特征规律,利用多年某月的数据作为训练样本推理该月的能见度等级具有较高的准确性,且相同样本形式下样本数据数量与推理结果准确性呈正相关。

In this paper,considering the lack of visibility data in assessing the risk along marine trade routes,we present a new approach in inferring visibility data based on Bayesian Network.The inference model of visibility data based on Bayesian Network is built through the determination of study area,choice of nodal index,generation of sample datasets,configuration of inference model,parameter learning and inference calculation.The inference model is applied to Korean Strait as an experiment.It is found that visibility reveals annual and inter-annual features.It is of high accuracy to infer the visibility level of a specific month using multiple-year data of that month as training samples.Furthermore,we also demonstrate that the amount of sample data has a positive effect on the inference accuracy.