基于贝叶斯网络的地震液化概率预测分析

发布时间：2018-10-26 16:30

【摘要】：基于解释结构模型和因果图法,选取12个具有代表性的定性和定量因素,在大量数据不完备的情况下提出了建立贝叶斯网络液化模型的方法。以2011年日本东北地区太平洋近海地震液化不完备数据为例,采用总体精度、ROC曲线下面积、准确率、召回率和F_1值5项指标对模型进行综合评估,并与径向基神经网络模型进行对比。结果表明:贝叶斯网络液化模型的回判和预测效果都优于径向基神经网络模型,且对于数据缺失的样本的预测效果也较理想。此外,该模型对于不同土质的液化评估均有较好的适用性。分类不均衡和抽样偏差会对模型的学习和预测效果产生很大影响,建议应同时采用上述5项评估指标进行综合评估模型的优劣。
[Abstract]:Based on the explanatory structure model and causality diagram, 12 representative qualitative and quantitative factors are selected, and the method of establishing the Bayesian network liquefaction model is proposed under the condition of incomplete data. Taking the incomplete seismic liquefaction data of the Pacific Ocean off the northeast Japan in 2011 as an example, the model is evaluated synthetically by 5 indexes, such as total precision, area under ROC curve, accuracy rate, recall rate and F1 value. And compared with the radial basis function neural network model. The results show that the prediction effect of Bayesian network liquefaction model is better than that of radial basis function neural network model. In addition, the model is applicable to the liquefaction assessment of different soils. The disequilibrium of classification and sampling deviation will have a great influence on the learning and prediction effect of the model. It is suggested that the above five evaluation indexes should be used to evaluate the advantages and disadvantages of the model at the same time.
【作者单位】：大连理工大学海岸和近海工程国家重点实验室;大连理工大学土木工程学院岩土工程研究所;大连理工大学信息管理与信息系统研究所;
【基金】：国家重点基础研究发展计划项目(973计划)(No.2011CB013605-2) 国家自然科学基金(No.51078062)~~
【分类号】：TU435

【相似文献】