岩土工程研究

基于贝叶斯网络的地震液化概率预测分析

展开
  • 1. 大连理工大学 海岸和近海工程国家重点实验室,辽宁 大连 116024;2. 大连理工大学 土木工程学院岩土工程研究所,辽宁 大连 116024; 3. 大连理工大学 信息管理与信息系统研究所,辽宁 大连 116024
胡记磊,男,1986年生,博士研究生,主要从事海洋土动力学和地震液化预测等方面研究

收稿日期: 2015-12-14

  网络出版日期: 2018-06-09

基金资助

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)。

Prediction of probability of seismic-induced liquefaction based on Bayesian network

Expand
  • 1. State Key Laboratory of Coastal and Offshore Engineering, Dalian University of Technology, Dalian, Liaoning 116024, China; 2. Institute of Geotechnical Engineering, School of Civil Engineering, Dalian University of Technology, Dalian, Liaoning 116024, China; 3. Institute of Information Management and Information systems, Dalian University of Technology, Dalian, Liaoning 116024, China

Received date: 2015-12-14

  Online published: 2018-06-09

Supported by

This work was supported by the National Program on Key Basic Research Project of China (2011CB013605-2) and the National Natural Science Foundation of China (51078062).

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F1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。

本文引用格式

胡记磊 ,唐小微,裘江南, . 基于贝叶斯网络的地震液化概率预测分析[J]. 岩土力学, 2016 , 37(6) : 1745 -1752 . DOI: 10.16285/j.rsm.2016.06.027

Abstract

Based on the interpretive structural model and cause-sequence mapping approach, twelve representative factors, either qualitative or quantitative, of seismic liquefaction are selected to construct a Bayesian network (BN) model of seismic-induced liquefaction under the condition of a large number of incomplete data. Based on a set of incomplete data of the 2011 Pacific Coast liquefaction induced by Tohoku Earthquake, the performances of proposed model are assessed comprehensively with regard to the following five indexes: the overall accuracy, the area under the ROC curve, precision, the recall rate and F1 score, and then compared with a radial basis function (RBF) neural network model. It is shown that both the back evaluation and forward prediction of the BN model are better than those of the RBF neural network model, and the BN model also performs well for the case of incomplete data. In addition, the BN model is also suitable for predicting the liquefaction of different soils. Classification imbalance and sampling bias can influence the performances of the models significantly. Hence it is suggested that the five indexes mentioned above can be used to evaluate the performances of evaluation models.
文章导航

/