王凤山, 戎全兵, 张宏军. 2016: 地下结构孕险环境危险性预警的改进SVM模型. 工程地质学报, 24(2): 324-330. DOI: 10.13544/j.cnki.jeg.2016.02.020
    引用本文: 王凤山, 戎全兵, 张宏军. 2016: 地下结构孕险环境危险性预警的改进SVM模型. 工程地质学报, 24(2): 324-330. DOI: 10.13544/j.cnki.jeg.2016.02.020
    WANG Fengshan, RONG Quanbing, ZHANG Hongjun. 2016: RISK PREGNANT ENVIRONMENT EARLY-WARNING MODEL FOR UNDERGROUND STRUCTURE BASED ON IMPROVED SUPPORT VECTOR MACHINE. JOURNAL OF ENGINEERING GEOLOGY, 24(2): 324-330. DOI: 10.13544/j.cnki.jeg.2016.02.020
    Citation: WANG Fengshan, RONG Quanbing, ZHANG Hongjun. 2016: RISK PREGNANT ENVIRONMENT EARLY-WARNING MODEL FOR UNDERGROUND STRUCTURE BASED ON IMPROVED SUPPORT VECTOR MACHINE. JOURNAL OF ENGINEERING GEOLOGY, 24(2): 324-330. DOI: 10.13544/j.cnki.jeg.2016.02.020

    地下结构孕险环境危险性预警的改进SVM模型

    RISK PREGNANT ENVIRONMENT EARLY-WARNING MODEL FOR UNDERGROUND STRUCTURE BASED ON IMPROVED SUPPORT VECTOR MACHINE

    • 摘要: 针对地震作用下孕险环境样本数据有限、复杂非线性等特点,提出了一种基于最小二乘支持向量机的地下结构孕险环境危险性预警方法。围绕地震后效应的地下工程区域孕险环境危险性预警目标和特征参数,设计孕险环境危险性问题的向量机表示形式,提出孕险环境危险性支持向量机训练工作机制,利用支持向量机结构风险最小化原则和非线性映射特性,建立基于最小二乘支持向量机的地下结构孕险环境危险性预警模型及其算法组件,并利用遗传算法优化其惩罚函数和核函数参数,隐式表达孕险环境危险性与其影响因素之间的非线性关系。结果表明,模型具有有效的小样本学习能力,具有较高的拟合和预测精度,明显优于神经网络等预测模型。

       

      Abstract: This paper addresses the limited complex and nonlinear characteristics in earthquake-induced risk pregnant environmental sample data. It puts forward a risk early-warning method for such pregnant environment around underground structure on least squares support vector machine. Around risk early-warning object and parameters in after-earthquake risk pregnant to underground engineering, the risk pregnant environment is expressed and designed with Support Vector Machine, and SVM training mechanism was proposed for risk pregnant environment. Such risk early-warning mold and the component is erected for risk pregnant environment around underground structure on least squares support vector machine, which utilizes structural risk minimization principle and nonlinear mapping feature of SVM, and optimizes the penalty function and kernel function parameters with Genetic Algorithm. The model implicitly expresses the non-linear relationship among the risk pregnant environment and factors. Case study shows such model has an effective small sample learning ability, well fitting and forecasting accuracy, which excels the predicting model with BP nerve network.

       

    /

    返回文章
    返回