首页 | 本学科首页   官方微博 | 高级检索  
     检索      

基于支持向量机-马尔可夫链的位移时序预测
引用本文:徐飞,徐卫亚.基于支持向量机-马尔可夫链的位移时序预测[J].岩土力学,2010,31(3):944-948.
作者姓名:徐飞  徐卫亚
作者单位:1. 河海大学 岩土力学与堤坝工程教育部重点实验室,南京 210098;2. 河海大学 岩土工程科学研究所,南京 210098
基金项目:国家自然科学基金重点资助项目,国家科技支撑计划,国家科技支撑计划,国家自然科学基金资助项目 
摘    要:结合支持向量机和马尔可夫链,提出了一种新的位移时序预测模型--支持向量机-马尔可夫链预测模型(SVM-MC)。通过对实测位移值的学习,利用经粒子群算法优化的支持向量机对位移时间序列的宏观发展趋势进行滚动预测;在此基础上应用马尔可夫链确定位移时序的状态转移概率矩阵,通过对状态的划分、实测值与支持向量机拟合值的绝对误差及相对误差等指标的分析,实现了对预测结果的改进。将该模型应用到某工程永久船闸高边坡的位移时序预测中,结果表明,该模型具有科学可靠、预测精度高的优点,在岩土体位移时序预测中具有有一定工程应用价值。

关 键 词:支持向量机  马尔可夫链  位移时间序列  粒子群优化  
收稿时间:2008-09-08

Prediction of displacement time series based on support vector machines-Markov chain
XU Fei,XU Wei-ya.Prediction of displacement time series based on support vector machines-Markov chain[J].Rock and Soil Mechanics,2010,31(3):944-948.
Authors:XU Fei  XU Wei-ya
Institution:1. Key Laboratory of Ministry of Education for Geomechanics and Embankment Engineering, Hohai University, Nanjing 210098, China; 2. Geotechnical Research Institute, Hohai University, Nanjing 210098, China
Abstract:A new displacement time series predicting model was proposed by combining the support vector machines and Markov Chain, named as support vector machines-Markov chain (SVM-MC) model. Through studying the measured displacements, SVM optimized by particle swarm was used to dynamically forecast the trend of macro development. Markov chain was applied to compute state transition probability matrix. By classifying system state and calculating absolute error and relative error between measured values and SVM fitting values, the predicting results are improved. The model was used to predict displacement time series of a high slope of a permanent shiplock. The engineering case studies indicate that the model is scientific and reliable; and there is engineering practical value for displacement time series predicting.
Keywords:support vector machines  Markov chain  displacement time series  particle swarm optimization
本文献已被 万方数据 等数据库收录!
点击此处可从《岩土力学》浏览原始摘要信息
点击此处可从《岩土力学》下载免费的PDF全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号