岩溶水系统支持向量机泉流量预报模型研究.pdfVIP

岩溶水系统支持向量机泉流量预报模型研究 杨军耀 , 赵 　涛 (太原理工大学水利科学与工程学院 , 太原 　030024) 摘要 : 针对岩溶系统结构不甚清晰、基础资料不完备条件下泉水流量预报问题 , 引入了能较好地解 决小样本、非线性、高维数和局部极小点等问题的支持向量机 ( Support Vector Machines , SVM) 方 法 , 将泉流量影响因子时间序列与支持向量机方法有机结合 , 建立了岩溶水系统支持向量机泉流量 预报模型 , 并与BP 神经网络模型进行了实例比较。结果表明 , SVM 模型具有泛化能力强、预报精 度高的特点 , 可很好地克服神经网络的过学习问题 , 同时 , 针对 SVM 模型“峰值”预报精度差的 缺点 , 提出了“峰值”预报解决方案。 关键词 : 支持向量机 ; 神经网络 ; 时间序列 ; 泉流量预报模型 中图分类号: P6418 文献标识码 : A Abstract : The method of Support Vector Machines ( SVM) which is characterized by a small amount of samples , nonlinearity , multidimensions as well as the minimum in certain areas is introduced to solve the problems in predicting the flow of springs based on the indistinct karst topography and incomplete reference. The factor of time sequence which exerts the influence on the flow of springs is combined with the method of SVM to produce the SVM model for predicting the flow of springs in karst topography. The comparison of the SVM model with the cases of the back propagation neural network is made to indicate the sweeping generalization and high accuracy of the SVM model in predicting so as to solve the overfitting problems of neural network. Meanwhile , a formula is proposed to settle the problems in predicting the peak value which is aimed at the low accuracy of SVM in predicting the peak value. Key words : Support Vector Machines ; neural network ; time sequence ; predicting model for spring flow 0 　前言 1 　支持向量机的基本原理[ 7 ,8] 支持向量机是由 Vapnik 等于 1995 年在统计学 支持向量机是从线性可分情况下的最优分类面 理论的基础上首次提出的一种新的算法 , 它是建立 发展而来的