岩溶地下河日流量预测的小样本非线性时间序列模型  

Daily Discharge Forecast of Karst Underground River on Non-Linear Time Series Model of A Small Sample

作　　者：温忠辉[1] 任化准[1] 束龙仓[1] 王恩[1] 柯婷婷[1] 陈荣波 

机构地区：[1]河海大学水文水资源与水利工程科学国家重点实验室,南京210098

出　　处：《吉林大学学报（地球科学版）》2011年第2期455-458,464,共5页Journal of Jilin University：Earth Science Edition

基　　金：国家'973'计划项目(2006CB403204)

摘　　要：针对岩溶含水系统高度的非线性特征,在小样本时间序列条件下,引入了能较好解决小样本、非线性问题的支持向量回归方法,利用偏最小二乘回归对影响地下河流量的诸多因素进行综合分析,并提取主成分作为支持向量机的输入变量,采用遗传算法优化模型参数,建立了地下河日流量预测的偏最小二乘-遗传-支持向量回归模型;将该模型用于后寨典型岩溶地下河流域日流量模拟和预测,并与BP人工神经网络、多元线性回归模型预测结果进行对比。偏最小二乘-遗传-支持向量回归模型模拟期的均方误差(MSE)、平均绝对百分比误差(MAPE)分别为0.25%、6.89%,预测期为0.65%、6.03%;BP神经网络模拟期的MSE、MAPE分别为0.24%、7.30%,预测期为0.84%、7.39%;多元线性回归模型模拟期的MSE、MAPE分别为0.28%、9.30%,预测期为1.10%、10.54%。结果表明,偏最小二乘-遗传-支持向量回归模型预测精度明显优于BP人工神经网络和多元线性回归模型。Considering the high nonlinearity of karst aquifer system and under the conditions of time-series on a small sample,the authors introduce the support vector regression method,which can be used to solve the small sample size and non-linear problem,use the partial least-squares regression to analyze the numerous factors impacting the daily discharge of underground river and extract the principal component as the input variables of support vector machine and use genetic algorithms to optimize model parameters.PLS-Genetic-Support vector regression model is established for the daily flow forecast of underground river and is use to forecast the daily flow of the typical karst underground river area in Houzhai,the mean square error and average relative error of PLS-Genetic-Support vector regression model is 0.25% and 6.89% in simulation period,and it is 0.65% and 6.03% in forecast period,the mean square error and average relative error of artificial neural network model is 0.24% and 7.30% in simulation period,and it is 0.84% and 7.39% in forecast period,and the mean square error and average relative error of multiple regression model is 0.28% and 9.3% in simulation period,and it is 1.10% and 10.54% in forecast period.The results show that prediction accuracy of the model is significantly better than the BP neural network and multiple regression model.

关 键 词：地下河 小样本 偏最小二乘 遗传算法 支持向量回归 

分 类 号：P641.8[天文地球—地质矿产勘探]