城市日用水量需求预测的等维新息SVR建模方法  

Equal-dimension and new information SVR forecasting model of urban daily water consumption

作　　者：柳景青[1] 俞亭超[1] 张土乔[1] 

机构地区：[1]浙江大学土木工程学系

出　　处：《浙江大学学报（工学版）》2006年第7期1230-1233,1271,共5页Journal of Zhejiang University：Engineering Science

基　　金：国家自然学基金资助项目(50078048)

摘　　要：针对社会经济因素和气象因素对城市日用水量需求的复杂非线性影响以及这种影响的动态变化特性,提出一种基于等维新息的支持向量回归(SVR)预测建模方法并用于城市日用水量需求预测.鉴于传统留一交叉验证法确定SVR模型参数时,存在样本分组数过少及容易出现预测误差评价失真等问题,采用测试集的等维新息SVR模型预测结果,提出了一种基于等维样本集的概率统计参数确定方法.实例分析表明:SVR模型的引入及新的参数确定方法的提出有利于提高城市日用水量需求预测精度.A support vector regression （SVR） forecasting model with equal-dimension and new information was developed. The model considered the following two aspects： one was the complex nonlinear relations between daily water consumption and its influence factors; the other was that the above relations changed with time. To go around the difficulty existing in traditional methods to decide the parameters for samples with noise or errors, a probability statistic method was proposed based on the forecasting results of the test set. Examples showed that the introduction of SVR forecasting model was helpful for improving the precision of forecasting urban daily water consumption; The SVR forecasting model with the parameters calculated by the probability statistic method had better results than that with other parameters.

关 键 词：日用水量 预测 SVR方法 模型参数 概率统计 

分 类 号：TU991.31[建筑科学—市政工程]