摘要
鉴于日用水量的时变性,本文提出一种基于变结构支持向量回归的动态预测模型.利用日用水量的历史数据训练支持向量机,得到模型结构参数历史数据序列,然后利用扩展卡尔曼滤波器对模型结构参数组进行估计,最后用模型结构参数估计量来更新模型结构并预测下一天日用水量.在实例分析中分别利用变结构支持向量回归模型和支持向量机预测模型对实际用水量性进行预测分析.结果表明,前者具有更好的动态跟踪能力和更高的预测精度,可应用于城市日用水量的预测.
In view of the time-varying characteristic of daily water consumption,a dynamic forecasting model based on variable structure support vector regression is proposed. Train support vector machine( SVM) using historical data of daily water consumption,and get historical data series of forecasting model structure parameters. According to the historical series of model structure parameters,estimate the next-day model structure parameters using extended kalman filter. Then the estimated structure parameters are used to update the SVM structure. Finally forecast the next-day water consumption with new structure SVM model. Application examples show that compared with SVM model,the variable structure SVM model has the better dynamic tracking ability and forecasting performance,and it can be used in daily water consumption forecasting.
出处
《应用基础与工程科学学报》
EI
CSCD
北大核心
2015年第5期895-901,共7页
Journal of Basic Science and Engineering
基金
重庆市科技项目(CSTC2006AB7020)
建设部科技资助项目(2001-45)
高等学校学科创新引智计划资助项目(B13041)
关键词
变结构
支持向量机
日用水量
动态预测
variable structure
support vector machine
daily water consumption
dynamic forecasting