摘要
支持向量回归机(Support Vector Regression,SVR)在供热负荷预测中得到了一些研究,然而模型的拟合度和泛化能力依赖于其相关参数的选取,需要足够的先验信息,寻优过程存在难度.针对上述情况,提出采用交叉验证(Cross Validation,CV)的思想对其中的重要参数(惩罚因子C和RBF核函数参数γ)进行网格划分,在训练集中自动寻找最佳参数,从而得到最佳的训练模型,并用该模型对测试集进行回归预测.以某热源数据进行了实验研究,结果表明:该方法能够快速建立预测模型,有效地预测供热负荷,具有较高的拟合度和较强的泛化能力.
Support Vector Regression（SVR） has achieved some results in the research of heating load forecast- ing. However, the fitting degree accuracy and generalization performance of the SVR models depend on the selection of its parameters, optimization process is difficult to find, with adequate prior information. In view of this situation, the thought of Cross Validation（CV） was put forward to carry on the grid division for sever- al important parameters（the penalty factor C and RBF kernel function parameter y）, which could search the best parameter in the training set automatically, so as to obtain the optimal model of regression prediction for the test set. In the experimental study of a heat source data, the results illustrate that this method can quickly establish prediction model, and effectively predict the heating load, with a high fitting degree and strong gen- eralization ability.
出处
《中北大学学报（自然科学版）》
CAS
北大核心
2014年第5期565-570,共6页
Journal of North University of China(Natural Science Edition)
基金
国家科技支撑计划项目(2012BAJ04B02)
关键词
支持向量回归
交叉验证
参数优选
热负荷预测
support vector regression
cross validation
parameter optimization
heating load prediction
