摘要
支持向量机(Support Vector Machines,SVM)是一种具有坚实理论基础的新颖小样本学习方法。采用支持向量机回归(Support Vector Machine Regression,SVR)算法,用libsvm-2.89软件包对我国近年来的粮食产量进行回归预测,选择交叉验证法进行参数寻优,建立粮食产量和其影响因素的支持向量机回归模型。粮食产量预测平均相对百分误差为1.209%,均方根误差为581.191,相关系数为0.962 24。将预测结果与指数平滑模型、生产函数模型及多元线性回归模型进行了比较,用平均绝对百分误差、希尔不等系数及均方根误差对4种模型预测结果进行评价。结果表明,基于支持向量机的径向基核函数(RBF)模型预测粮食产量的精度优于其他预测方法。
Support Vector Machine Regression, which has sound theoretical foundation,is a novel learning way for small sample. A SVR prediction model of grain output in recent years was made by libsvm-2.89, with parameter optimization using cross-validation method in China. The model between grain production and its impact factors was also set up with the optimal parameters. The average relative percentage error and root mean square error between actual test and pre- dicted grain yield are 1. 209% and 581. 191, respectively,and the correlation coefficient is 0. 96224. At the same time, in order to further explain the advantages of SVM regression prediction, the predicted results were compared with ex- ponential smoothing model, the production function model and multivariate linear regression models. We valued the four kinds of models of prediction with the average absolute percentage error, theil inequality coefficient and root mean square error. The results of the evaIuation showed that, the accuracy of SVM Radial Basis Function （RBF） model on grain yield prediction was higher than the other three models.
出处
《山西农业大学学报（自然科学版）》
CAS
2013年第4期357-361,共5页
Journal of Shanxi Agricultural University(Natural Science Edition)
基金
高等学校博士学科点专项科研基金(20121403110002)
山西省归国留学人员科研基金(2013-061)
关键词
支持向量机
回归预测
参数选择
粮食产量
Support vector machine
Regression Prediction ~ Parameters choice
Grain output
