摘要
以对数线性去趋势法平稳化多维时间序列因变量,结合地统计学的后效时间长度与支持向量回归进行因变量自动快速定阶及自变量非线性筛选,最后利用一步预测法验证模型的外部预测能力。将该方法应用于两种害虫发生量预测,其预测均方误差均优于其他参比模型。该方法具有地统计学半变异函数模型与支持向量回归的诸多优点,适用于受多因素影响的非线性多维时间序列的预测预报。
Using logarithm linear detrended analysis method to make the dependent variable of multidimensional time serial stationary, and then combined with the length of timeliness from geostatistics and support vector regression to conduct automatically quick order selection for dependent variable and nonlinear screening for independent variable,finally using one-step prediction method to verify the external predicting capacity of the model. Applying the method in predicting of two pests amount, and the mean square error of its predicting results were better than the other selected model's. Therefore, the method has both the advantages of semi variation function model from geostatistics and the advantages of support vector regression, thereby suiting to predict non-linear multidimensional time serial affected by multi-factors.
出处
《湖南农业科学》
2013年第10期64-67,共4页
Hunan Agricultural Sciences
基金
教育部高等学校博士点专项基金(200805370002)
关键词
多维时间序列
地统计学
支持向量回归
害虫发生量预测
模型
multidimensional time serial
geostatistics
support vector regression
predicting of pest amount
model
