摘要
提出一种基于ARIMA和动态ε支持向量机(ε-DSVM)的组合预测模型(ARIMA-ε-DSVM),预测松毛虫发生面积.先采用ARIMA模型进行时间序列线性趋势建模,为非线性部分确定输入阶数,根据确定的输入阶数进行时间序列样本重构,再采用ε-DSVM模型进行时间序列非线性特征建模,将这两模型预测值相加得到组合模型预测值.对辽宁省朝阳市松毛虫时间序列进行仿真试验,结果表明,ARIMA-ε-DSVM模型预测精确度比单一模型ARIMA和SVM及简单组合模型ARIMA-SVM要高,ARIMA-ε-DSVM模型大幅度改善预测效果,显著地减少预测误差,泛化能力强.
A novel forecasting model combinating autoregressive integrating moving average（ARIMA） with dynamic ε-insensitive cost function support vector machine（ε-DSVM）was brought forth, which showed the complicated and dynamic characteristics of Dendrolimus punctatus occurrence. ARIMA model was used to capture the linear feature of the time series and ε-DSVM model to fit the nonlinear component of the time series to obtain the ensemble forecasting result by adding ARIMA to ε-DSVM. The prediction performances of the method was tested by Dendrolimus punctatus occurrence, and the results showed that the hybrid model, which took advantage of the unique strength of the two models in linear and nonlinear modeling, had better accuracy than the single model and simple ensemble forecasting model incorporating ARIMA and SVM. As a novel model combinated ARIMA with ε-DSVM , the combinatining model had the advantages of structural risk minimization and non-linear characteristics, which was suitable for small samples, being able to avoid the over-fit. It is a new, powerful tool in pests forecasting work.
出处
《湖南农业大学学报（自然科学版）》
CAS
CSCD
北大核心
2010年第4期430-433,共4页
Journal of Hunan Agricultural University(Natural Sciences)
基金
国家自然科学基金项目(30570351)
关键词
支持向量机
松毛虫
时间序列
差分自回归移动平均
support vector machines
Dendrolimus punctatus
time series
auto regression integrated moving average（ARIMA）
作者简介
向昌盛（1971-），男，湖南怀化人，博士，副教授，从事生物信息学及农业昆虫与害虫防治研究，cx5243879@sohu．com
