基于支持向量机的害虫多维时间序列预测

Pest multiple dimension time series forecasting based on SVM

针对害虫发生数据高度非线性特点导致传统方法预测准确率低的难题,提出一种基于支持向量机(SVM)的多变量自回归(CAR)的害虫时间预测方法(SVM_CAR)。SVM_CAR首先利用SVM以留一法的MSE最小化原则进行时间序列非线性定阶;然后用SVM对害虫发生的影响因子进行非线性筛选,并同时通过强制汰选给出各保留因子对预测结果的相对重要性;最后建立基于保留对预测结果影响较大因子的SVM_CAR预测模型。以大豆食心虫虫食率与晚稻第5代褐飞虱发生量两个实例数据集进行验证性实验,SVR-CAR比五种参比模型的预测精度都要高,实验结果表明,SVM_CAR更能反映害虫发生时间序列样本集的非线性动态规律,在害虫预测中有着广泛的应用前景。

To solve the problems of pest prediction accuracy,this paper proposed a novel high precision method of pest forecasting based on integrated controlled autoregressive （ CAR） and support vector machine （ SVM） . Firstly,determined the model order by leave-one-out method based on the minimum mean square error with SVM,secondly,screenning the better variable and the order of reserved variable screened compulsorily with SVM,lastly,built model with the reserved variable based on SVM. The prediction results of the two samples set,which are moth-eaten,ratio of Leguminivora glycinivorella Mats and occurrence of 5th generation of rice brown planthopper,show that SVM_CAR has the highest prediction precision in all reference models,and can reflect nonlinear dynamic discipline of pest occurrence time series,can be widely used in the prediction of pest time series forecasting.