基于支持向量机回归模型的稻田二化螟历史数据预测

Forecasting on historical data of Chilo suppressalis in paddy field based on support vector regression model

通过1991—1996的历史数据分析稻田早稻生物学特征与不同土壤处理对二化螟发生株率的非线性相关关系,测试支持向量机回归(SVR)模型在二化螟测报的可行性。结果表明,应用epsilon-SVR模型预测水稻综合因子观测场1996年的早稻二化螟平均发生株率预测准确率达97.95%,而阴离子观测场的平均发生株率预测准确率达96.97%。该回归模型表现出良好的鲁棒性和自学习能力。因此,SVR模型适于二化螟田间发生株率的预测,在虫害测报中应用前景广阔。

The aim was to analysis the non-linearity relationship between the Chilo suppressalis occurring rate and the data of the early-season rice＇s biological characters and the agrological treatments in paddy field from 1991 to 1996, and the another aim was to test the probability of forecasting C. suppresscdis in field with support vector regression （SVR） model. Using an epsilon-SVR model, the data of this station in 1996 were forecasted. The results showed that veracity of the average occurring rate in the observation field of paddy comprehensive factors was 97.95%, and the predictive veracity in the observation field of anion was 96.97%, furthermore this model had the better robustness and flexibility. The SVR model could be used in forecasting Chilo suppressalis occurring rate in paddy field, and SVM method had a better outlook for the pest＇s forecasting.