局部支持向量回归在小麦蚜虫预测中的研究与应用

Research and Application of Local Support Vector Regression in Prediction of Wheat Aphid

针对小麦蚜虫预测预警准确率不高的问题,本文提出了一种基于局部支持向量回归的小麦蚜虫短期预测算法。首先用相关分析法进行特征选择,然后进行归一化处理,最后使用局部支持向量回归进行小麦蚜虫百株蚜量短期预测模型的构建,并对未知样本进行预测。利用1990~2013年山东省烟台地区的小麦蚜虫数据及气象数据进行实验,并与标准的支持向量回归进行对比试验。局部支持向量回归的预测以及回代的均方误差为196362和198780,准确率为82.69%和91.03%;支持向量回归的预测以及回代的均方误差为199366和213108,准确率为80.77%和91.03%。实验结果表明,对于小麦蚜虫的短期预测,局部支持向量回归在准确率和推广能力上均明显优于支持向量回归。

Aiming at the accuracy of wheat aphid prediction is low, this paper proposed a short-term forecasting algorithm of wheat aphid based on local support vector regression. Firstly, feature selection was realized by correlation analysis. Secondly,the normalized processing of selected features was calculated. Finally, the short-term forecasting model was established and the prediction value of test sample was obtained by the established model. Experiments were conducted on the wheat aphid data and meteorological data of Yantai area from 1990 to 2013 year and contrast test was conducted by the standard support vector regression. The standard support vector regression achieved the Mean Square Error at 199366 in prediction and213108 in back-substitution check, the accuracy at 80.77% in prediction and 91.03% in back-substitution check, while local support vector regression achieved the Mean Square Error at 196362 in prediction and 198780 in back-substitution check, the accuracy at 82.69% in prediction and 91.03% in back-substitution check. The results showed that local support vector regression has better performance in accuracy and generalization ability for the short-term prediction of wheat aphid.