支持向量机在害虫发生量预测中的应用  

Application of Pest Occurrence Prediction Based on Support Vector Machine

作　　者：向昌盛[1] 周子英 张林峰[3] 

机构地区：[1]湖南农业大学东方科技学院,湖南长沙410128 [2]湖南辰溪县进修学校,湖南辰溪419500 [3]湖南农业大学信息科学技术学院,湖南长沙410128

出　　处：《生物信息学》2011年第1期28-31,共4页Chinese Journal of Bioinformatics

基　　金：湖南省教育厅科学研究资助项目(10C0803)

摘　　要：害虫发生量与其影响因子之间具有复杂的非线性和时滞性关系,传统方法不能很好的分析和拟合高度非线性的害虫发生量变化规律,导致预测精度不理想。为了有效构建害虫发生量与其影响因子之间复杂的非线性关系模型,提高害虫发生量预测精度,提出一种基于支持向量机的害虫发生量预测方法。该方法首先通过F测验对害虫发生量的最佳时滞阶数进行确定,并利用最佳时滞阶数对样本进行重构;然后利用前向浮动因子筛选法对害虫发生量的影响因子进行筛选,筛选出对预测结果贡献大的影响因子;最后采用10折交叉验证得到害虫发生量的最优预测模型。采用粘虫的幼虫发生密度数据在Mat-lab7.0平台下对该方法进行测试与分析,实验结果表明,相对于其它预测方法,支持向量机提高了害虫发生量的预测精度,克服了传统方法的缺陷,更适合于非线性、小样本的害虫发生量预测。There are complex nonlinear relations between pest occurrence and its influence factors,the traditional method can not analysis and fit the nonlinear factors and prediction accuracy is low.In order to improve the pest occurrence prediction accuracy and build the complicated nonlinear prediction model between pest occurrence and its influence factor,a pest occurrence prediction model is proposed based on support vector machine.Firstly,the time series model-order is determined by nonlinear expansion of samples with F test;secondly,a floating search method for factor selection is carried out;lastly,the optimal pest occurrence prediction model is got by ten fold cross-validation based support vector machine.The method is tested and analysis by armyworm larvae density data on Matlab 7.0,the experimental results show that,compared with other prediction method,support vector machine prediction accuracy has improved a lot and overcame the defects of the traditional method;the method is more suitable for nonlinear,small sample pest occurrence prediction.

关 键 词：害虫发生量 支持向量机 预测模型 交叉验证 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]