基于BP神经网络的稻瘟病预测技术研究  

Rice Blast Forecasting Based on BP Neural Network

作　　者：李晓菲[1] 刘振华[1] 陈涛[1] 王园媛[1] 王云月[1] 

机构地区：[1]云南农业大学,农业生物多样性与病害控制教育部重点实验室,云南昆明650201

出　　处：《云南农业大学学报（自然科学版）》2013年第4期551-560,共10页Journal of Yunnan Agricultural University:Natural Science

基　　金：云南省现代农业水稻产业技术体系项目(A3006517)

摘　　要：利用BP神经网络技术,以云南省的勐海和石屏作为试验点,选取稻瘟病发生相关气象因子及田间叶瘟病情指数开展稻瘟病的预测预报研究。采用Trainlm与Traingdx训练函数讨论了隐节点数及学习率取值并建立了相应的预测模型。勐海两个预测模型10年历史数据拟合精度分别为87.65%,92.93%,石屏两个预测模型9年数据平均拟合精度分别为93.48%,87.8%。2011年,勐海模型预测精度分别为95.96%,97.6%,石屏模型预测精度分别为94.74%,83.35%,优于逐步回归模型的72.33%,34.02%。BP神经网络预测模型的拟合精度和预测精度都达到80%的预期目标,预测效果较为理想,在稻瘟病预测中具有优势,对稻瘟病防治工作的开展及云南省稻瘟病预测技术的更新具有一定指导意义。In this study,we selected meteorological factors and leaf blast disease index as predictive factors to study the application of BP neural network in rice blast forecasting in Menghai and Shiping.BP neural network forecasting models were built up using Traingdx and Trainlm and the best value of hidden nodes and learning rate were discussed.The ten years historical data fitting accuracy of Traingdx and Trainlm were 87.65%,92.93% respectively in Menghai,and the nine years historical data fitting accuracy respectively were 93.48%,87.8% respectively in Shiping.In 2011,the predict accuracy were 95.96%,97.6% respectively in Menghai and the predict accuracy were 94.74%,83.35% respectively in Shiping,which were superior to the predict value（72.33%,34.02%） of stepwise regression model.The fitting and predict accuracy all reached the expected target of 80%,indicating that BP neural network used in rice blast forecasting is more preponderant.The forecasting model built by BP neural network could provide scientific basis for rice blast management and update of forecasting technology in Yunnan.

关 键 词：稻瘟病 预测模型 BP神经网络 

分 类 号：S431.9[农业科学—农业昆虫与害虫防治]