中国农业气象 ›› 2009, Vol. 30 ›› Issue (S2): 290-294.

• 论文 • 上一篇    下一篇

基于遗传神经网络混合模型预测仙居县马尾松毛虫的发生量

朱寿燕;陈绘画;罗家进;   

  1. 浙江省仙居县气象局;浙江省仙居县林业局;
  • 出版日期:2009-12-20 发布日期:2009-12-20
  • 基金资助:

    仙居县科技局“仙居县林业主要有害生物数值预报的研究”(200628)

Forecast of Occurrence Quantity of Masson Pine Caterpillar(Demdrolimus punctatus Walker) based on GA-BP Mixed Model of Neural Network

ZHU Shou-yan1,CHEN Hui-hua2,LUO Jia-jin1(1.Meteorological Bureau of Xianju County,Xianju 317300,China;2.Forest Bureau of Xianju County,Xianju 317300)   

  • Online:2009-12-20 Published:2009-12-20

摘要: 针对BP算法易陷入局部极小、遗传算法具有全局寻优的特点,将二者结合起来形成一种训练神经网络的新算法——GA-BP算法。用均生函数法提取前期虫情信息,根据相关系数法和逐步回归法选择与仙居县马尾松毛虫有虫面积、虫口密度、虫株率相关关系密切的延拓均生函数序列和气象因子作为各预测模型的输入特征,分别建立马尾松毛虫有虫面积、虫口密度、虫株率与气象因子的GA-BP混合模型。结果表明:所建立的各GA-BP混合预测模型,具有令人满意的拟合精度和预测精度。当预报因子数为6个时,隐含层神经元个数为13个,3组预留有虫面积的平均预测误差为4.41%;虫口密度GA-BP混合模型的预报因子数为4个时,隐层神经元个数为9个,3组预留样本的平均预测误差为2.17%;虫株率GA-BP混合模型的预报因子数为4个时,隐层神经元个数为9个,3组预留样本的平均预测误差为4.25%。

关键词: 马尾松毛虫, 遗传神经网络, 发生量, 预测预报, GA-BP混合模型

Abstract: In order to overcome the shortcoming of BP neural network,which was easily sinking into partial smallest value,the Genetic Algorithm(GA) was combined with the BP neural network.A new algorithm,the GA-BP mixed model of trained neural network was applied to forecast the occurrences quantity of Masson Pine Caterpillar(Demdrolimus punctatus Walker).The results showed that the model had a satisfied fitting and forecast precision.

Key words: Masson pine caterpillar(Demdrolimus punctatus Walker), Genetic Algorithm(GA) neural network, Occurrence quantity, Forecast, GA-BP mixed model