摘要
根据相关系数法、均生函数法及逐步回归法分别选取与马尾松毛虫有虫面积、虫口密度、虫株率相关关系密切的气象因子和延拓均生函数序列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。当隐含层神经元个数为13个,预报因子数为6个时,3组预留有虫面积的平均预测误差为4.41%;虫口密度GA-BP混合模型的隐层神经元个数为9个,预报因子数为4个时,3组预留样本的平均预测误差为2.17%;虫株率GA-BP混合模型的隐层神经元个数为9个,预报因子数为4个时,3组预留样本的平均预测误差为4.25%。
According to correlation coefficient, mean-generating function and stepwise regression methods, meteorological factors corresponding to the area of infested wood, the population density and the attacking rate of Dendrolimus punetatus Walker, and continuation-mean-generating function sequence weie selected as the imported characters to establish the GA-BP mixed models. Results showed that the established GA-BP mixed models had satisfactory fitting accuracy and forecasting precision. Under the condition of the number of hidden neurons was 13 and the number of forecasting factors was 6, the average forecasting error of 3 groups of reserved area of infested wood was 4.41% . Set the number of hidden neurons with 9 and the number of forecasting factors with 4 under the GA-BP mixed model for the population density and the attacking rate, the average forecasting error of 3 groups of reserved samples were
2.17% and 4.25%, respectively.
出处
《安徽农业科学》
CAS
北大核心
2009年第12期5548-5551,共4页
Journal of Anhui Agricultural Sciences
基金
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)
