基于人工神经网络的马尾松毛虫发生量预测模型的研究

运用人工神经网络的原理和方法，根据相关系数法和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作为样本的输入特征，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BP网络模型。结果表明，所建立的各BP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15个，预报因子数为8个时，2组预留有虫面积的2a平均预测误差为3.15%，虫口密度BP模型的隐层神经元个数为8个，预报因子数为6个时，预留样本的平均预测误差为5.91%，虫株率BP模型的隐层神经元个数为4个，预报因子数为5个时，预留样本的平均预测误差为10.65%。

A Study on the Forecast Model of Dendrolimus punctatus Occurrence Based on Artificial Neural Network

The principle and methodology of artificial neural network were used to select some meteorological factors closely correlated to the occurrence area, population density and damage rate by the methods of correlation coefficients and step regression. The BP network models of meteorological factors and occurrence area, population density and damage rate of Dendrolimus punctatus were established. The results showed that these BP models established have satisfied fitting and forecast precision. When the amount of implicit layer neuron is 15 and the amount of forecast factor is 8, the mean error of forecast of 2 groups of reserved occurrence zone was 3.15% in two years. When the amount of implicit layer neuron is 8 and the amount of forecast factor is 6,the mean error of forecast of reserved occurrence zone was 5.91%, while when the amount of implicit layer neuron is 4 and the amount of forecast factor is 5, the mean error of forecast of reserved occurrence zone was 10.65%.