模型参数对赤霉病人工神经网络预报模型精度的影响

The Influence of Parameters on Prediction Accuracy of ANN Model for Winter Wheat Scab

运用最优化相关普查方法 ,选取确定了对江苏省太湖地区小麦赤霉病发生程度有着显著指示意义的预报因子 ,将其作为输入变量经多层前馈型神经网络的 BP算法进行学习训练 ,建立了赤霉病病穗率的人工神经网络预报模型。分析了结构参数对模型效果的影响情况 ,发现训练的总体误差平方和对模型的效果影响最为显著 ,历史样本的拟合率随着总体误差平方和的减小而稳定上升 ,但总体误差平方和取值偏小时模型对独立样本的预报精度下降 ;当总体误差平方和取适当值使模型稳定时 ,隐含层节点数。

By dint of optimum correlation method the predictors,which are good indicators of winter wheat scab occurrence in Taihu area in Jiangsu province,were input into the feed forward multi layer artificial neural network(ANN) based on back propagation study.ANN prediction model for occurrence of winter wheat scab was developed.The influence of model parameters on the fitting and prediction accuracy of model was studied.It can be concluded that total error square sum(E) has the most outstanding impact on model function than other parameters.The smaller is the value of E,the higher is fitting accuracy for historical samples.Extreme small E will result in lower prediction accuracy for independent samples.The action of number of neurons in the hidden layer (N h),study factor( α ) and momentum factor (β) may be ignored when suitable E was given to make model stable.