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、动量因子和学习因子对模型效果的影响可以忽略不计 By using the method of optimizing relevant census, we selected the predictor that has a significant indication to the extent of wheat scab in Taihu Lake, Jiangsu Province, and used it as input variables to study training with BP algorithm of multi-layer feedforward neural network. Artificial neural network prediction model of scab rate in panicle disease. The influence of structural parameters on the model results is analyzed. It is found that the square sum of training errors has the most significant effect on the model. The fitting rate of historical samples increases steadily with the decrease of the sum of square errors. However, When the value is too small, the prediction accuracy of the model for the independent samples decreases. When the sum of the square of the total error and the appropriate value make the model stable, the influence of the hidden layer nodes, momentum factor and learning factor on the model’s effect is negligible