小麦赤霉病神经网络和偏最小二乘预测模型

选取2004-2014年间发病期的气象因素为自变量，小麦赤霉病病穗率为因变量，借助神经网络的函数映射能力，采用Fletcher-Reeves算法的变梯度反向传播算法，建立了小麦赤霉病的气象预报模型.由于神经网络无法提供直观的函数来反映病穗率与气象因子之间的关系，为了进一步分析气象因子间的相关性，采用主成分分析法提取主成分，并利用回归分析得到线性函数关系，建立了偏最小二乘模型.神经网络预测模型平均预报精度达到99%，但只提供病穗率和气象因子之间的拓扑关系;偏最小二乘预测模型可以得到病穗率和气象因子之间直观的函数关系，模型平均预报精度达到97%.2种模型均具有较高的预报精度，对小麦赤霉病的预防工作具有一定的参考价值.

Prediction Models of Neural Network and Partial Least Squares for Wheat Scab

Prediction models of neural network and partial least squares for wheat scab are proposed，which provides a scientific basis for prevention of wheat scab.The meteorological factors and morbidity rate between 2004 and 2014 in Anhui are chosen as independent variables and dependent variable，respectively.The alternating gradient back-propagation algorithm of Flecher-Reeves is employed to build the neural network model.Since the simple functional relationship between meteorological factors and morbidity rate can't be provided by the model，the correlation among meteorological factors is analyzed and the partial least squares model is built through principal component analysis and regression analysis.The accuracy of the neural network model is about 99%，through which the topology between meteorological factors and morbidity rate is provided.The linear function is obtained through the partial least squares model and the accuracy is about 97%.Both the neural network model and the partial least squares model can achieve high prediction accuracy and provide scientific guidance for the prevention of Anhui wheat scab.