基于神经网络的冬小麦生育期日土壤水分预测模型研究

Prediction Model of Soil Moisture in Winter Wheat Growth Period Based on Neural Network

准确掌握土壤水分动态变化,对精准制定灌溉计划至关重要。采用五道沟实验站2018-2019年蒸渗仪日土壤水和同期7个气象要素(气温、降雨、水面蒸发、日照时数、风速、绝对湿度、地温)资料,采用BP神经网络方法建立冬小麦生育期不同土层(10、30、50 cm)的土壤水分预测模型,模型分别为BP(7-9-1)、BP(7-12-1)和BP(7-14-1),并用遗传算法优化上述BP神经网络模型。结果表明:两种模型均可用于冬小麦生育期土壤水分预测,其中遗传算法优化BP神经网络能够更好提高预测精度,且随着土层厚度增加,预测精度提高。BP神经网络土壤水分预测10、30、50 cm土层平均相对误差分别为6.2、4.0、2.9;遗传BP神经网络土壤水分预测10、30、50 cm土层平均相对误差为3.8、1.7、1.3。

Accurate mastery of soil moisture dynamic changes is crucial for accurate formulation of irrigation plans.In this study,according to the data of soil water and 7 meteorological elements(temperature,rainfall,evaporation of water surface,sunshine hours,wind speed,absolute humidity and ground temperature)of the same period from 2018 to 2019 by the evapotranspiration apparatus of Wudaogou experimental Station,the BP neural network method was used to establish soil moisture prediction model of different layers(10,30,50 cm)in winter wheat growth period.The models were BP(7-9-1),BP(7-12-1)and BP(7-14-1).And the genetic algorithm was used to optimize the BP neural network model.The results showed that both models could be used to predict soil moisture during the growth period of winter wheat,among which the BP neural network optimized by genetic algorithm could better improve the prediction accuracy,and the prediction accuracy increased with the increase of soil thickness.The average relative errors of BP neural network for soil moisture prediction in 10,30 and 50 cm soil layers were 6.2%,4.0%and 2.9%,respectively.The mean relative errors of soil moisture prediction in 10,30 and 50 cm soil layers by genetic BP neural network were 3.8%,1.7%and 1.3%.