摘要
为实现参考作物蒸散量(Reference Crop Evapotranspiration,ET_0)的准确计算和预测,利用广义回归神经网络(Generalized Regression Neural Network,GRNN)进行非线性映射,为减少人为因素影响采用粒子群算法(Particle Swarm Optimization,PSO)自动寻找神经网络最优参数,建立了基于粒子群算法和广义回归神经网络的参考作物蒸散量预测模型(PSO-GRNN)。研究气象数据缺失情况下模型模拟效果,在缺失风速和日照时数情况下,模型仍能取得较好效果(模型均方根误差和自相关系数分别为0.437%和91.865%)。将模型计算结果与Hargreaves、Priestly-Taylor、Makkink、Irmark-Allen等4种经验模型进行对比,得到模型的均方根误差和自相关系数为0.23%和97.709%,优于4种经验模型。以P-M模型求得的参考作物蒸散量为标准值,对2011-2015年预测得到的参考作物蒸散量进行求和,得到PSO-GRNN模型计算得到历年ET_0总和的相对误差为0.44%,优于4种经验模型(Hargreaves模型18.29%、Priestly-Taylor模型2.89%、Makkink模型3.27%、Irmark-Allen模型18.49%)。该研究建立的PSO-GRNN模型预测精度高,稳定性好,人为影响少,能够较好的进行ET_0模拟计算,为作物需水量智能决策提供参考。
Reference crop evapotranspiration （ETo ） is an essential parameter of water resources planning and management. Accurate estima- tion of ETo becomes vital in planning and optimizing irrigation schedules and irrigation system management. In order to calculate and predict the reference crop evapotranspiration .accurately, a forecast model is built by using Generalized Regression Neural Networks, and in order to reduce artificial factors＂ effect, Particle Swarm Optimization algorithm is used to automatically search for the optimal parameters of the Neural Network. The model is trained and tested by meteorological data of Xi＇an, Yulin and Yan＇an of Shaanxi, China. The evaluation criteria of root mean squared error（RMSE） , coefficient of determination and model efficiency（EF） is used for the comparison. The results indicate that PSO -GRNN model performs better than Hargreaves model, Priestly-Taylor model, Makkink model and Irmark-Allen model. The research can provide a model to estimate ETo accurately and build water demand decision support systems.
出处
《中国农村水利水电》
北大核心
2017年第6期1-7,共7页
China Rural Water and Hydropower
基金
"十三五"国家重点研发计划项目"西北典型农区高效节水灌溉技术与集成应用"(2016YFC0400201)
国家科技支撑计划项目"规模化农业综合节水技术集成与示范"(2015BAD24B00)
关键词
参考作物蒸散量
广义回归神经网络
粒子群算法
预测模型
reference evapotranspiration
generalized regression neural network
particle swarm optimization algorithm
prediction model
