摘要
引入遗传算法优化BP神经网络权重和阈值的方法建立黄土坡面产流入渗模型。模型以雨强、降雨历时、表层40 cm土壤前期含水量、坡度值为输入项,径流量、入渗量为输出项,用实测资料对网络进行模拟和预测。模拟结果平均误差6.32%和1.93%,预测结果平均误差为5.71%和1.92%。并与传统BP神经网络模型和定雨强Philip回归模型的预测入渗结果进行了对比,结果表明:遗传算法优化BP神经网络模型的预测效果要明显优于传统BP神经网络模型和定雨强Philip入渗模型,三种模型入渗预测结果平均误差分别为1.92%,5.29%,9.10%,最大误差分别为6.48%,25.88%,20.36%。
The method of back-propagation neural networks optimized by genetic algorithms was used to establish a hillslope runoff and infiltration model.The rainfall intensity,rainfall duration,initial soil water content and slope gradient were selected as the model inputs,and the runoff volume and infiltration volume were the model outputs.Through simulating and predicting,the results showed that simulation mean reletive errors were respectively 6.32% and 1.93%,and the prediction mean reletive errors were 5.71% and 1.92%.In order to compare the prediction effects of the model with those of other models, the unoptimized back-propagation neural network model and the Philip regression model under the condition of fixed rainfall intensity were applied to predict the infiltration amount,and the comprasion results showed that the mean reletive errors of the three models in infiltration amount prediction were 1.92%,5.29% and 9.10%,respectively,while the maximum mean reletive errors were 6.48%,25.88%,20.36%,which showed that the prediction effects of optimized back-propagation networks had better performance than the other two models obviously.
出处
《干旱地区农业研究》
CSCD
北大核心
2011年第2期209-212,共4页
Agricultural Research in the Arid Areas
基金
国家科技支撑计划项目(2007BAD88B05)
陕西省自然科学基金项目(2007E235)
陕西省教育厅自然科学计划项目(08JK406)
关键词
神经网络
遗传算法
入渗
产流
预测模型
neural networks
genetic algorithm
runoff
infiltration
prediction model
