烟台地区水面蒸发BP神经网络模型研究  

Water Surface Evaporation BP Neural Network Model in Yantai Region

作　　者：李清翠[1] 张振华[1] 姚付启[1] 

机构地区：[1]鲁东大学地理与资源管理学院,山东烟台264025

出　　处：《灌溉排水学报》2008年第4期122-124,共3页Journal of Irrigation and Drainage

基　　金：国家自然科学基金项目(50609022)

摘　　要：运用BP神经网络模型对水面蒸发进行研究,并与多元线性回归和主成分回归2种方法的拟合结果进行比较。结果显示,多元线性回归各参数均通过t检验,拟合较好;主成分回归中,参数b2未通过t检验,拟合效果不如多元线性回归好。BP神经网络模型、多元线性回归、主成分回归建立的水面蒸发量观测值与拟合值的回归方程中决定系数分别为0.8986、0.7993、0.7984。应用BP神经网络进行分析,相对误差小于10%的样本个数超过总样本个数的40%,相对误差不超过30%的样本个数接近80%;而其它2种方法相对误差大于10%的样本个数超过总样本数的80%,相对误差大于50%的接近总样本个数的30%。可见,应用BP神经网络模型进行水面蒸发量计算,远优于其它2种方法,应用此方法进行水面蒸发量的预测是非常理想的。This paper try to use BP neural network model on the research of water-surface evaporation, and the fitting result is compared with multiple linear regression and main component regression. The result shows that all the parameters have passed t test in the multiple linear regression, and the fitting result is good, but in the main component regression, parameter b2 does not pass t test, the fitting result is worse than multiple linear regression. The water surface evaporation measured value and fitted value regression equation found by BP neural network.multiple linear regression and main component regression, and their coefficient of determination is 0. 8986,0. 7993,0. 7984. When analyzing with BP neural network model, nearly 40% sample＇s relative error less than 10%, more than 80%sample＇s relative error less than 30%. More than 80% sample＇s relative error is more than 10% in the other two methods; relative error which more than 50%is nearly all sample＇s 30%. We know that BP neural network model is far off excelled than the other two methods, The use of BP neural network model on the research of water surface evaporation is quite ideal.

关 键 词：水面蒸发 BP神经网络模型 多元线性回归 主成分回归 

分 类 号：P333.1[天文地球—水文科学]