BP网络模型在沙尘暴预测中的应用研究  

Study on Sandstorm Forecasting with BP Neural Network Method

作　　者：左合君[1] 勾芒芒 李钢铁[1,3] 李兴[4] 

机构地区：[1]内蒙古农业大学,内蒙古呼和浩特010019 [2]内蒙古水利科学研究院,内蒙古呼和浩特010020 [3]北京林业大学,北京100083 [4]内蒙古师范大学,内蒙古呼和浩特010022

出　　处：《中国沙漠》2010年第1期193-197,共5页Journal of Desert Research

基　　金：国家林业软科学研究先导专项"国外干旱土地退化治理技术与政策研究"项目(2007-14);内蒙古农业大学博士基金"沙尘暴时空分布规律的研究"项目(BJ-04-24)资助

摘　　要：依据锡林郭勒地区30a气象资料,应用人工神经网络(ANN)中不同BP网络结构和算法,探索建立沙尘暴预测模型的方法。研究认为,在建立锡林郭勒地区沙尘暴预测模型时,选择年大风日数、年平均地温、年蒸发量、相对湿度4个气象因子作为模型的输入因子是合理的;经过输入因子确定,层数、节点选择,每层激活函数和输出因子的确定,锡林郭勒地区沙尘暴预测模型可采用三层网络结构(4-6-1)。比较和试算显示,快速BP算法较普通BP算法的训练速度快,收敛精度高64.47%;快速BP神经网络的沙尘暴预测模型的预测精度可达到98%,较传统的多元线性回归数学模型高。因此,应用快速BP神经网络建立沙尘暴预测模型简捷、方便,具有精度高、智能化等特点,可在区域沙尘暴预测预报领域推广。The different BP structures and algorithm of artificial neural network（ANN）are applied to seek the sandstorm forecasting method based on the meteorological data for 30 years in Xilingela,Inner Mongolia.It is logical to select the four meteorological factors yearly strong wind days,mean annual ground temperature,evaporation and relative humidity,as the input in the forecasting model.The forecasting model of sandstorms in Xilingela has three network structures（4-6-1）,flowing from input factors determination to layer and node choice then to function activation of each layer and output factors determination.It can be conclude that the accelerated BP algorithm has faster training speed and higher convergence accuracy（6447% higher）compared with the normal BP,and can reach the high forecasting precision of 80%,much larger than that of traditional multi-linear regression model.This sandstorm forecasting model based on accelerated BP neural network has characteristics of simplicity,convenience,high precision and intelligentization,and so can be extended in field of regional sandstorm forecast.

关 键 词：沙尘暴 BP神经网络 预测模型 

分 类 号：P457[天文地球—大气科学及气象学]