Study on Sandstorm Forecasting with BP Neural Network Method

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.