Study On Sandstorm Forecast Model By Using BP Neural Network In The Xilin Gol Area

Sand storm is a common natural disasters in arid and semi-arid areas in China,and it is an an important indicator of desertification.This phenomenon has caused serious harm to people.With the development of computer performance artificial intelligence technology has been researched and applied in various fields,and it is a very important mean to study sand strom.Based on the meteorogical date during 1971-2000 from around 15 meteorogical stations, this paper analyzed the features and changes of sand storms in Xilin Gol league in Inner Mongolia.Fistly, SPSS software is applied to study main meteorological factors which is relate to sand storm by related analysis and cluster analysis.Secondly, Surfer software is applied to study the features and changes of time and space about sandstorm.Finaly, apply to different BP structure and algorithm of artificial neural network(ANN) to explore model of sandstorm.The main conclusions as follow:1. There are many climate factors which influence the frequency and strength of the sandstorm in Xilin Gol league in Inner Mongolia, which is many dimensions problem.Influence of climate conditions are most obvious.In the number of days of strong winds, the average annual ground temperature, the annual evaporation, relative humidity are correlated to the number of days of sand storm.Strong wind is the most important and correlation coefficient is up to 90% with sand storm.2. In the recent 30 years, the trend of sand storm.is declining and achieves the lowest level in the 90s; Occurrence season of sand storm is mainly spring and the frequency of the year accounted for more than 70%; Days of sand storm in west is more than in east, Which the annual average number of days of sand storm is maximal in SU Youqi for 10 days.3. Through determining the input factors, layer, node choice, each activation function and output factor, can use three network structures (4-6-1) in the forecast model of sandstorm. After compare and calculation, the accelerated BP algorithm is better than normal BP both in training speed and convergence accuracy, and restraining precision high 64.47%. The comparison is made between the accelerated BP algorithm and the mathematic model of linear regression, and the conclusion of comparison shows that precision of the accelerated BP algorithm is near 98%.4. It can be shown that days of sand storm are increasing from 2001 to 2010 by using BP neural network forecast model.