Research On Prediction Of PM₁₀ Pollution In Heating Season In Xi'an City

The prediction of PM₁₀ pollution is studied according to the PM₁₀ concentration data daily in heating season(November 15 of each year?the March 15 of the following year) and the ground meteorological data daily that monitored from 2001 to 2007.The essay analyzes characteristic of PM₁₀ concentration's changes yearly of Xi'an City,contrasted the characteristics of PM₁₀ concentration in heating season and non-heating season.The average concentration of PM₁₀ in heating season is more than non-heating season.Therefore,the article analyses PM₁₀ concentration of heating season and synchronous meteorological data of Xi'an City.PM₁₀ pollution is affected by many meteorological factors,through correlation analysis between PM₁₀ pollution and meteorological factors,choosing PM₁₀ concentration of former day,14 visibility,08visibility,average wind speed,average humidity,08 humidity,08 dew temperature,sunshine hours,08 temperatures-08 dew temperature,average pressure,08 Pressure,08 temperature altogether 12 meteorological factors affecting the heating season PM₁₀ pollution as the forecasting model's input factors.The genetic algorithms combined with BP neural network in this article.First of all,the genetic algorithms optimize connection value of BP neural network.And then, putting the optimized connection value into the 3 layers BP neural network model which has 12-4-1 structure to train.Finally,using the trained GA-BP models to forecast heating season's PM₁₀ concentration of Xi' an.The results showed that the required precision was reached when genetic iterations reach to 6,training step reaches to 60.The model make the learning time shorter,make the speed of convergence faster and the better fitting.The forecast accuracy is to be 84.9%.The article also use the stepwise regression method,BP neural network model to forecast the same data with the same input factors.The forecast accuracy rate were 76.4%,81.9%respectively,and compare the results of stepwise regression method and BP neural network with the results of GA-BP.Through the comparison,we found that GA-BP model's accuracy is higher and capability is better relatively.