PM₁₀ Pollution Forecast Based On BP Neural Network And MATLAB Implementation In Xi'an City

Inhalable particulates(PM₁₀) has been the primary pollutant in the air of Xi'an city for a long time.For better reflecting the pollution change tendency,strengthening pollution control and preventing serious pollution accidents,to study the pollution forecasting methods,develop pollution forecasting work are of great significance.As atmospheric environment complicated,variable,and so much monitoring data has been accumulated,it is hard for traditional forecasting methods to take full advantage of the useful information to achieve accurate forecast,so does the popularization.In this paper,Artificial Neural Network technology was applied to the field of air pollution forecast,with the powerful non-linear processing capacity of neural network,PM₁₀ pollution forecasting model based on BP neural network in Xi'an was built up by means of software MATLAB.First,with simple linear regression and principal components analysis,the alternative forecasting factors were retrenched from 28 to 11,that's the input of the model.Then,by comparing the training effects of the six different BP algorithms such as momentum BP algorithm,BFGS quasi-Newton algorithm,SCG algorithm and so on,it came to the conclusion that SCG algorithm was the training algorithm most suitable for the forecasting model.Finally,through changing the hidden note number and training times,by horizontal and vertical comparison of the training and forecasting effects,it showed that the optimal hidden note number was 5.Till then,the PM₁₀ pollution forecasting model of Xi'an was established.To improve the adaptability of the network,early termination algorithm based on SCG algorithm was applied in the paper,and testing sample sets were used to test the simulation effects of the forecasting model.The results indicated that the correlation coefficient of forecast value and actual value was 0.801,of the 265 testing samples,the number of days when forecast value completely conformed to the actual value was 212,which accounted for 80%;regarding the difference not more than one grade as accurate,then 262 days met the requirements,which accounted for 98.87%.In a word,the forecast results were basically consistent with the actual situation,the conclusion was intuitive and the effects were ideal.The research proved in practice that it was feasible to apply the artificial neural network in air pollution forecasting in Xi'an city.It provided a new way of thinking and method for urban air pollution forecasting in information society, meanwhile,an effective,convenient software for modeling was found-MATLAB.