Neural Network Based Prediction Of Harbin Urban Air Quality

In recent years,the heavy pollution weather in Harbin is frequent,which is concerned by the masses,the society and the government,however,the acquisition of real-time data of air environmental quality can not meet the requirements of environmental air quality management of predictive early warning,and the sources of air pollution are difficult to predict and have low precision.Therefore,it is necessary to adopt scientific and effective methods to analyze and predict air quality based on a large amount of air quality monitoring data.In this paper,Harbin city is taked as the research object,according to the monitoring data of the 2013-2016 Harbin air automatic monitoring point and regular surface meteorological data at the same time,using BP neural network,the prediction model of air quality in Harbin city was established,and the model was used to predict the air quality in summer and winter in 2016.First of all,three prediction methods used in air quality prediction are introduced in this paper,and the premise,advantages and disadvantages of each method are analyzed and compared.Then the BP neural network method is selected,and the sample selection,training steps and implementation process of air quality prediction are described.Secondly,the annual variation of six pollutants from 2013 to 2016 and the monthly variation of PM_2.5,PM₁₀,SO₂,NO₂,CO and O₃ concentrations from January to December 2015 were analyzed in this paper.The result of the annual variation of SO₂ from 2013 to 2016 was between 0.028 and 0.057,and the result of the 24h variation of CO was between 1.3 and 1.7,the above two indicators were lower than level 2 of the national standard.The result of the annual variation of NO₂was between 0.044 and 0.056,and the result of the annual variation of PM ₁₀0 was between 0.074 and 0.119,and the result of the annual variation of PM_2.5 was between 0.052 and 0.081,and the result of the 8h variation of O₃ was between 0.061and 0.092,the above four indicators were higher than level 2 of the national standard.Throughout 2015,the concentration of PM_2.5.5 was higher;The concentration of NO₂ exceed the standard 8 months;Except November,the concentration of PM₁₀ was lower than level 2 of the national standard;The concentrations of CO and O₃ were lower than level 1 of the national standard;The concentration of SO₂ was lower than level 2 of the national standard.Thirdly,based on PM _2.5,PM₁₀,SO₂,NO₂,CO,O₃ six pollutant concentration data and three meteorological data of highest temperature,wind speed and wind direction in 2015,the relationship among nine factors was analyzed by SPSS software,and the correlation coefficient above of nine factors with AQI was 0.967,0.982,0.596,0.853,0.849,-0.304,-0.504,0.057,-0179,It is concluded that the correlation between PM₁₀ factor and AQI is the most significant,and the correlation between wind direction factor and AQI is the least significant.The results showed that eight factors of PM_2.5,PM₁₀,SO₂,NO₂,CO,O₃,temperature and wind speed were significantly correlated with AQI,therefore,it is necessary to predict the air quality of Harbin based on the above eight factors.Finally,based on the annual air quality daily and meteorological daily data of Harbin in 2015,this paper establishes the AQI index prediction method of Harbin air quality using BP neural network.Using this method,the AQI index for the period from June to August 2016 and December 2016 to February 2017 is predicted based on 8 factor data per day.Compared with the actual monitoring data,the acceptance rate of training data is 100%,the acceptance rate of the test results is98.9%,it is verified that BP neural network has good generalization ability and high prediction accuracy in the Harbin air quality prediction application after effective training.