Analysis And Prediction Of Daily Air Pollution Index In Nanjing

Air pollution is one of the urgent environmental problems that China needs to solve.Its analysis and prediction is an important part of "winning the battle to win the blue skies" and is a test of the effectiveness of atmospheric environmental governance.It also aims at conducting more accurate and effective prevention and treatment to heavy polluted areas and contamination-prone periods.This article,taking Nanjing City as an example,mainly uses daily air pollution index(API)data to analyze and predict its API.In terms of analysis,firstly,we use wavelet analysis and visual statistics to analyze monthly API changes in the city from 2001 to 2010.Secondly,we make statistical analysis on its seasonly API change.Then,we use fuzzy comprehensive evaluation to judge the air quality in 10 years so as to obtain the yearly API change rule.The Daniel trend test was used to analyze the trend of the monthly,seasonly and annual API.We can find out the API change rule from analysis and we can find out that API varies from Season to season according to its seasonly rule.Therefore,we need to establish spring,summer,autumn and winter forecast model to forecast API.In terms of prediction:based on the BP neural network prediction model,progressive optimization is performed to improve the accuracy of the API in the period of sharp rise or fall.Firstly,Fruit Fly Optimization Algorithm is used to optimize BP network to build a new Model FOA-BP model.Secondly,we decompose the seasonal data of the city by wavelet and then configure singlel-reconstructed subsequences with appropriate models_BP network and FOA-BP network to predicate.We reconstruct the predicated subsequence by weight calculation and the reconstructed sequence needs to predicate API sequence.The research conclusions of the dissertation are as follows:(1)In the 10 years,the probability of API peak value in Nanjing in January,March,May,and December is relatively large,and the low value of API is mainly in July and August.In the winter and spring of each year,the API fluctuates greatly,the air quality is poor,while the summer air quality is the best.In the visual monthly value visual API statistics,the average monthly API values in the months of June,July and August show a decreasing trend,while the average monthly values of September,October and November show an increasing trend.We can draw the conclusion that the regularity of API monthly average value in summer and autunmn is stronger than that in winter and spring.According to the analysis of Daniel’s trend test,the average monthly API value has a downward trend from January to November,with a significant downward trend in January,March,April,August,September,and September.The monthly average of API in December has an upward trend,but the trend has no significant change.(2)The average API value of each season showes the highest API in spring and the lowest API in summer while the fall and winter are between the two.According to the statistics of air quality level days,the number of days without air pollution in autumn is greater than that in winter,and the number of air pollution days is less than winter,and the air quality in autumn is better than that in winter.According to the Daniel trend test on the number of air quality level days,the air pollution status in the four seasons has been improved.The API trend in the spring and summer has dropped significantly.Although the fall and winter APIs have decreased,they have not been significant.(3)After the fuzzy comprehensive evaluation,the inter-annual API changes of Nanjing City in 10 years were indirectly calculated:2002>(higher than)2001>2004>2006>2005>2007>2003>2008>2010>2009.The city’s air quality is moving toward better and better trends and the trends are changing significantly.(4)Four forecasting models were used to predict the four seasons of spring,summer,autumn and winter in Nanjing.The forecasted mean square error was wavelet FOA-BP network model<wavelet-BP network model<FOA network model<BP neural network model.The prediction accuracy of the four seasons are wavelet-FOA-BP network model>wavelet-BP network model>FOA network model>BP neural network model.The wavelet-FOA-BP network is most suitable for the API prediction of Nanjing City’s four seasons.The prediction accuracy rates are:70.96%,93.9%,95.4%,and 91.8%respectively,and the prediction accuracy during the period of sudden increase or decrease in API is also high.