Research On Air Quality Analysis And Forecast Methods Based On Multimodality

With the accelerated development of industrialization and urbanization in China,the rapid expansion of urban population,energy and transit construction scale,unreasonable industry structure,intensive energy consumption and low utilization ratio lead to continuous sharp decline in air quality.Atmospheric pollution is increasingly serious,which is mainly caused by inhalable particle,sulfur dioxide and nitrogen oxides and so on.Deteriorating air quality not only threats the public health,but also affects sustainable development of national economy.A comprehensive,scientific and accurate air quality analysis and forecast has great theoretical and practical value in conducting the public effectively avoiding health damage caused by air pollution,environmental government department strengthening pollution source supervision and improving emergency capacity in heavy pollution days.On the basis of understanding the factors influencing urban air quality and establishing air quality analysis model,the paper deeply studies the methods like Grey Theory,Neural Network,Markov Chain and Conditional Random Field,establishes various urban air quality forecast models according to different forecast terms,improves the forecast accuracy of air quality and conducts systematical analysis and experimental study on urban air quality change trend through the analysis on urban air quality monitoring data.First of all,aiming at the relatively complex problems of air quality interactions and associations between region and region,the spatio-temporal attribute of air quality influence factors and the spatio-temporal relations of the air quality characteristics are analyzed.A method using a weighted-directed graph to establish the air quality model and characterization methods is proposed.These will serve as a theoretical underpinning for air quality forecast and analysis.Secondly,for the lack of information abonut influence factors like meteorology,socio-economic and so on in air quality forecast,an improved GM(1,1)model is designed.On the basis of analyzing the influence of raw sequence interference and background value change on forecast accuracy of the model during the establishment of the model,new buffer operator and background value are defined to improve GM(1,1)model at the same time.Improved GM(1,1)model can weaken model randomness and external interference,decrease hysteresis error in the model fitting,which improves the stability and forecast accuracy of GM(1,1)model in air quality forecast.Improved GM(1,1)model solves the problem of urban air quality forecast only using the air pollutant concentration data under the condition of deficient information.Thirdly,to solve the problem that single grey model has a worse fitting effect on the data of larger volailty in air quality forecast of the condition of deficient information,two improved grey combination forecast models are proposed.Improved grey neural network model and improved grey markov model are established by combining GM(1,1)model respectively with neural network model and markov chain model and improving the combination.According to the features of traditional grey neural network model,improved grey neural network model makes improvement respectively in the background value of grey model and the form of training sample data of neural network.On the basis of analyzing traditional grey markov model,improved grey markov model establishes dynamic transition probability matrix by replacing traditional GM(1,1)model with metabolic GM(1,1)model and defines new buffer operators.The application of two improved grey combination forecast models in urban air quality forecast can improve the reliability and accuracy of air pollutant concentration forecast results and avoid the limitation of single model.Two improved grey combination forecast models solve the problem of deficient information air quality forecast under the condition of the larger volailty data.Fourthly,in addition to meteorological factors,socio-economic factors also exert influence on air quality.Therefore,a collaborative forecast model is designed,which combines GM(1,N)with artificial neural network model based on the analysis of the influence of factors on air quality.According to the characteristics of two kinds of influence factors,meteorological factors with nonlinear characteristic are taken as the input of artificial neural network to forecast air pollutant concentrations and socio-economic factors with incomplete information filtered through grey correlation model are taken as the input of GM(1,N)to forecast air pollutant concentration.On the basis of using Pearson correlation coefficient to respectively determine the influence of meteorological factors and socio-economic factors on urban air quality,the forecast results of two models are assigned weights and sum,thus the final forecast result is gained.The forecast accuracy of collaborative forecast model is higher than that of any single model and that with high forecast accuracy,effectively improve the prediction accuracy and comprehensiveness,solves the problem of both considering the effects of meteorological and socio-economic factors of air quality forecast.Finally,with the demand of air quality level real-time forecast,real-time forecast analysis on urban air quality level is proposed by using Conditional Random Field model on the basis of analyzing the influence of meteorological factors on air quality.According to the features of metrological factors,a new feature template is defined during the forecast modeling in order to improve the model forecast accuracy.Conditional Random Field model has high forecast accuracy in forecasting air quality level and that as a new effective forecast method and solves the problem of real-time forecast of air quality level.