Air Quality Forecasting Using BP Neural Network And Random Forest Model

Great wealth is created for mankind with the development of urban and industrial civilization,billions of tons of waste gas is discharged into the atmosphere at the same time.When the harmful gases in the atmosphere reach a certain concentration,it will have a great impact on the physical and mental health of human beings.However,in recent years,with the continuous expansion of urban scale,the acceleration of the process of industrialization and the continuous increase of the population,air quality problems occur frequently in China,which has aroused widespread concern.The desire for a better understanding of the future air quality is more and more strong.Based on this background,by taking main pollutants in a city as research objects,spatial influence factors and time factors influence on air quality as research direction.Prediction of pollutant concentration of an air quality monitoring station in the next one hour,using a data-driven method is the mainly studied.With the help of machine learning,in depth,systematic analysis and Experimental Research on the existing air quality data.In this paper,the main works completed are as follows:Firstly,feature extraction,including spatial correlation factors(such as interest point data and road network data)and temporally-related features(e.g.,temperature,humidity),it is helpful to study and solve the problem of air quality prediction.Secondly,aiming at the problem of interest point data and the feature dimension of the road network,this paper through calculates the Pearson correlation coefficient to reduce the latitude of the algorithm input,thereby reducing the cost of model training.Then,according to the spatial and temporal characteristics,the spatial prediction model based on BP neural network and temporally prediction model based on random forest are established respectively.Aiming at the problem that the BP neural network is prone to over fitting,a method to dynamically change the structure of the network model is proposed to improve the generalization ability of the algorithm.Besides,the parallel neural network algorithm and random forest algorithm based on Spark are implemented.Finally,prediction of hourly concentrations of various pollutants in each monitoring site by spatial-temporal model and then this paper evaluated the model with extensive experiment,the results show that the prediction accuracy of this model is higher than that of traditional neural network algorithm and random forest algorithm.