Neural Network Model For PM2.5 Concentration Prediction Based On C-SVM-RFE

There are many factors that affect the concentration of PM2.5,resulting in the expression of PM2.5 concentration is complex and changeable.How to analyze the PM2.5concentration and its related influencing factors,find the internal laws,and then realize the prediction of PM2.5 concentration.The prediction of PM2.5 and PM2.5 is of great significance to improve people’s health and development.A neural network model of PM2.5 concentration prediction based on relevance recurrent SVM is proposed by taking the pollutant concentration and meteorological data of Beijing as the research object.The method fully considers the dynamic correlation of air quality evolution characteristics,so as to improve the accuracy and stability of PM2.5concentration prediction.The main research work is as follows.Firstly,in order to solve the problem of sparse pollutant data,based on the correlation between other pollutants and meteorological data and the principle of "the first law of geography",an interpolation method based on spatial transformation is proposed.Aiming at the data set with missing values,the method of region division is used to divide the data,and the rules between each region are mined.Finally,the inverse distance weighted method(IDW)is used to interpolate the missing value region combined with the data of other regions.Secondly,based on the influence of meteorological factors and other pollutant concentrations on PM2.5 concentration changes,the correlation is introduced into SVMRFE,and the support vector machine recursive feature elimination(C-SVM-RFE)method based on correlation is constructed.The support vector machine feature elimination method(SVM-RFE)is improved,The key features are easy to be deleted by mistake.Thirdly,the FCM method was used to cluster the features,and the long-term and shortterm memory artificial neural network prediction model was established to predict the PM2.5 concentration.Dropout mechanism is introduced to solve the problem of LSTM over fitting.Finally,the real meteorological data and air pollutant concentration data of 12 monitoring stations in Beijing from March 1,2013 to February 28,2017 are used in the experiment.By comparing with the results of other PM2.5 concentration prediction models,the accuracy of the proposed model in interpolation and prediction of PM2.5 concentration is proved,and the effectiveness of the proposed method is verified.