Prediction Of PM_2.5 In Tianjin Based On Data Mining

Since the 21st century,a series of environmental problems have been brought by rapid industrialization and urbanization in China.The consumption of mineral energy,population growth,traffic expansion and other human activities have produced a lot of pollutants and discharged into the air.In recent years,PM_2.5,one of the major air pollutants in China,gradually attracted people's attention.On the side of residents,the results of air pollution forecasting can be used as an important reference for daily activities.So that it can reduce the harm of pollutants on the human body.On the side of relevant government departments,they can develop accurate control measures timely according to the results.This is important for environmental management,urban public utilities and the development of urban economy.The innovative points of this study are to discuss the relationship between meteorological conditions on the previous day and PM_2.5 concentration,to mine the relationship between meteorological data and PM_2.5 using association rules,to establish forecasting models in different seasons and to use association rules as one of the input vectors of genetic neural network.Firstly,the influencing factors of different seasons and PM_2.5 were analyzed by regression.And the influencing factors with higher correlations in different seasons were obtained.Then,Apriori algorithm was used to mine association rules between different influencing factors and PM_2.5 in different seasons.Finally,the association rules are transformed to obtain the correlation vector.The trial-and-error method is used to determine the number of hidden layer nodes in the neural network.The meteorological data of the day before the forecast day is used as the input vector for training and prediction.The predictive models are established in different seasons.The joint model of genetic neural network-association rules is established respectively in different seasons.Using the meteorological conditions monitored on the day before as an input vector and establishing models by sub-seasons can improve the accuracy of prediction.