Research On Air Quality Prediction Model

With the improvement of living standards,people pay more and more attention to environmental pollution which can affect their own health and the health of the next generation.Among them,air pollution is the closest to people's lives.Among all kinds of air pollutants,PM2.5 is the most concerned pollutant in recent years,people have done a lot of research around it.However,due to its complex causes,especially in cities,in the case of a combination of various conditions,it is difficult to accurately predict the concentration of PM2.5.There are three main problems in the existing research: First,the traditional numerical prediction methods have bottlenecks.Compared with statistical prediction methods,the accuracy of prediction is lower;Second,the existing statistical prediction methods tend to be too singular in feature selection,and the number of factors considered is too small.The resulting prediction model can be applied with too many scene constraints and cannot be predicted in the spatial dimension;Third,the existing research does not consider the planning problem of the monitoring site.At present,the selection of the location of the monitoring site has a strong human subjectivity,which leads to the unreasonable distribution of the air quality monitoring site.In response to the above problems,this paper has carried out the following research:First,using the statistical prediction method,using the improved boosting tree model,the XGBoost model,modeling in the time dimension,and combining air quality data with meteorological condition data,using feature importance to filter features.n addition,the method of adding tags by misplacement expands the data,and the data set is expanded to more than ten times the original scale,making the predicted model more accurate.The accuracy of the model was verified by experiments.Secondly,using the idea of spatial interpolation to model the PM2.5 concentration in the spatial dimension,extract the information of the data in the spatial dimension by latitude and longitude coordinate transformation,and set the importance of the spatial dimension information as the necessary attribute to ensure the strict modeling process.Based on site space coordinate information.Combining the prediction model on the spatial dimension with the prediction model on the time dimension,a time-space prediction model of PM2.5 concentration is constructed,which can predict all monitoring within 24 hours when the current air quality information and meteorological condition information are known.The PM2.5 concentration of any latitude and longitude in the rectangular area formed by the upper and lower latitude and longitude points.Finally,using the conclusions of the above research,this paper constructs regional gridded PM2.5 concentration data,and uses these data to study the planning of air quality monitoring sites.Firstly,a site distribution evaluation criterion is proposed,which is the regional absolute deviation control rate.Then,based on the K-means clustering algorithm,a fixed site location planning method based on quadratic clustering is proposed.The shortcoming further proposes a dynamic site planning method.Finally,the advantages and disadvantages between the sites obtained by the two methods and the real sites are compared through experiments.