Research On Air Quality Classification Forecasting Hybrid Model Based On Position Data

With the increasingly serious air pollution problem in China,citizens are paying more and more attention to air quality.The air quality in the future and the unknown area can be predicted to provide suggestions for people's life.Ordinary people in the city are concerned with air quality level rather than specific pollutant concentrations.So,the prediction of air quality level has a certain practical significance.General meteorological methods such as atmospheric prediction models,pollution dififusion models,are often complex and computationally expensive.Traditional machine learning methods can avoid these problems,but the data source and model structure are relatively singular.It deserves further investigation on how to use the large amount of data generated by the rapid development of the Internet in the city,and deepen the understanding of the air quality prediction project to extract the feature and build model,so as to improve the effect of air quality prediction.This dissertation proposes a mixed air quality classification prediction model based on position data,the main research works are as follows:1)Study the correlations between air quality and meteorological features,temporal characteristics,POI characteristics,and positional data.2)For the future air quality level prediction,temporally we use SVM algorithm,spatially we segment the surrounding area and then integrate them to generate features,use ANN algorithm to classify,finally integrate the two prediction results through decision tree.3)For the air quality level prediction of unknown space,temporally we use the CRF algorithm,spatially we use KNN to select several similar known area and calculate the correlation coefficients between them as the feature,then use the GBDT algorithm for classification,and finally integrate them with decision tree.4)Through a series of experiments,we study the influence of the granularity of the position hot data,or simply position data on the forecasting effect,compare the performance of the hybrid classification model with the regression model,the single model and other traditional models.The results show that our model has certain improvements.