Prediction Of Urban Air Quality Based On Machine Learning Technology

In recent years,with the steady and rapid development of China's economy,urban and rural industrialization and the living standards of residents have increased substantially.The scale of industry and transportation has continued to expand,the construction of factories has increased,and the number of motor vehicles has increased year by year.This has also led to an increase in fossil energy consumption,and increased emissions of automobile exhaust,industrial emissions,construction dust,and waste incineration.The problem of air pollution is becoming more and more serious,and it has gradually become a major livelihood issue which the people are take care of.For many current air pollutant concentration prediction models,mainly for time series feature modeling and prediction,spatial information is only used as general feature data.In this,the convolutional neural network is used to mine the potential spatial relationship between the predicted site and its neighbors.A predictive model of spatial correlation to predict PM2.5 concentration.On the other hand,for the one-sidedness of modeling from the perspective of time series information,this research proposes a time model based on gradient lifting framework and a spatial model of neural network framework through Stacking integration model to construct a predictive model of integrated spatiotemporal information,which is flexible and effective.The main research contents of this included:1)Constructing training data sets and test sets based on historical data of air quality and meteorological historical data.The training subsets of different time scales are sampled at different time intervals,and then the gradient boosting models were respectively trained on each subset.The result of time series prediction model is the weighted sum of each sub-model prediction result.2)The data of the adjacent air quality monitoring station were aggregated into the spatial neighborhood of the station to be predicted by spatial interpolation method,and the neighborhood matrix data set was constructed.Then the convolutional neural network regression model is constructed as the training data.3)By using the fully connected neural network as the secondary learner algorithm,this research constructed a stacking-based integration model,and aggregated the results of the time series gradient boosting model and the convolutional network model.4)Finally,compared the prediction performance of each model on the test set and verified the effectiveness of the stacking model.