Research On Air Quality Prediction Algorithm Based On Multi-source Spatial-temporal Data Fusion

Air quality is a major social issue affecting people's livelihood.President Xi has repeatedly stressed the importance to improve the level of environmental governance,strengthen the control of air pollution,and resolutely win the blue sky defense war.Air quality monitoring stations as an infrastructure for environmental monitoring in cities,which are distributed in many areas of the city,and a large amount of air monitoring data is collected every day.These data provide a data foundation for mining data linkages and establishing scientific air quality prediction models.Therefore,this thesis applies data fusion,data mining,machine learning and other methods and techniques to construct an air quality prediction model,and selects multi-source data to predict air quality.The main contents are as follows:(1)Air Quality Prediction Method based on GTWRThe air monitoring data has sequence characteristics in the presentation mode,and there are temporal and spatial characteristics.In order to make full use of the spatial and temporal correlation between data,an air quality prediction method based on geographical and temporal weighted regression is presented.First,for the data missing problem,the combination of linear interpolation and neural network prediction is used to recover the missing data.Secondly,for the multi-source characteristics of data,the spatial and temporal characteristics are used to fuse the air monitoring data and the meteorological data,and the temporal information is discretized to enhance the correlation of the data in time.Finally,for the problem of high data dimension and collinearity between variables,the principal component analysis method is used to select the optimal variables,and the geographical and temporal weighted regression model is used for analysis and prediction.Experiments show that the proposed method is superior to the ordinary linear regression model and the traditional geographically weighted regression model in accuracy.(2)Air Quality Prediction Method based on Fast LSTMThe composition of atmospheric pollutants is complex and diverse,longer dependencies between data are prone to long-term dependency problems during training process,which affects the prediction accuracy of the model.To this end,an air quality prediction method based on fast long short-term memory networks is presented.First,the K-means algorithm is selected to cluster multi-source data and add classification labels.Secondly,the long short-term memory network is used to solve the long-term dependence problem caused by the long sequence data,and the gradient disappearance and gradient explosion are effectively avoided.Finally,for the problem of slow training of long short-term memory network models,simple recurrent unit is introduced to simplify state calculation and achieve faster state update.Experiments show that this method greatly reduces the training time while ensuring the prediction accuracy.(3)Design and Implementation of Air Quality Prediction Prototype System based on Multi-source DataIn order to verify the research results of this subject and strengthen the combination of theory and practice,a prototype system for air quality prediction based on multi-source data fusion was designed and implemented.The system includes three modules: algorithm selection,model training and model prediction.The model is trained by applying historical data from different cities and different periods.Users can select different models and different time spans to predict the air quality in the future.The system has important practical value for urban life,intelligent transportation,and environmental protection.