| Air pollution is one of the most concerned problems.With the increasingly complex causes of air pollutants and the increasing types of air pollutants,how to comprehensively use the massive monitoring data to analyze and predict air quality is a key problem in many research fields.Beijing Tianjin Hebei area is the political and cultural center of China,the important core area of northern China’s economy,and also the area with the most prominent air pollution status and the most urgent demand for prevention.It is urgent to predict the concentration of air pollutants,then prevent the occurrence of serious pollution events,and carry out effective air quality control.Based on the air pollutant concentration data,meteorological data and emission inventory data from 2015 to 2018 in Beijing Tianjin Hebei region,this paper firstly analyzes the spatial and temporal characteristics of air pollutants in this region.In the spatial dimension,the spatial autocorrelation and Kriging interpolation are used to analyze the distribution characteristics and spatial aggregation of PM2.5 emissions and concentrations in the cities of the region.In the time dimension,wavelet analysis is used to decompose and reconstruct the AQI time series of each city,and the annual,seasonal and monthly change characteristics of each city are studied.Through the multi-scale analysis of wavelet variance,it is found that there are 30 d and 250 D shock periods in each city in this area.In order to comprehensively consider the influence of time and space on air quality prediction,this paper designs an integrated deconv LSTM air pollution prediction model,which combines deconvolution network and short-term memory network as a weak classifier,and uses Ada Boost integration algorithm to integrate the deconv LSTM air pollution prediction model Now we make full use of big data,the bottom deconvolution neural network is used to extract the feature association of input data in the spatial dimension;the top long-term memory network is used to extract the feature association of input data in the time dimension,so as to achieve the spatiotemporal fusion prediction of air quality data.Finally,the integrated deconv LSTM air pollution prediction model proposed in this paper is verified by experimental analysis.Compared with traditional prediction models and algorithms such as LSTM,ARIMA,BP neural network and SVM support vector machine,it has higher accuracy and better performance. |
