Regional Air Pollutant Concentration Prediction Based On Deep Learning

Air pollution not only seriously affects people’s health,but also becomes one of the important reasons restricting urban development.It can accurately and effectively predict air pollutants,which plays a very important role in pollution prevention and control.In the field of air pollution prediction,the method of predicting air pollution by simulating the physical and chemical processes of the atmosphere is not only time-consuming,but also very difficult to collect reliable information about the source of pollutants and the physical and chemical characteristics of the atmosphere.With the development of artificial neural network,deep learning technology provides a new idea for the prediction of air pollutants.However,in the current research,there are still weak links in the research on the processing of air pollutant concentration data and the correlation between various pollutants.The paper focuses on two air pollutant concentration prediction models based on deep learning,and the main work is as follows:1.Aiming at the problem of single-factor air pollutant concentration prediction,the CEEMDAN-PE-GRU air pollutant prediction model based on time series decomposition is established.Firstly,the original sequence is decomposed into a group of intrinsic mode functions(IMF)and a residual component(RES)with different frequencies and complexity by using the adaptive decomposition ability of CEEMDAN decomposition algorithm for nonlinear signals.Secondly,IMF components with similar complexity and residual components are recombined according to PE algorithm.Finally,the recombined subsequences are predicted using GRU model respectively,The final prediction result is obtained by adding the prediction results of subsequences.The experimental results show that the prediction error based on CEEMDAN-PE-GRU model is significantly lower than other models.2.A STL-ConvLSTM-GRU air pollutant concentration prediction model integrating time series decomposition and spatiotemporal feature extraction is constructed to solve the problem that the air pollutant concentration cannot be accurately predicted under the influence of multiple factors.Firstly,the original sequence is decomposed by STL algorithm,and the decomposed sequence is fused with other factors respectively;Build the ConvLSTM-GRU model,and use Bayesian optimization algorithm to optimize the parameters;The fused data is imported into the ConvLSTM network for spatiotemporal feature extraction,and then the extracted feature sequence is imported into the GRU network for prediction.Compared with the prediction results of other models,it is proved that the proposed model has the characteristics of small error and good prediction effect.