Prediction Study Of Regional Air Pollutant Concentrations Based On Deep Learning

Air pollution is one of the most important factors affecting people’s urban life,and the increase of its concentration will aggravate the threat to people’s health.Therefore,accurate prediction of atmospheric pollutant concentration is of great significance to human health protection.At the same time,the prediction of air pollutant concentration is always considered to be the key problem of air pollution warning and control management.The prediction of atmospheric pollutant concentration is an effective method to prevent pollution events by providing early warning for harmful substances in the atmosphere.However,due to the dynamic nature of atmospheric flow,the task of pollutant concentration prediction has great uncertainty and instability,which makes it difficult for traditional models to effectively predict the change trend of regional pollutant concentration.At present,with the popularity of urban space-time big data and easy access to big data of pollutants and meteorology,novel deep learning technology provides an effective solution for complex pollutant concentration prediction.At present,according to the characteristics of the research methods of pollutant concentration prediction,it can be divided into two categories: the prediction method based on numerical model and the prediction method based on data-driven model.The prediction method based on numerical model is to simulate the emission,diffusion and disappearance of atmospheric pollutants according to meteorological principles and statistical methods,so as to achieve the prediction of pollutant concentration.However,due to the uncertainties of pollution sources,meteorological conditions and transformation process,as well as the high complexity and large amount of computation of numerical models,it is difficult for the prediction method based on numerical models to be applied to the field of regional pollutant concentration.The prediction method based on data-driven model is to predict the pollutant concentration by learning and analyzing the historical data of pollutants.However,the spatiotemporal relationship between various complex and non-stationary air pollutants and meteorological data needs to be modeled for air pollution prediction.The simple network model structure of machine learning in data-driven models may limit the prediction accuracy.At present,deep learning method is emerging in the field of air pollutant concentration prediction.The deep learning model can obtain better robustness by processing complex spatiotemporal data with deeper hiding layer and excellent self-learning ability.According to the research,deep learning method has achieved remarkable results in the field of pollutant concentration prediction,but it still faces some problems:(1)In the field of pollutant concentration prediction,the prediction model based on CNN has limited ability to extract spatial features between pollutant meteorological data;(2)In addition,due to the large amount of pollutants and meteorological data,problems such as feature extraction difficulties and gradient disappearance or explosion may occur in the prediction model.In view of the above deficiencies,this paper is devoted to combining regional pollutants and meteorological historical data to study the prediction of regional air pollutant concentration from two dimensions of time and space.Firstly,this paper constructs CBAM-CNN-Bi LSTM model on the basis of CNN-LSTM model.CBAM-CNN-Bi LSTM provides model basis for regional pollutant concentration prediction.In order to solve the problem of feature degradation of convolutional neural network,Bi LSTM cannot extract high-dimensional spatiotemporal features.A regional pollutant concentration prediction model CBAM-Res Conv LSTM is also constructed.The main work of this paper is as follows:(1)In the CBAM-CNN-Bi LSTM model,the structure of the model is first introduced;Secondly,the parameter setting of the model is introduced.Finally,the training process of the model is described.CBAM-CNN-Bi LSTM makes full use of the convolutional attention module to assign weight to the original input from the spatial dimension,so as to enhance the essential information of the data.CNN is responsible for extracting spatial features.Bi LSTM acts as the output layer to make the final prediction of pollutant concentration.(2)In the CBAM-Res Conv LSTM model,the structure of the model is firstly introduced;Secondly,the parameter setting of the model is introduced.Finally,the training process of the model is described.With the assistance of the convolutional attention module,the residual neural network strengthens the feature propagation and avoids the problems of gradient explosion or network degradation when dealing with the big data of pollutants and weather.Finally,Conv LSTM receives the time series with high dimensional spatial characteristics from the residual network,extracts the temporal and spatial characteristics of pollutants and meteorological data,and makes predictions of pollutant concentrations.(3)Data preprocessing,correlation and spatiotemporal analysis.In this paper,pollutant and meteorological data are filled with missing values and data standardization.After data preprocessing,the correlation and spatiotemporal dimension of pollutants and meteorological data are analyzed to provide the basis for data analysis of the prediction model.(4)Simulation experiments show that the prediction performance of CBAM-CNN-Bi LSTM and CBAM-Res Conv LSTM models based on deep learning is superior to other classical models and has a good application prospect in regional air pollutants.At the same time,the experiment also proves that CBAM-Res Conv LSTM is more suitable for regional air pollutant concentration prediction than CBAM-CNN-Bi LSTM.