Research On Air Pollution Prediction Based On Hybird Model Of Spatio-temporal Features

In recent years,with the continuous deepening of urbanization and industrialization in my country,air pollution has become more and more serious,and the problem of air pollution in our country has received extensive attention.Compared with the”Air Quality Guidelines”issued by the World Health Organization,our country still needs to work hard in air pollution control.In order to minimize the short-term and long-term effects of air pollution on people’s health,it is important to quickly and accurately forecast pollutants such as PM_2.5,PM₁₀,CO and so on.Significance.At present,the numerical prediction models at home and abroad cannot meet the current needs of people in terms of com-putational efficiency and accuracy.Most of the published artificial intelligence forecast models are time-series forecasts for the overall air quality of the city and the observation data of a single monitoring station,ignoring the correlation factors between the surround-ing monitoring stations and the local monitoring stations,which affects the accuracy of the forecast.In this thesis,on the basis of time series prediction,and considering the influence of surrounding monitoring stations on local monitoring stations,a deep learning model-based PM_2.5concentration prediction method considering spatiotemporal characteristics is proposed.Its main contents are as follows:（1）The data of air pollutants and their related air pollutants are all time series,and the time series analysis of the data set can be carried out using the method of statistical test quantity.Firstly,the data features are selected for PM_2.5and its related objects,and then the time series linear analysis is carried out for the strongly correlated data features,and it is concluded that PM_2.5and its related objects are nonlinear Time series,setting the direction for model design.Then,two filling algorithms are proposed for the frequently-occurring missing values of time series,and relevant nonlinear models are designed according to the nonlinearity of time series to verify the filling effect of missing values.（2）It is more effective to use data from monitoring stations in urban areas to fore-cast air quality in various areas.However,air pollutants will change with factors such as wind speed and geographical topography.In view of the problem that the influence of geography has not been deeply considered in previous research,a PM_2.5based on LSTM（Long Short Term Memory）neural network is built.Hybrid prediction model.The model consists of a time prediction model,a spatial prediction model,and an aggregation model.The time prediction model uses the historical data of local monitoring stations to make predictions as temporal features,and the spatial prediction model uses the historical data of surrounding monitoring stations to make predictions as spatial features.The spatial features are mixed and selected to obtain the final result.（3）Use the data of air quality monitoring stations in the Beijing-Tianjin-Hebei region to train the air pollution prediction model,and design different experiments to verify the effectiveness of the model in this thesis.First,the air pollution prediction model is com-pared with other deep learning methods using the same data set to predict the effect.The results show that the model in this thesis is better than other deep learning methods.Sec-ondly,the effect of the air pollution prediction model is compared with the model using only air quality data and the model using air quality data and meteorological data.The results show that the spatial characteristics of the model in this thesis are effective.