Research On PM_2.5 Concentration Prediction And Monitoring Network Optimization Of Urban Air Quality

In recent years,with the apid development of China’s economy,air pollution has become more and more serious. Fine paticulate matter (PM_2.5),as one of the main pollutants in the air,contains a large number of toxic and harmful substances.Due to its small particle size,it is easy to enter the human body and harm human health. In addition,it not only stays in the atmosphere fora long time,but also has a long transport distance,which can promnote the formati on of haze. This wll reduce the visibility of the air and seriously affect the air quality and atmospheric environment. Although most cities in China have their own air quality monitoring networks,they can only provide real-time information on pollutants. Therefore, timely and accurate prediction of current and future PM_2.5 concentrations can provide a basis for the control,management and early warming of air pollution,as well as help guide people’s travel and protect human health. With the continuous.development of the city scale,the number of monitoring stations in the urban air quality monitoning network is insufficient,so it is crucial to optimize the monit oring network.Aiming at the problems of PM_2.5 concentration prediction and monitoring network optimization,the work of this paper is as follows.First,air quality and meteorological dlata of Taiyuan are collected and prepiocessed. After data preprocessing,data mining and statistical methods are used to analyze the factors affecting PM_2.5 concenttation. Since PM_2.5 is produced and discharged along with the emision of other pollutants in the air,air polutants.can also be converted to each other to a certain extent. Therefore,there is a corelation between PM_2.5 concentration and the concentration of other pollutants.By visualizing the change of PM_2.5 concentration in different time scales,the influence of temporal features on PM_2.5 concentration is summarized.Second,in tems of PM_2.5 concentration prediction,this paper proposes an AC-LSTM （Attention-based CNN-LSTM） model based on deep learning method to predict PM_2.5 concentration in the next 24 hours. Correlation coefficient and autocorrelation function are used to analyze the spatiotemporal correlation of PM_2.5 concentration. The analysis results show that PM_2.5 concentration in the past will affect current and future PM_2.5 concentration,andthe predicted regional PM_2.5 concentration is closely related to the PM_2.5 concentration of surrounding stations.Therefore,the pollutant concentrations,meteorological features and PM_2.5 concentrations of a djacent monitoring stations are used as input of AC-LSTM model,to construct the multiscale predictors. The AC-LSTM model can process the temporal data related to air quality and effectively simulate the spatiotemporal correlation of PM_2.5 concentration. The paper also adds a network layer based on attention mechanisminto the prediction model,which can measure the importance of the influence of feature states in the past on PM_2.5 concentration in the future.Compared with six machine leaming methods,AC-LSTM model has achieved high prediction accuracy in predicting PM_2.5 concentrations at diferent times.Finally,this paper uses pincipal component analysis and hierarchical dlustering to analyze the changes of PM_2.5,PM₁₀ and SO₂ concentrations in air quality monitoning network of Taiyuan,revealing the similarity of pollutant concentration changes in the monitoring network The results show that there is redundant monitorning equipment in the monitoring network Combining with the geographical environment and industrial distribution of Taiyuan city,this paper puts forward an optimization plan. In addition,in view of the curent Taiyuan city monitoring network of other problems,the corresponding solutions are put forward.