Research On Air Quality Prediction Based On Space-Time Mixed Model

With the rapid development of urbanization in China,urban population,traffic scale and energy consumption are expanding,resulting in a large increase in air pollutants such as inhalable particles,carbon monoxide,sulfur dioxide,nitrogen oxides and so on,causing serious air pollution.Concentration of Haze and PM2.5 exceeding the standard is an obvious feature.Air pollution seriously damages the ecological environment,having an affect not only on people's daily life,physical and mental health,but also on the city's environment for investment,talent attraction and economic development.Reducing air pollution is of great significance to the sustainable development of cities.It has become a problem to be solved urgently by various governments,and has attracted the attention of large institutions and large enterprises,and thus having become a hot spot of research by researchers and scholars.Facing the above challenges,this paper studying the urban air quality prediction method based on space-time mixed model,uses various data source such as urban air quality,meteorology and weather forecast,considers the various factors that affect the urban air quality and combines the spatial-temporal correlation of air pollution,establishing a multi-view spatial-temporal mixed model.The specific research contents are as follows: 1.Study the temporal dependence of air quality on air pollutants and weather conditions in the target city,and use multiple linear regression method to establish the air quality prediction model based on time series correlation of meteorological data and air quality historical data.Air pollutants include air pollutant concentration data collected by air quality monitoring stations in target cities,and weather conditions include meteorological data and weather forecast data recorded by meteorological stations in target cities.Air pollutants include air pollutant concentration data collected by air quality monitoring stations in target cities,and weather conditions include meteorological data and weather forecast data recorded by meteorological stations in target cities.2.Study the influence of the surrounding environment of the target city on the air quality of the city.In a word,study the spatial correlation between air pollutants and weather conditions in surrounding cities and air quality in target cities,and use artificial neural network method to establish an air quality prediction model based on spatial correlation between meteorological data and air quality historical data.The air pollutant dataincludes the air pollutant concentration collected by the surrounding urban air quality monitoring stations,and the meteorological data includes the meteorological data and weather forecast data recorded by the surrounding urban meteorological stations;3.Based on the monthly and sectional average of air quality over the years,use the aggregation method to integrate the foregoing two independent air quality prediction results,forming the final spatio-temporal mixed air quality prediction mode;4.Finally,use the air quality data and meteorological data of Beijing and Shenzhen to verify the model proposed in this paper.The experimental results show that the model proposed in this paper is better than the previous methods in air quality prediction accuracy.There are three innovations in this paper:1.In the establishment of air quality prediction model based on time-series spatial relationship,this paper considers the correlation between air quality historical data in a wider range and a longer time from a new perspective.Specifically,this paper proposes the concept of the sectional mean of air pollutant concentration,and adds the monthly mean and sectional mean of air pollutant concentration to the multivariate regression model as a related factor.Existing time-series-based air quality forecasting methods usually use local data,from the past few hours,predicting future air quality.The proposed method in this paper utilizes data from a period of time in the past month(sectional average)and monthly average data from the past few years,considering not only local factors of air quality change but also the middle and long term trend of air change.Therefore,this method is more conducive to improving the accuracy of air quality prediction.2.In establishing the air quality prediction model based on spatial relations,we consider the spatial and temporal asynchronism of multi-sensor(air quality monitoring station)data in a large area to solve the time alignment problem of asynchronous data,then train the model.The data asynchronization has a great impact on the accuracy of the model prediction.One obvious problem is that the data of air pollution far from the target city do not affect the air quality of the target city at t+1 time(i.e.the time of t delay of one hour).For example,if Zhangjiakou is250 km away from Beijing and its wind speed is 25 km per hour,the pollutants in Zhangjiakou will reach Beijing in 10 hours.Therefore,when predicting the air quality of Beijing at a certain time,the pollutant concentration at time T-10 in Zhangjiakou can be regarded as the data related to the air quality of Beijing.In other words,use the data with a certain amount of time advance as the input of the neural network prediction model to reflect the spatial correlation of air pollutants.3.In this pager proposes a new model aggregation method,which aggregates the results oftime series multiple regression prediction model,artificial neural network prediction model,monthly average and sectional average of air quality over the years into a new prediction result to form the final spatio-temporal mixed air quality prediction model.The contributions of this paper are as follows: 1.in constructing the prediction model based on time series multiple linear regression,extend the application scope of historical data.The original method used the past few hours data to predict the future air quality,this pager proposes segmental averaging values(the average data from the beginning of the month to the current time)and the monthly average(the monthly mean value of the past few years).2.In the establishment of air quality prediction model based on spatial relationship,this paper found a group of data with better spatial correlation as the input of neural network model.In a word,first solve the time alignment problem of asynchronous data,and then train the model.3.This pager proposes a new model aggregation method.The original aggregation method is to select different training models according to different conditions for prediction.The method in this paper is to aggregate the results of time series multiple regression prediction model,artificial neural network prediction model,monthly average and sectional average of air quality over the years into a new prediction result,forming the final time and space mixed air quality prediction model.These are the innovations of this article.