Research On The Prediction Of Air Pollution Concentration In Xinyang Based On Combination Model

In recent years,China's air quality is improving,due to the superposition of many high-pollution and high-emission industries,the high proportion of fossil energy consumption,urbanization and other factors,it will be a long way to reach the ideal air quality.While air pollution is harmful to human health,can increase the incidence of respiratory diseases and so on,and affects local economic development by hindering the entry and accelerating the outflow of talent.It's helpful for local management by predicting air quality and warning the peak concentration and occurrence time of air pollutants,the local government can take effective measures to prevent the occurrence of extreme air pollution events,and reduce the impact of air pollution on production and the public.So it is necessary to explore the hourly concentration prediction model with high accuracy and lower use conditions.In this paper,the first step is analyzing the correlation between air pollutants in Xinyang,and finds that the correlation between PM2.5 and SO2 and O3 is weak,and these three kinds of air pollutants are more harmful to health,so the paper selects them for research and prediction.Then,A combined prediction model based on time series and LSTM is proposed.The variation trend of air pollution concentration is decomposed into the long-term trend(annual and monthly trend),medium-term trend(daily trend)and short-term trend(hourly trend),and they are predicted respectivel.According to the correlation between air pollutant concentration and meteorological factors,the paper proposes another combined prediction model based on LSTM and multiple regression model to process air pollutant and weather data on multiple time steps.LSTM model can extract information from time-related data,can process the data with high dimensions,and can fit nonlinear data.LSTM model is used to make a preliminary prediction,and then ridge regression was used to optimize the prediction results based on weather forecast data.Finally,experimental verification of the model is carried out by using the historical concentration data of PM2.5,SO2,O3 and meteorological data of Xinyang every 3 hours from 2017 to 2019.The prediction performance of the two combined models proposed in this paper was compared and evaluated on several dimensions.The combined model based on time series and LSTM is more stable,the average prediction accuracy of this model for PM2.5,SO2 and O3 is 81.19%,74.33% and 73.70%.The average prediction accuracy of the combined model based on LSTM and multiple regression model for PM2.5,SO2 and O3 is 72.39%?77.13%?73.97%.Compared with the comparison model,the two combination models constructed in this paper have effectively improved the prediction accuracy.In general,they are able to predict the time and value of the peak of air pollutants accurately which can provide a more scientific reference for the management decisions of local government to guarantee the production and promote economic development.