Prediction Of Near-surface Ozone Concentration In Beijing Based On Stacking Model

In recent years,China’s government has made precise efforts on air pollution,and the effectiveness of treatment is obvious to all,2013-2020 China’s air quality has been greatly improved,the annual average concentration of PM2.5 in 2020 is only 33μg/m~3,reaching the national standard attainment line for the first time,at the same time,ozone concentration is gradually increasing,showing a rising trend,located in the troposphere of near-ground ozone as the main air pollutant as a major air pollutant in the troposphere,near-ground ozone is extremely harmful to human body,and the high concentration of ozone pollution poses a serious threat to the health of Chinese residents and economic development.Therefore,it is necessary to establish a high-precision ozone concentration prediction model to provide advance warning for relevant departments’decision making and people’s life and work.In this paper,we select hour-by-hour ozone concentration,pollutants and meteorologi cal data in Beijing from 2020 to 2021,and propose an ozone pollution Stacking fusion early warning model to achieve high precision pollution warning.Firstly,we use random forest to fill in the missing values and 3σadjust the anomalous values to complete the data pre-processing;we combine the random forest variables’importance ranking and entropy method for feature selection,and keep seven variables,including temperature,dew point temperature,barometric pressure,wind direction,cloudiness,rainfall within 6h and Vocs,to build the ozone concentration prediction index system;then we analyze the temporal variation pattern of ozone concentration from monthly,weekly and hourly aspects,analyze the correlation between variables using Pearson correlation coefficient;input the data set into KNN,support vector machine,GBDT,XGBoost,and LightGBM for modeling,combine grid search and 5-fold cross-validation,establish Stacking fusion model,and evaluate the model based on MSE,RMSE,and MAE;finally,establish the new dataset of hour-by-hour ozone concentration and pollutants and meteorological data from January 1 to June 30,2022,to evaluate the generalization ability of each model,and performs graded warning for ozone concentration according to the Ambient Air Quality Standard(GB 3095-2012),and derives the model warning correctness.The practice shows that(1)Stacking fusion model has the best effect on the validation set with MAE of 0.3538,MSE of 0.2502 and RMSE of 0.5018;(2)Stacking fusion model has the best effect on the new data set with MAE of 0.6698,MSE of 0.8374and RMSE of 0.9150,and the accuracy of ozone pollution early warning is 0.8169,which is the best performance,and indicates that the model is effective and feasible,and can make early warnings of ozone concentration exceeding the near-ground level in Beijing.