PM_2.5 CONCENTRATION PREDICTION AND UNCERTAINTY ANALYSIS BASED ON A COMPOSITE MODEL

In this paper, GIS software and Kriging interpolation method were used to analyze the spatial and temporal distribution of PM_2.5 concentration in Hefei city circle. According to historical environment monitoring data, ground meteorological stations and historical meteorological data of Hefei, multiple regression analysis, correlation analysis and other methods were adopted to study the influencing factors of PM_2.5 concentration in Hefei. The results showed that: 1) the overall change of PM_2.5 concentration in the above cities was in the order of winter > autumn > spring > summer, and PM_2.5 concentration in most cities peaked in January, then gradually declined, reached the lowest value in July, and then gradually increased; 2) PM_2.5 concentration showed a highly positive correlation with CO, with the correlation coefficient as high as 0.875. The correlation with PM₁₀, SO₂ and NO₂ was also high. There was a negative correlation with O₃. PM_2.5 concentration was negatively correlated with air pressure, wind speed, rainfall and visibility, and strongly positively correlated with temperature and relative humidity. Based on PM_2.5 concentration monitoring data of Hefei from 2018 to 2019, a composite model was built to predict PM_2.5 concentration, and three times exponential smoothing model was compared to determine that simulated annealing+genetic+three times exponential smoothing was the optimal composite model, with the fitting degree reaching 95%. Kappa and MAPE indexes were used to analyze and evaluate the uncertainty of the composite model. Kappa and MAPE indexes were 0.654 and 0.072 respectively, indicating that the model was highly stable. The proper combination of prediction factors and the study of model uncertainty were helpful to improve the model prediction accuracy, so as to provide theoretical basis and method for the monitoring and evaluation of atmospheric environment quality.