In order to improve the prediction accuracy of wide-area and continuous PM2.5 concentration monitoring,the XGBoost algorithm was used to screen the key meteorological influencing factors in Zigong area from 2017 to 2019,and the daily aerosol optical depth(AOD)was obtained by combining remote sensing MCD19A2 product inversion as input variables,and an intelligent model of PM2.5 concentration optimization optimized by improved particle swarm optimization(PSO)was established.The fitting results of PSO-SVM and PSO-PP models were used for RMSE,MAE and R2 validation analysis and evaluation of improved TOPSIS.The results showed that:1)The XGBoost algorithm was used to determine the rainfall,wind speed and temperature as the meteorological representative factors input to the PM2.5 concentration fitting model.2)The fitting effect of the improved PSO-BP model was more stable and the highest accuracy,with RMSE and MAE ranges of 0.215～0.381 and 0.146～0.234,respectively,and R2 was larger(range of 0.611～0.998),and the ranking range of most sites in TOPSIS evaluation was 1～21.3)The improved PSO-BP model fits the mean correlation coefficient R of PM2.5 concentration,which is between 0.859～0.994,and the mean R value is higher than 0.900 for 90％.The combination of AOD and meteorological factor values at 1 h and 2 h before and after the transit of the remote sensing satellite had the best modeling effect,and the mean R values of the four stations were 0.934,0.928,0.932 and 0.944,respectively.4)The seasonal fitting results of the improved PSO-BP model showed that the standard deviation had an increasing trend in the seasonal transition period,especially from summer to autumn and autumn to winter.Moreover,the fitting mean results and the real-time correlation coefficient both passed the significance test of 0.05 or above,and the correlation performance was the best in winter at 0.920,followed by spring.
万方数据