| Guanzhong area is located in the Weihe Basin.In recent years,the rapid development of social economy has led to prominent atmospheric problems.Therefore,it is particularly important to carry out air quality forecast and early warning to obtain accurate and timely air pollution information.Compared with the traditional statistical forecast,the regional air quality model is based on atmospheric dynamics,considering a variety of physical and chemical processes,and then quantitatively describes the diffusion and transport laws of pollutants,which is widely used in air quality forecast and pollution process analysis.However,due to the time lag and low spatial resolution of the emission inventory at the present stage,as well as the uncertainties of the chemical mechanism of the model and the meteorological model,there are errors between the simulated results of the air quality model and the actual observed values.In order to improve the prediction accuracy of pollutant concentration in Guanzhong region,this study aimed at the existing problems of regional air quality model in Guanzhong region,combined with multi-model numerical prediction and machine learning algorithm,carried out optimization and correction method research of regional air quality model.The main research contents of this paper are summarized as follows:(1)A multi-model air quality forecast and early warning system in Guanzhong region was established based on atmospheric chemistry numerical model.In this paper,WRF,CMAQ and CAMx models were used to carry out meteorological numerical simulation and air quality forecast and reanalysis in Guanzhong region.The numerical simulation reanalysis data and WRF model meteorological data in Guanzhong region were obtained,and the related accuracy of WRF,CMAQ and CAMx models was verified.The results show that the WRF model can accurately simulate the spatiotemporal evolution of regional meteorological factors,and the simulation accuracy of temperature and sea level pressure were relatively high,with correlation coefficients of 0.94 and 0.91,respectively.The simulation accuracy of wind speed and relative humidity were lower.The simulation results of CAMx and CMAQ modes were consistent with the observation values in time series,and PM2.5 overestimation and O3 underestimation existed in the simulation results.The simulation accuracy of CAMx was lower than that of CMAQ.(2)The machine learning algorithm was used to optimize the simulation results of single site and optimize the evaluation.At present,most of the researches focused on the optimization of forecast values by a single machine learning model,while there were relatively few researches on the comparison of optimization results of multiple machine learning models and the evaluation of optimization effects of different pollutants by models.Therefore,based on the air quality model simulation results,this paper proposed to use a variety of machine learning algorithms to correct and optimize the numerical simulation results.Meteorological factors such as temperature,relative humidity,boundary layer height,sea level pressure and wind speed were selected to verify the numerical simulation optimization results of a variety of machine learning algorithms for the five cities in Guanzhong,and to analyze the optimization performance of different algorithms for the numerical simulation of a single station.The results showed that the random forest algorithm had the highest optimization accuracy for PM2.5,and the correlation coefficient between the simulated value and the observed value after optimization was 0.74-0.8.The support vector machine algorithm had the best result for O3 optimization,and the correlation coefficient was improved to 0.79?0.88 after optimization.(3)The regional optimization network was established to optimize and verify the simulation results in Guanzhong region.At present,the research of machine learning algorithms mainly focused on the optimization of urban sites,while the regional optimization of numerical models based on algorithm combined with ensemble prediction was less.In order to realize the regional optimization of numerical model,based on the single station optimization,this paper added the data of 33 state-controlled monitoring stations,DEM and land use data in five cities of Guanzhong to build a regional optimization network,and verified and compared the optimization performance of XGBoost algorithm and LSTM deep learning neural network algorithm.The results showed that the XGBoost algorithm was superior to the LSTM algorithm in terms of regional optimization.The correlation coefficients between the simulated and observed PM2.5 values optimized by XGBoost and LSTM algorithms were 0.93-0.99 and 0.73-0.82,respectively.The correlation coefficients between the simulated and observed values of 03 were 0.85-0.97 and 0.70-0.84,respectively.In terms of regional verification,the algorithm could significantly improve the phenomenon of overestimation of PM2.5 and underestimation of O3,providing more accurate pollutant concentration value for the region.(4)The XGBoost algorithm was used to achieve the regional numerical simulation and optimization forecast of Guanzhong region in the next seven days.Based on the feasibility of regional optimization algorithm,this paper used the algorithm to revise and optimize the regional forecast data.The results showed that the RMSE value of PM2.5 prediction and optimization results in Guanzhong area was about 14.2-22.3?g·m-3,and the RMSE value of O3 prediction and optimization results was about 5.5?10.6?g·m-3.Meanwhile,the evolution process of PM2.5 and O3 concentration were consistent with the time series variation results of single station observation results.Therefore,the algorithm optimization could improve the accuracy of air quality forecast in the future,and could also achieve accurate forecast of air pollutants in areas without air quality monitoring stations,thus providing reference for pollution control and prevention. |
PDF Full Text Request