Abstract:To solve the problems that support vector regression(SVR) models cannot actively select optimal parameters and kernel functions, we optimize the support vector regression model by the aquila optimizer(AO) and construct the aquila optimized support vector regression (AO-SVR) model. The four models, AO-SVR and SVR, gray wolf optimized support vector regression(GWO-SVR), and whale optimization algorithm support vector regression(WOA-SVR), are combined with atmospheric pollutants, meteorological factors and hourly zenith tropospheric delay(ZTD) data in five cities of Lhasa, Urumqi, Changchun, Wuhan, and Shanghai from 2020-01-01 to 30 to predict the changes of PM2.5 concentrations in the five cities on 2020-01-31, respectively. The results show that the AO-SVR model has better applicability; the predicted values in Shanghai are the closest to the actual observed values.