To achieve accurate prediction and effective control of PM2.5 concentration in dry bulk port areas, a multivariate hybrid prediction model based on support vector regression (SVR) integrating XGBoost feature selection and particle swarm optimization (PSO) intelligent optimization was proposed, namely XGBoost-PSO-SVR model. The differences in PM2.5 formation mechanisms and diffusion processes under meteorological conditions in ports were fully considered by the model. XGBoost was employed to quantify the key influencing factors, and the SVR optimized by PSO was utilized for accurate prediction. The case verification results of typical port areas in the Nanjing section of the Yangtze River show that the XGBoost-PSO-SVR model has strong robustness in the high-concentration range; compared with the univariate STL-ARIMA-SVR model, its MAE, MAPE, and RMSE decrease by 15.6%, 14.3%, and 24.9%, respectively; compared with the conventional multivariate SVR model, they decrease by 52.5%, 57.1%, and 37.7%, respectively. The XGBoost-PSO-SVR model has good prediction performance and can provide decision support for intelligent dust suppression in dry bulk port areas.