Abstract:A combined model called CNN-GRU-SSA is proposed to address the challenging issue of hyperparameter selection for Gated Recurrent Unit (GRU). This model integrates Convolutional Neural Networks (CNN), Gated Recurrent Units, and the Sparrow Search Algorithm (SSA). The proposed approach begins by employing CNN to extract features from the multidimensional dataset inputs. Subsequently, the features extracted by CNN are fed into the GRU model. Lastly, the SSA algorithm is utilized to optimize the hyperparameters of the GRU model, which is then applied to predict PM2.5 concentrations. This study focuses on the western city of Chengdu and the eastern city of Hangzhou as the research areas. Hourly data from December 1, 2021, to February 13, 2022, including atmospheric pollutants, meteorological factors, Boundary Layer Height (BLH), and Precipitable Water Vapor (PWV), were utilized for modeling. The aim was to predict the changes in PM2.5 concentrations from February 14 to February 28, 2022, for both cities. The experimental results indicate that the CNN-GRU-SSA model exhibits a significantly improved predictive accuracy compared to other models. Among these, the predicted values for Chengdu closely match the actual observations.