To address the insufficient consideration of hydrological physical processes in current intelligent streamflow simulation, an intelligent streamflow simulation method incorporating the spatiotemporal distribution of watershed runoff generation was proposed. Taking the Yong’anxi River Basin in Zhejiang Province as a case study, a tri-SWAT-Transformer model was developed using the three daily runoff generation components simulated by the SWAT model for each sub-basin as model inputs. The simulation performance of the proposed model was compared with that of the SWAT model and other comparative models. The results show that the tri-SWAT-Transformer model achieves the best streamflow simulation performance, with Nash-Sutcliffe efficiency (NSE) values of 0.95 and 0.89 during the calibration and validation periods, representing improvements of 90.0% and 107.0%, respectively, over the SWAT model. Under high-flow conditions, the proposed model also performs best, with the NSE increased by 200.0%, and the percent bias and root mean square error reduced by 89.6% and 68.8%, respectively, compared with the SWAT model. Furthermore, correlation analysis between runoff generation components and streamflow components explains why the tri-SWAT-Transformer model outperforms the other models. The results indicate that the tri-SWAT-Transformer model can fully capture the spatiotemporal distribution characteristics of various runoff generation hydrological physical processes through the multi-head attention mechanism of Transformer, and makes better use of the spatiotemporal distribution of runoff generation information than other reference models.