<p>The accuracy of daily runoff prediction is of great significance for environmental management, water resource management, and effective utilization of water resources. This paper proposes a data-driven modeling approach dominated by the Informer deep learning model for accurate daily runoff prediction. Compared to traditional learning models, the Informer model digs deeper into the feature information of time series and makes more accurate predictions. However, due to the non-stationarity and strong volatility of the runoff sequence, it is still difficult for a single model to realize accurate prediction. For this purpose, we coupled the Informer model with the black kite algorithm (BKA), Variational Mode Decomposition (VMD), and Error Correction strategy (EC) to propose the combined BKA–VMD–Informer–EC model for daily runoff prediction. Firstly, the VMD is optimized using BKA to obtain two parameters for the VMD to decompose the daily runoff sequence, and then a series of sub-sequences and a residual are obtained by decomposing the daily runoff sequence, and then each sub-sequence and residual are predicted using the Informer model and then superimposed, and finally, the final prediction results are obtained after error correction by the Informer model. To verify the reliability of the proposed combined model, it was applied to the daily runoff prediction at Shebu, Fenghuang, and Shuangpai hydrological stations, and the prediction model was estimated using four evaluation indexes, Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Nash-Sutcliffe Efficiency Coefficient (NSEC), and Correlation coefficient (R), and compared with eight models, including Transformer, LSSVM, LSTM, Informer, EEMD–Informer, CEEMDAN–Informer, BKA–VMD–Informer*, and BKA–VMD–Informer. The prediction results indicate that the BKA–VMD–Informer–EC model has the highest prediction accuracy; compared with the Informer single model, the MAE is reduced by 61.36%, the RMSE is reduced by 68.15%, the NSEC is improved by 72.62%, and the R is enhanced by 30.43%, for example, in terms of the Shebu hydrological station, and the prediction result is much closer to the real observation values. This study provides a new combined model and method for daily runoff prediction.</p>