Abstract: Deep learning models have demonstrated exceptional capabilities in managing the intricate interactions among hydrological factors，leading to their adoption in hydrological forecasting.Nonetheless，there remains a gap in researches on the integration of ensemble learning with deep learning models.This study introduced a novel combined model，termed AdaBoost-Informer model，which integrates the AdaBoost algorithm with the Informer deep learning model to enhance flood runoff forecasting accuracy.The model utilizes historical precipitation and runoff data as input，with the Informer model，known for its proficiency in capturing long-term dependencies，serving as the weak learner within the ensemble framework.Hyperparameters are optimized using grid search，and AdaBoost is employed to weight and aggregate the weak learners into a robust predictor.Evaluation in the Jiaojiang River Basin in Zhejiang Province revealed that the AdaBoost-Informer model outperforms other models such as Random Forest，AdaBoost，Transformer，and Informer，achieving an RMSE of 62.08 m3/s，an MAE of 23.83m3/s，an NSE of 0.980，and a forecasting success rate of 100%.This model can significantly enhance the precision of flood forecasts and offer a valuable basis for decision-making in flood prevention and emergency management.