基于集成学习与深度学习的洪水径流预报研究

Research on flood runoff forecasting based on ensemble learning and deep learning

深度学习模型凭借其对水文因素间复杂作用的优秀处理能力,在水文预报领域得到了一定的应用,然而,针对集成学习与深度学习耦合模型的研究仍有所缺失。通过融合集成学习AdaBoost算法与深度学习Informer模型,提出了一种组合模型,称为AdaBoost-Informer模型,以提高洪水径流预报的精度。该模型以历史雨量和径流数据作为数据输入,将具备长时序依赖捕获能力的Informer作为集成学习的弱预测器,使用网格搜索法进行超参数调优,使用AdaBoost集成学习算法对弱预测器进行加权组合得到强预测器。在浙江省椒江流域的应用分析表明:对比Random Forest、AdaBoost、Transformer、Informer等模型,AdaBoost-Informer模型表现最佳,RMSE为62.08 m^(3)/s,MAE为23.83 m^(3)/s,NSE为0.980,预报合格率为100%。所提模型可有效提高洪水预报精度,为防汛抢险和防洪系统调度提供决策依据。

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 m^(3)/s,an MAE of 23.83 m^(3)/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.