【摘要】 在小流域水文数据稀缺的情况下，传统水文模型存在适应性差、预报精度不高问题。以典型小流域石砭峪水库为研究对象，引入前期影响雨量(Pa)构建卷积神经网络-长短期记忆网络(CNN-LSTM+Pa)融合模型，对比分析Pa对模型预报性能影响。结果表明:融合模型可显著提升降雨径流关系的学习能力，NSE均值达0.843，洪峰、洪量误差均值显著降低。研究方法可供小流域洪水预报参考。更多还原

【Abstract】 In case of scarce hydrological data in small watersheds,traditional hydrological models are of problems such as poor adaptability and lower forecasting accuracy. Taking the Shibianyu Reservoir in a typical small watershed as the research objective,the antecedent precipitation index( Pa) is introduced to construct a convolutional neural network plus long short-term memory network hybrid model( CNN-LSTM+Pa),and the impact of Pa on the forecasting performance is analyzed comparatively. The results show that the hybrid model significantly improves the learning ability of the rainfall-runoff relationship,with the mean Nash-Sutcliffe efficiency( NSE) reaching 0.843,and the mean errors of flood peak and flood volume significantly reduced. The research can be useful reference for flood forecasting of small watersheds.更多还原