河道洪水反向演算在河库联合调度和推求设计洪水的区间组成等方面具有重要意义,直接使用传统的马斯京根法反演效果较差。基于河道上下断面洪水的非线性关系,采用BP神经网络模型训练历史洪水,并用训练得到的网络对下游洪水进行反演。实例应用结果表明,该模型反演结果与实际洪水过程更为接近,具有一定的精度和实用性。 The reverse calculation of river flood is of great significance in the joint regulation of the river bank and the calculation of the interval of design flood, and the direct inversion using the traditional Muskingum method is less effective. Based on the nonlinear relationship between the upper and lower sections of the river, the historical neural network model was used to train the historical floods and the network was trained to reverse the downstream floods. The results of practical application show that the inversion result of this model is closer to the actual flood process and has certain accuracy and practicability.