概念性水文模型与神经网络模型的偶和应用研究

onceptual and black box models have advantages and disadvantages. The coupling model is constituted by evapotranspiration and runoff generation modules of the Xinanjiang model, the neural network flow concentration model and the K-nearest neighbor error correction model. In the coupling model, the runoff generation module has a high accuracy and the flow concentration module has a strong nonlinear simulation ability. The flow concentration module of the coupling model does not adopt the real-time forecasting mode and can simulate the discharge continuously without using observed previous outlet flow. A global optimization algorithm by combining the SCE-UA algorithm and the genetic early stopping SCG algorithm is used to calibrate the model so the influences of experiential factors on calibration are greatly eliminated. Two types(lumped and sub-block) model is applied to Dongwan watershed and the results prove that the coupling model has a higher accuracy than the Xinanjiang model. The coupling model has advantages of conceptual and black-box model and is easy to be used.