用人工神经网络反演扩散波方程

Identification of parameters in flood diffusion wave equation by artifical neural network

本文通过对人工神经网络与传统扩散波洪水演算方法的一致性研究,提出利用神经网络反演线性扩散波模型的方法,探索从一种新的角度研究洪水波运动传播特性的途径。本文利用断面实测洪水流量资料,应用神经网络对其进行训练,根据网络连接权值辨识扩散波方程的参数,并依据扩散波差分方程预测河道下游断面的流量过程。应用实例表明该方法的模拟精度满足要求,模拟效果良好。说明采用神经网络方法辨识出的扩散波模型能较好地描述洪水波的运动规律,而且,这将极大地扩展人们研究洪水波传播规律的手段和方法。

In this paper a new approach is proposed, in which the parameters in the flood diffusion wave equation are identified with artificial neural network (ANN). Flood diffusion wave equation is reinstated through trained ANN, and forecasting down stream outflow. This study opens up a new way to mine observed data for knowledge discovery. This approach was applied to QingJiang channel, and satisfactory results were acquired. It appears to be quite generally applicable.