基于BP神经网络模拟闸坝下游冲刷坑空间分布

Simulation on Spatial Distribution of Downstream Scour Pit for Sluice Dam Based on BP Neural Network

闸坝下游河床的冲刷是一个复杂的非线性系统,现有的计算方法无法得到冲刷坑的空间分布。基于BP神经网络,利用某个试验工况冲刷坑的空间坐标信息训练网络,建立其他工况闸坝下游冲刷坑的插值和预测模型,此法可精准地识别出表现优异的神经网络。将BP神经网络在特征断面的插值结果与反距离加权、克里金、局部多项式插值法的插值结果对比,分析预测值和实测值的最低点差值、均方根误差和相关系数。结果表明,BP神经网络在冲刷上游段和过渡段的插值结果明显优于其他3种插值方法且能较好地预测冲刷坑发展,可利用该方法结合BP神经网络模拟闸坝下游冲刷坑的空间分布特性。

The scouring of riverbed downstream the sluice dam is a complex nonlinear system,and the existing calculation formula cannot reflect the spatial distribution of scour pit.By using the spatial coordinate information of the scour pit under a test condition,the interpolation and prediction model of downstream scour pit of sluice dam under other working conditions are established based on the BP neural network.This method can accurately identify the neural network with excellent performance.The interpolation results of BP neural network in the characteristic section are compared with the interpolation results of inverse distance weight,Kriging and local polynomial interpolation,and the lowest point difference,root mean square error and correlation coefficient between predicted and measured values are analyzed.The results show that the interpolation results of BP neural network in the upstream section and transition section of scouring are obviously better than the other three interpolation methods.The adopted method can be combined with BP neural network to simulate the spatial distribution characteristics of scour pits downstream of the sluice dam.