山区流域RBF神经网络洪水预报方法

Research on flood forecasting model of radial basis function neural network in mountainous basin

建立以高斯核函数为径向基函数的神经网络模型,以崇阳溪山区流域为例进行分析。利用泰森多边形法将流域划分为6个子流域;选取流域1997-2014年中15场洪水过程作为训练样本,以流域内6个雨量站各时段雨量资料和武夷山水文站前期流量过程作为输入,以流域出口断面武夷山水文站相应流量过程作为输出,运用自组织正交最小二乘法确定径向基函数的中心,采用伪逆规则求解其权值,建立RBF神经网络洪水预报模型;利用余下的6场洪水过程对模型进行检验。结果表明:各场次洪水流量过程平均相对误差和洪峰流量相对误差绝大部分在10%以内,确定性系数均大于0.9,预报精度符合要求,可以为防汛部门预测洪水提供依据。

A neural network model with Gauss kernel function as radial basis function was established,and the mountainous watershed of Chongyang River was taken as an example in this paper.Firstly,the watershed was divided into six sub-watersheds by Thiessen polygon method.And then 15 flood processes in the basin from 1997 to 2014 were selected as training samples,the rainfall data of six rainfall stations in the basin and the pre-discharge process of Wuyishan Hydrological Station were taken as input,and corresponding discharge process of Wuyishan Hydrological Station at the outlet section of the basin was taken as output.Self-organizing orthogonal least squares method was used to determine the center of radial basis function,the pseudo-inverse rule was used to solve its weight,and then RBF neural network flood forecasting model was established.Finally,the remaining six flood processes were used to test the model.The results showed that the average relative error of each flood discharge process and the relative error of peak discharge were mostly less than 10%,the deterministic coefficient was greater than 0.9,and the prediction accuracy met the requirements,which could provide a basis for flood control departments to predict flood.