具有峰值识别的神经网络模型对水沙过程的预报

Flood Process Forecasting Model with Recognition of Flood Peak Based on Neural Network Theory

在经典BP神经网络模型的基础上,增设误差修正系数,实现网络误差修正权重倾向于输出样本的较大值.同时提出了一种计算输入输出向量的归一化公式.在此基础上建立了具有洪峰识别的BP网络预报模型.该模型能根据实测资料模拟和预报不同特征年的流量或含沙量过程.采用建立的模型,对宜昌水文站典型年实测流量过程及含沙量过程进行了预测检验,其结果与实测值吻合较好,对峰值的预报较经典BP模型有所提高.

In this paper, the BP neural network model for flood forecasting is improved when add the correct coefficient,which error weight for bigger output example value is introduced into BP model to meet the peak flood. This model can simulate and forecast different characters＇ discharge process and their peak flood. The standardization formulas are proposed for calculating the vector of input and output. Finally, two years＇ result of measuring and forecasting discharge process at Yichang Station have been proved. It can be found that forecast result agrees well with the measured data.