不同组合小波神经网络模型对径流预测的适用性

针对水文时间序列的非平稳性特征，以长江三峡宜昌站1904~2003年年平均流量为例，分别建立了小波分析（WA）与BP神经网络和径向基函数神经网络（RBF）耦合的预测模型，探究了两种组合模型的预测效果，并与传统的单一人工神经网络模型对比；并采用5种常见的预测性能评价指标分析预测效果。结果表明：组合模型预测成果的精度较单一模型显著提高；组合和单一模型中RBF网络模型均优于BP网络模型；小波径向基函数神经网络组合模型具有较优的预测精度和泛化能力，是提高预测精度的有效方法，在径流预测中具有可行性。

Applicability of different combination of wavelet neural network models to runoff prediction

According to the non-stationary characteristics of hydrological time series, taking the average annual runoff at Yichang Station from 1904 to 2003 for example, two coupling forecasting models, wavelet analysis (WA) and the BP neural network, wavelet analysis (WA) and radial basis function (RBF) neural network, are established, and the forecasting results are compared with the traditional single artificial neural network model, also the prediction effects are analyzed by five frequently used prediction performance evaluation index. The results show that the prediction accuracy of the integrated model is higher than that of the single model; the RBF network model is superior to the BP network model both in the integrated model and single model; WARBF integrated model has better prediction accuracy and generalization ability, which is a useful way to increase prediction accuracy and is feasible for runoff prediction.