基于小波分析—模糊神经网络的径流预报模型

Wavelet analysis-fuzzy neural network based runoff forecasting model

根据不同的小波分解及重构技术及不同的模糊神经网络模型训练周期,本文提出了四种小波分析与模糊神经网络相结合的径流预报模型,即:基于Mallat算法的母周期径流预报模型、基于Mal-lat算法的子周期径流预报模型、基于小波包算法的母周期径流预报模型、基于小波包算法的子周期径流预报模型,并阐述了模型建立的原理、结构及步骤。而且,以黄河源区出口水文站——唐乃亥站月径流量为应用实例,采用周期分解系数及模拟效率系数对上述四种模型进行对比评价。结果表明:基于Mallat算法的母周期径流预报模型预报效果最好,基于小波包算法的子周期径流预报模型则模拟效果最差。文中对导致这一现象的主要原因进行了分析,并对小波分析及模糊神经网络在水文模型中的应用提出了合理化建议。

In accordance with the wavelet decomposition and reconstruction technology as well as the training cycles of various fuzz- y neural network models, four runoff forecast models based on the combination of wavelet analysis with fuzzy neural network, i.e. Mallat algorithm based long cycle runoff forecasting model; Mallat algorithm based short cycle runoff forecasting model; wavelet packet algorithm based long runoff forecasting model ; wavelet packet algorithm based short cycle runoff forecasting model, are put forward herein, and then the principle, structure and step of the establishment of the models are expatiated as well. Moreover, by taking the monthly runoff data from Tangnaihal Hydrological Station one of the outlet hydrological stations at the source regions of the Yellow River as the applied case, the four models mentioned above are comparatively evaluated with the cycle decomposition coefficient and Nash-Sutcliffe efficiency coefficient. The result shows that the forecasting effect is best from the Mallat algorithm based long cycle runoff forecasting model and that from the wavelet packet algorithm based short cycle runoff forecasting model is worst. Thereby, the main causation of this phenomenon is also analyzed hereirL Furthermore, some reasonable suggestions on the application of both the wavelet analysis and the fuzzy neural network to hydrological model are presented as well.