基于Wavelet-Anfis模型的陈家湾水库长期径流预测

针对径流序列特性变化复杂且难以预测的问题,提出了基于Wavelet-Anfis的径流预测模型。首先对长期径流序列进行拟合得到原始信号,通过Sym8小波对该信号进行尺度为4的分解,得到对应的低频信号A4和高频信号D1~D4,利用Anfis对这些分解信号逐步训练以确定最佳模型,从而预测出低、高频信号预测期的序列值,最后将各预测序列重构回原尺度生成径流预测值。将该方法应用于陈家湾水库年径流序列中,并与直接Anfis预测结果进行对比。结果表明,Wavelet-Anfis模型预测值的相对误差较小,且具有较快的收敛速度,既有效又准确。

Long-Term Runoff Forecast of Chenjiawan Reservoir Based on Wavelet-Anfis Model

Aiming at the runoff series of complicated variation characteristics and prediction difficultly, a model of runoff prediction was proposed based on the Wavelet-Anfis method. The long-term runoff series was fitted to get the original runoff signal. Then the runoff signal was decomposed with Sym8 wavelet at the scale of 4 into five signals which were low-frequency A4 and high-frequency D1-D4. To obtain an optimization model, these five signals were trained gradually by Anfis. A series of values about low-frequency and high-frequency signals were predicted by this model. Finally, each prediction series were reconstructed to the original scale and get the predicted value of runoff. This method was applied to the annual runoff series of Chenjiawan Reservoir. Compared with the predicted results with Anfis, the results show that the relative error was smaller and convergent speed was faster. It indicates that the proposed model is effective and accurate.