基于小波广义回归神经网络耦合模型的月径流预测

Monthly runoff prediction using wavelet transform and generalized regression neural network model

针对中长期水文预报方法预测结果精度低的问题,将离散小波变换(DWT)与广义回归神经网络(GRNN)耦合,建立了月径流预测模型。通过DWT处理将原始月径流序列分解重构为确定性成分和随机性成分两个分量,对两个分量的GRNN模型预测结果叠加作为预测值的方法称为WGRNN1模型。将WGRNN1模型与剔除随机序列的GRNN模型(WGRNN2)和不进行离散小波变换的GRNN模型结果进行对比,采用平均绝对误差(MAE)、确定性系数(DC)和相关系数(R)为模型评价指标。将模型应用于黑河干流莺落峡站的月径流预测,结果表明:模型WGRNN2的评价指标优于WGRNN1,且这两个模型预测效果都优于GRNN模型。说明与离散小波变换的耦合可以提高GRNN模型对月径流的预测精度,同时剔除随机成分的小波广义回归神经网络模型有更好的预测效果,可应用于实际生产。

In this study, discrete wavelet transform（DWT） and a generalized regression neural network（GRNN） were integrated to forecast monthly runoff and improve the accuracy of medium-and long-term hydrologic forecasting models. First, DTW was used to decompose the runoff series into deterministic and stochastic components, then these two components were inputted into two different GRNN models respectively, and finally the prediction results of the two models were summed up as the final forecasts of monthly runoff. To estimate the forecasting accuracy of this superposition model, we compared it with the best model taking only the deterministic component as GRNN input and the traditional GRNN model without DWT, in terms of three indexes： mean absolute error（MAE）, determination coefficient（DC）, and correlation coefficient（R）. Its application to the monthly runoff of the Yingluoxia station at the Heihe River shows that it has an accuracy slightly higher than that of the best single component model, but these two models are more accurate than the traditional GRNN. Thus, GRNN coupled with DWT improves the accuracy of monthly runoff forecasting and is useful for runoff prediction in practice.