摘要
为改善传统循环神经网络预测梯度消失的问题,准确预测水位变化,采用门控循环单元(gated recurrent unit,GRU)和支持向量回归(support vector regression,SVR)构建组合预测模型,对广州市猎德涌的源头西湖水位进行预测。选择了3种不同核函数下的GRU-SVR(多项式核、RBF核、Sigmoid核)模型,并确定了最佳核函数组合,探索了GRU组合模型在水文时序预测中的有效性。该组合模型通过GRU提取雨量与水位间时空特征,SVR增强整体的非线性预测能力。结果表明,GRU-SVR(多项式核)适用于湖泊降雨时期预测,与CNN-GRU及GRU、SVR相比,该模型整体预测精度分别提升了3.2%、10.3%和59.3%。
For the sake of perfecting the problem of gradient disappearance predicted by traditional cyclic neural network and accurately predicting the change of water level,gated recurrent unit(GRU)and support vector regression(SVR)were used to construct a combined prediction model to the prediction of water level of West Lake where is located in Liede Chong,Guangzhou.GRU-SVR model with three different kernel functions were compared among polynomial kernel,RBF kernel and Sigmoid kernel,which was confirmed to be the best kernel function combination.And the validity in hydrological time series prediction was explored as well.In the combined model,GRU was used to extract the spatial-temporal characteristics between rainfall and water level,and SVR was used to enhance the overall nonlinear prediction ability.The results show that GRU-SVR(polynomial kernel)is suitable for the prediction of lake rainfall period.Compared with CNN-GRU,GRU and SVR,the overall prediction accuracy of this model has been improved by 3.2%,10.3%and 59.3%respectively.
作者
刘惟飞
陈兵
余周
LIU Wei-fei;CHEN Bing;YU Zhou(School of Environment and Energy,South China University of Technology,Guangzhou 510006,China;Guangdong Provincial Engineering and Technology Research Center for Environmental Risk Prevention and Emergency Response,Guangzhou 510006,China)
出处
《科学技术与工程》
北大核心
2022年第33期14870-14880,共11页
Science Technology and Engineering
基金
广东省科技计划(2014A020216006)
广州市科技项目(201604020010)。
