水库年径流改进EEMD-BNN神经网络耦合预测模型研究  

Research on Coupling Prediction Model of Improved EEMD-BNN Neural Network for Annual Runoff of Reservoirs

作　　者：侯超新 HOU Chaoxin(Dezhou Hydrological Center,Dezhou 253016,China)

机构地区：[1]德州市水文中心,山东德州253016

出　　处：《水资源开发与管理》2023年第5期24-28,共5页Water Resources Development and Management

摘　　要：为提高年径流预测预报精度,促进水库防汛抗旱、优化调度和水资源管理与保护工作顺利开展,引入三次样条插值对EEMD经验模态分解进行优化,并与BNN神经网络相融合构建EEMD-BNN水库径流预测耦合模型。三次样条插值能改进EEMD对上、下包络线的光滑拟合,便于模型准确提取径流特性的IMF模态分量和趋势项。基于变分推理的贝叶斯神经网络对IMF分量进行学习训练后,经聚合重构获得能真实反映径流时间序列特征的预测数据。结果表明,改进EEMD-BNN模型对水库径流具有很好的预测适用性和有效性,相比传统EEMD模型和EEMD-BP模型,收敛性好、精度高且具备全局寻优稳定性,可为水库中长期径流预测提供一种新的参考方法。In order to improve the accuracy of annual runoff prediction and promote the smooth implementation of flood control and drought relief,optimization scheduling,and water resource management and protection of reservoirs,the study introduces cubic spline interpolation to optimize the empirical mode decomposition of EEMD and integrates it with BNN neural network to construct an EEMD-BNN coupling model for reservoir runoff prediction.Cubic spline interpolation can improve the smooth fitting of EEMD to the upper and lower envelope lines,making it easier for the model to accurately extract the IMF mode components and trend terms of runoff characteristics.Based on variational inference,the Bayesian neural network learns and trains the IMF components,and the aggregated reconstruction obtains predicted data that can truly reflect the characteristics of the runoff time series.The results show that the improved EEMD-BNN model has good prediction applicability and effectiveness for reservoir runoff,with better convergence,higher accuracy,and global optimization stability compared to the traditional EEMD model and EEMD-BP model,and can provide a new method reference for long-term reservoir runoff prediction.

关 键 词：EEMD模态分量 三次样条插值 BNN神经网络 年径流预测 

分 类 号：TV697.21[水利工程—水利水电工程]