基于CNN-Seq2seq的河道水位区间预测方法  

A river water level interval prediction method based on CNN-Seq2seq

作　　者：孙英军 唐为昊 王成[2] 李英德[2] SUN Yinjun;TANG Weihao;WANG Cheng;LI Yinde(Zhejiang Hydrological Management Center,Hangzhou 310009,China;College of Mechanical Engineering,Zhejiang University of Technology,Hangzhou 310023,China)

机构地区：[1]浙江省水文管理中心,浙江杭州310009 [2]浙江工业大学机械工程学院,浙江杭州310023

出　　处：《浙江工业大学学报》2022年第4期381-392,405,共13页Journal of Zhejiang University of Technology

基　　金：国家自然科学基金资助项目(L1924063);浙江省自然科学基金资助项目(LY18G020018)。

摘　　要：准确的河道水位预测在水资源利用和防洪减灾辅助决策中具有重要作用。在水文流域相关物理过程未知的情况下,构建了一种基于一维卷积和长短期记忆网络的混合深度学习区间预测模型——卷积-序列到序列网络(CNN-Seq2seq),结合卷积网络和长短期记忆网络能够提取不同数据特征的优势,使构建的模型能充分提取水文过程的隐含统计特征。选择其他5种预测模型,利用流域内水文测量站点的数据记录完成模型训练和对比试验。实验结果表明:相较于其他模型,CNN-Seq2seq具有更好的泛化能力,在洪水过程的水位预测上具有更高的精度。Accurate prediction of water level in the river plays an important role in water resources utilization and auxiliary decision-making for disaster mitigation.In the case that the relevant physical processes of the basin hydrological are unknown,a hybrid deep learning water level interval prediction model(CNN-Seq2seq)based on one-dimensional convolution and long short-term memory network is constructed,which combines the strengths of convolutional network in data feature extraction and long short-term memory network in time series data processing.The constructed model is given the ability to extract the water level,rainfall and discharge information of the hydrological process.Five other prediction models were selected,model training and comparison experiments were completed using data records from survey stations in the basin.The experimental results show that compared with other models,CNN-Seq2seq has better generalization ability,and has higher accuracy in the prediction of the water level in the river,especially the water level during the flood.

关 键 词：数据驱动水位预测 一维卷积网络 长短期记忆网络 CNN-Seq2seq 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] P333.9[自动化与计算机技术—控制科学与工程] P334.9[天文地球—水文科学]