《计算机应用》唯一官方网站 ›› 2022, Vol. 42 ›› Issue (5): 1607-1615.DOI: 10.11772/j.issn.1001-9081.2021050829

• 前沿与综合应用 • 上一篇    下一篇

基于图注意力网络与双阶注意力机制的径流预报模型

胡鹤轩1,2,3, 隋华超1,2, 胡强1,2(), 张晔1,2, 胡震云4, 马能武5,6,7   

  1. 1.河海大学 计算机与信息学院,南京 211100
    2.水利部水利大数据重点实验室(河海大学),南京 211100
    3.西藏农牧学院 电气工程学院,西藏 林芝 860000
    4.河海大学 商学院,南京 211100
    5.长江勘测规划设计研究有限责任公司,武汉 430010
    6.长江空间信息技术工程有限公司,武汉 430010
    7.湖北省水利信息感知与大数据工程技术研究中心,武汉 430010
  • 收稿日期:2021-05-19 修回日期:2021-10-08 接受日期:2021-10-09 发布日期:2021-10-08 出版日期:2022-05-10
  • 通讯作者: 胡强
  • 作者简介:胡鹤轩(1975—),男,江苏南京人,教授,博士,CCF会员,主要研究方向:人工智能、机器学习、水利大数据
    隋华超(1997—),男,山东青岛人,硕士研究生,CCF会员,主要研究方向:数据挖掘、人工智能、水利大数据
    胡强(1992—),男,江苏镇江人,博士研究生,CCF会员,主要研究方向:机器学习、人工智能 huqianghhu@163.com
    张晔(1976—),女,江苏南京人,讲师,博士,主要研究方向:水利大数据、人工智能
    胡震云(1968—),女,江苏南京人,教授,博士,主要研究方向:水资源管理
    马能武(1965—),男,湖北天门人,教授级高级工程师,博士,主要研究方向:水利水电安全监测。
  • 基金资助:
    国家重点研发计划项目(2018YFC0407904)

Runoff forecast model based on graph attention network and dual-stage attention mechanism

Hexuan HU1,2,3, Huachao SUI1,2, Qiang HU1,2(), Ye ZHANG1,2, Zhenyun HU4, Nengwu MA5,6,7   

  1. 1.College of Computer and Information,Hohai University,Nanjing Jiangsu 211100,China
    2.Key Laboratory of Water Big Data Technology of Ministry of Water Resources (Hohai University),Nanjing Jiangsu 211100,China
    3.College of Electrical Engineering,Tibet Agriculture and Animal Husbandry University,Linzhi Xizang 860000,China
    4.Business School,Hohai University,Nanjing Jiangsu 211100,China
    5.Yangtze River Survey Planning and Design Research Company Limited,Wuhan Hubei 430010,China
    6.Changjiang Space Information Technology Engineering Company Limited,Wuhan Hubei 430010,China
    7.Hubei Research Center of Water Conservancy Information Perception and Large Data Engineering Technology,Wuhan Hubei 430010,China
  • Received:2021-05-19 Revised:2021-10-08 Accepted:2021-10-09 Online:2021-10-08 Published:2022-05-10
  • Contact: Qiang HU
  • About author:HU Hexuan, born in 1975, Ph. D., professor. His researchinterests include artificial intelligence,machine learning,big data of water conservancy.
    SUI Huachao, born in 1997, M. S. candidate. His researchinterests include data mining,artificial intelligence,big data of waterconservancy.
    HU Qiang, born in 1992,Ph. D. candidate. His research interestsinclude machine learning,artificial intelligence.
    ZHANG Ye, born in 1976,Ph. D.,lecturer. Her research interestsinclude big data of water conservancy,artificial intelligence.
    HU Zhenyun, born in 1968,Ph. D.,professor. Her researchinterests include water resource management.
    MA Nengwu, born in 1965,Ph. D.,professor of engineering. Hisresearch interests include water conservancy and hydropower safety monitoring.
  • Supported by:
    National Key Research and Development Program of China(2018YFC0407904)

摘要:

为了提高流域径流量预报的准确率,考虑数据驱动水文模型缺乏模型透明度与物理可解释性的问题,提出了一种使用图注意力网络与基于长短期记忆网络(LSTM)的双阶注意力机制(GAT-DALSTM)模型来进行径流预报。首先,以流域站点的水文资料为基础,引入图神经网络提取流域站点的拓扑结构并生成特征向量;其次,针对水文时间序列数据的特点,建立了基于双阶注意力机制的径流预报模型对流域径流量进行预测,并通过基于注意力系数热点图的模型评估方法验证所提模型的可靠性与透明度。在屯溪流域数据集上,将所提模型与图卷积神经网络(GCN)和长短期记忆网络(LSTM)在各个预测步长下进行比较,实验结果表明,所提模型的纳什效率系数分别平均提高了3.7%和4.9%,验证了GAT-DALSTM径流预报模型的准确性。从水文与应用角度对注意力系数热点图进行分析,验证了模型的可靠性与实用性。所提模型能为提高流域径流量的预测精度与模型透明度提供技术支撑。

关键词: 图神经网络, 注意力机制, 编码器-解码器, 长短期记忆网络, 时间序列预测, 水文预报

Abstract:

To improve the accuracy of watershed runoff volume prediction, and considering the lack of model transparency and physical interpretability of data-driven hydrological model, a new runoff forecast model named Graph Attention neTwork and Dual-stage Attention mechanism-based Long Short-Term Memory network (GAT-DALSTM) was proposed. Firstly, based on the hydrological data of watershed stations, graph neural network was introduced to extract the topology of watershed stations and generate the feature vectors. Secondly, according to the characteristics of hydrological time series data, a runoff forecast model based on dual-stage attention mechanism was established to predict the watershed runoff volume, and the reliability and transparency of the proposed model were verified by the model evaluation method based on attention coefficient heat map. On the Tunxi watershed dataset, the proposed model was compared with Graph Convolution Neural network (GCN) and Long Short-Term Memory network (LSTM) under each prediction step. Experimental results show that, the Nash-Sutcliffe efficiency coefficient of the proposed model is increased by 3.7% and 4.9% on average respectively, which verifies the accuracy of GAT-DALSTM runoff forecast model. By analyzing the heat map of attention coefficient from the perspectives of hydrology and application, the reliability and practicability of the proposed model were verified. The proposed model can provide technical support for improving the prediction accuracy and model transparency of watershed runoff volume.

Key words: graph neural network, attention mechanism, encoder-decoder, Long Short-Term Memory network (LSTM), time series prediction, hydrological forecast

中图分类号: