基于CNN-LSTM-Attention的月生活需水预测研究  

Research on Monthly Water Demand Prediction Based on CNN-LSTM-Attention

作　　者：陈星[1] 沈紫菡 许钦 蔡晶[4] CHEN Xing;SHEN Zihan;XU Qin;CAI Jing(College of Hydrology and Water Resources,Hohai University,Nanjing 210098,China;The National Key Laboratory of Water Disaster Prevention,Nanjing 210098,China;Yangtze Institute for Conservation and Development,Nanjing 210098,China;Hydrology and Water Resources Department,Nanjing Hydraulic Research Institute,Nanjing 210098,China)

机构地区：[1]河海大学水文水资源学院,南京210098 [2]水灾害防御全国重点实验室,南京210098 [3]长江保护与绿色发展研究院,南京210098 [4]南京水利科学研究院水文水资源研究所,南京210098

出　　处：《三峡大学学报（自然科学版）》2024年第5期1-6,共6页Journal of China Three Gorges University：Natural Sciences

基　　金：国家自然科学基金项目(52209031);中央级公益性科研院所基本科研业务费专项(Y523008,Y522018,Y520009);2023年山东省重点研发计划项目(2023CXGC010905);湖北省重点实验室开放基金(2422020009)。

摘　　要：需水预测是进行水资源配置的重要部分,对于水资源合理开发利用和社会可持续发展有重要指导意义.本文以陕西省为研究区,结合大数据分析法,提出一种基于CNN-LSTM-Attention的月生活需水预测模型.首先,通过卷积神经网络(convolutional neural networks,CNN)提取数据动态变化特征,然后利用长短期记忆(long short-term memory,LSTM)网络对提取的特征进行学习训练,最后使用注意力(attention)机制分配LSTM隐含层不同权重,预测月生活需水量并对比实际数据.结果表明,CNN-LSTM-Attention模型的相对平均误差值和决定系数(R2)分别为2.54%、0.95,满足预测精度需求,相比于LSTM模型预测精度更高.进一步证明了模型预测的合理性,可为陕西省水资源规划提供指导.Water demand prediction is a crucial part of water resources allocation and holds significant guiding significance for the rational development and utilization of water resources as well as the sustainable development of society.This paper takes Shaanxi Province as the research area and combines big data analysis to propose a prediction model based on CNN-LSTM Attention.Firstly,dynamic changes in data are extracted using Convolutional Neural Networks(CNNs).Then,the Long Short Term Memory(LSTM)networks is used to learn and train the extracted features.Finally,attention mechanisms are used to assign different weights to LSTM hidden layers,predict monthly water demand,and compare actual data.The results show that the relative mean error and R2 of the CNN-LSTM Attention model are 2.54%and 0.95,respectively,which meets the prediction accuracy requirements.Compared with the LSTM model,its prediction accuracy is higher,further demonstrating the rationality of the model prediction and offering guidance for the water resources planning of Shaanxi.

关 键 词：月尺度 需水预测 卷积神经网络 长短期记忆网络 注意力机制 因子筛选 

分 类 号：TV213.4[水利工程—水文学及水资源]