| 1,037 | 16 | 284 |
| 下载次数 | 被引频次 | 阅读次数 |
针对传统的LSTM模型存在网络训练受阻、泛化能力减弱、预测精度和效率较低的问题，从模型结构和参数优选两方面进行改进。结构方面，在LSTM模型前加入具有多层结构的神经网络层；参数优选方面，采用多层网格搜索法选取模型参数。以长江中游典型通江湖泊——洞庭湖不同湖区的水位预测为例，与传统的LSTM模型、BP神经网络及水动力模型相比，改进型LSTM模型平均均方根误差分别减少58.80%、65.95%、44.14%;从预测计算时间来看，改进型LSTM模型所消耗的时间比传统的LSTM模型缩短62.12%,且明显少于水动力模型，总体来看改进型LSTM模型的整体性能优于其他三种模型。将改进型LSTM模型应用到三峡水库蓄水对洞庭湖水位的影响分析上，结果表明：三峡水库运行对洞庭湖不同湖区水位的影响具有明显的空间异质性，城陵矶站受其影响最为显著，其次为东洞庭湖鹿角站和西洞庭湖南咀站，南洞庭湖受影响最小。蓄水期间东洞庭湖城陵矶站水位平均下降0.44 m,最大降幅为1.55 m;鹿角站水位平均下降0.22 m,最大降幅为1.02 m;西洞庭湖南咀站水位平均下降0.27 m,最大降幅为1.28 m;南洞庭湖杨柳潭站水位平均下降0.15 m,最大降幅为1 m。研究成果为快速准确预测三峡水库影响下的洞庭湖水位提供了新的手段，同时也可为三峡水库的蓄水策略优化提供重要参考依据。
Abstract：As the traditional long short-term memory(LSTM) model has the problems of blocked network training, weakened generalization ability, low prediction accuracy and efficiency, the model structure and parameter optimization are improved in this study. As for structure, the neural network layer with multi-layer structure is added in front of the LSTM model; as for parameter optimization, multi-layer grid search method is used to select model parameters. The model is used for water level prediction of Dongting Lake, a typical river-connected lake in the middle reach of Yangtze River. Compared with the traditional LSTM model, BP neural network and hydrodynamic model, the average root mean square error(RMSE) of the improved LSTM model is reduced by 58.80%, 65.95%, and 44.14% respectively. From the perspective of predictive calculation time, the time consumed by the improved LSTM model is 62.12% shorter than that of the traditional LSTM model, and is significantly less than that of the hydrodynamic model, the overall performance of the improved LSTM model is better than the other three models. The improved LSTM model is applied to analyse the effect caused by the impoundment of the Three Gorges Reservoir(TGR) on the Dongting Lake level. The results show that the effect of the operation of the TGR on the water level of different lakes in Dongting Lake has obvious spatial heterogeneity: Chenglingji Station is the most significantly affected, followed by Lujiao Station(in the East Dongting Lake) and Nanzui Station(in the West Dongting Lake) and South Dongting Lake is the least affected. During the impoundment period, the water level of Chenglingji Station(in the East Dongting Lake) decreased by 0.44 m on average, and the maximum decrease is 1.55 m. The water level of the Lujiao station decreased by 0.22 m on average, and the maximum decrease is 1.02 m. The water level of Nanzui Station(in the West Dongting) decreased by 0.27 m on average, and the maximum decreased by 1.28 m. The water level of Yangliutan Station(in the South Dongting Lake) decreased by 0.15 m on average, and the maximum decrease is 1 m. The research results provide a new method for rapidly and accurately predicting the water level of Dongting Lake under the influence of the TGR, and also provide an important reference for the optimization of the water storage strategy of the TGR.
[1] YUAN Yujie,ZENG Guanming,LIANG Jie,et al.Variation of water level in Dongting Lake over a 50-year period:Implications for the impacts of anthropogenic and climatic factors[J].Journal of Hydrology,2015,525(10):450-456.
[2] 卢金友，姚仕明.水库群联合作用下长江中下游江湖关系响应机制[J].水利学报，2018,49(1):36-46.LU Jinyou,YAO Shiming.Response mechanism of the river and lakes in the middle and lower reaches of the Yangtze River under the combined effect of reservoir groups[J].Journal of Hydraulic Engineering,2018,49(1):36-46.
[3] 朱玲玲，许全喜，戴明龙.荆江三口分流变化及三峡水库蓄水影响[J].水科学进展，2016,27(6):822-831.ZHU Lingling,XU Quanxi,DAI Minglong.Runoff diverted from the Jingjiang reach to the Dongting Lake and the effect of Three Gorges Reservoir[J].Advances in Water Science,2016,27(6):822-831.
[4] LI Zhongwu,NIE Xiaodong,ZHANG Yan,et al.Assessing the influence of water level on schistosomiasis in Dongting Lake region before and after the construction of Three Gorges Dam[J].Environmental Monitoring Assessment,2016,188(1):1-10.
[5] 胡春宏.三峡水库175 m试验性蓄水十年泥沙冲淤变化分析[J].水利水电技术，2019,50(8):18-26.HU Chunhong.Analysis on sediment scouring and silting variation of Three Gorges Reservoir since 175 m trial impoundment for past ten years[J].Water Resources and Hydropower Engineering,2019,50(8):18-26.
[6] 王旭，肖伟华，朱维耀，等.洞庭湖水位变化对水质影响分析[J].南水北调与水利科技，2012,10(5):59-62.WANG Xu,XIAO Weihua,ZHU Weiyao,et al.Effects of water level variation on water quality in Dongting Lake[J].South-to-North Water Transfers and Water Science & Technology,2012,10(5):59-62.
[7] 方春明，曹文洪，毛继新，等.鄱阳湖与长江关系及三峡蓄水的影响[J].水利学报，2012,43(2):175-181.FANG Chunming,CAO Wenhong,MAO Jixin,et al.Relationship between Poyang Lake and Yangtze River and influence of Three Georges Reservoir[J].Journal of Hydraulic Engineering,2012,43(2):175-181.
[8] 周慧，毛德华，刘培亮.三峡运行对东洞庭湖水位影响分析[J].海洋湖沼通报，2014(4):180-186.ZHOU Hui,MAO Dehua,LIU Peiliang.The investigation on water lever of East Dongting Lake affected by Three-Gorge Reservoir[J].Transactions of Oceanology and Limnology,2014(4):180-186.
[9] XIE Yonghong,TANG Yue,CHEN Xinsheng,et al.The impact of Three Gorges Dam on the downstream eco-hydrological environment and vegetation distribution of East Dongting Lake[J].Ecohydrology,2015,8(4):738-746.
[10] 付湘，赵秋湘，孙昭华.三峡水库175 m试验性蓄水期调度运行对洞庭湖蓄水量变化的影响[J].湖泊科学，2019,31(6):1713-1725.FU Xiang,ZHAO Qiuxiang,SUN Zhaohua.Effects of 175 m experimental operation of the Three Gorges Reservoir on the storage capacity of Lake Dongting[J].Journal of Lake Sciences,2019,31(6):1713-1725.
[11] 陈斌，包为民，瞿思敏，等.双向线性回归法在椒江临海站水位预报中的应用[J].水文，2008,28(3):45-48.CHEN Bin,BAO Weimin,QU Simin,et al.Application of bidirectional propagation and multivariate linear regression method in stage forecasting at Linhai Station on Jiaojiang River[J].Journal of China Hydrology,2008,28(3):45-48.
[12] 张轩，张行南，江唯佳，等.秦淮河流域东山站水位预报研究[J].水资源保护，2020,36(2):41-46.ZHANG Xuan,ZHANG Xingnan,JIANG Weijia,et al.Study on water level forecast of Dongshan Station in Qinhuai River Basin[J].Water Resources Protection,2020,36(2):41-46.
[13] 杨佳，钱会.时间序列分析在地下水位动态预测中的应用[J].水资源与水工程学报，2015,26(1):58-62.YANG Jia,QIAN Hui.Application of time series analysis in prediction of groundwater level dynamic[J].Journal of Water Resources & Water Engineering,2015,26(1):58-62.
[14] 徐敏，王立兵，谢德尚.灰色系统GM(1,1)模型在地下热水水位预测中的应用：以河北省廊坊市为例[J].中国地质灾害与防治学报，2018,29(4):135-139.XU Min,WANG Libing,XIE Deshang.The application of GM(1,1) model of grey system in prediction of geothermal water level:a case study in Langfang City,Hebei Province[J].The Chinese Journal of Geological Hazard and Control,2018,29(4):135-139.
[15] 闫佰忠，孙剑，王昕洲，等.基于多变量LSTM神经网络的地下水水位预测[J].吉林大学学报(地球科学版),2020,50(1):208-216.Yan Baizhong,Sun Jian,Wang Xinzhou,et al.Multivariable LSTM Neural Network Model for groundwater levels prediction[J].Journal of Jilin University(Earth Science Edition),2020,50(1):208-216.
[16] 王鸿翔，朱永卫，查胡飞，等.东洞庭湖湿地生态水位阈值研究[J].长江流域资源与环境，2021,30(9):2217-2226.WANG Hongxiang,ZHU Yongwei,ZHA Hufei,et al.Study on ecological water level threshold of East Dongting Lake Wetland[J].Resources and Environment in the Yangtze Basin,2021,30(9):2217-2226.
[17] 张冬冬，戴明龙，徐高洪，等.三峡水库蓄水期洞庭湖出湖水量变化[J].水科学进展，2019,30(5):613-622.ZHANG Dongdong,DAI Minglong,XU Gaohong,et al.Research on change of the outflow of Dongting Lake during the refill period of the Three Gorges Reservoir[J].Advances in Water Science,2019,30(5):613-622.
[18] LAI Xijun,LIANG Qiuhua,JIANG Jiahu,et al.Impoundment effects of the Three-Gorges-Dam on flow regimes in two China′s largest freshwater lakes[J].Water Resources Management,2014,28(14):5111-5124.
[19] 马乐宽，邱瑀，赵越，等.基于改进的神经网络与支持向量机的小流域日径流量预测研究[J].水资源与水工程学报，2016,27(5):23-27.MA Lekuan,QIU Yu,ZHAO Yue,et al.Prediction of daily runoff in a small watershed based on improved neural network and support vector machine (SVM)[J].Journal of Water Resources and Water Engineering,2016,27(5):23-27.
[20] 黄群，孙占东，姜加虎.三峡水库运行对洞庭湖水位影响分析[J].湖泊科学，2011,23(3):424-428.HUANG Qun,SUN Zhandong,JIANG Jiahu.Impacts of the operation of the Three Gorges Reservoiron the lake water level of Lake Dongting[J].Journal of Lake Sciences,2011,23(3):424-428.
[21] 涂月明，付湘，杨会娟.基于互信息的湖泊日水位预测：以西洞庭湖为例[J].人民长江，2017,48(16):38-42.TU Yueming,FU Xiang,YANG Huijuan.Lake daily level forecast by using mutual information:case of west Dongting Lake[J].Yangtze River,2017,48(16):38-42.
[22] 王蒙蒙，戴凌全，戴会超，等.基于支持向量回归的洞庭湖水位快速预测[J].排灌机械工程学报，2017,35(11):954-961.WANG Mengmeng,DAI Lingquan,DAI Huichao,et al.Support vector regression based model for predicting water level of Dongting Lake[J].Journal of Drainage and Irrigation Machinery Engineering,2017,35(11):954-961.
[23] 郭燕，赖锡军.基于长短时记忆神经网络的鄱阳湖水位预测[J].湖泊科学，2020,32(3):865-876.GUO Yan,LAI Xijun.Water level prediction of Lake Poyang based on long shoet-term memory neural network[J].Journal of Lake Sciences,2020,32(3):865-876.
[24] ZHU Senlin,HRNJICA B,PTAK M,et al.Forecasting of water level in multiple temperate lakes using machine learning models[J].Journal of Hydrology,2020,585:124819.
[25] HRNJICA B,BONACCI O.Lake level prediction using feed forward and recurrent neural networks[J].Water Resources Management,2019,33(7):2471-2484.
[26] 王海涛，宋文，王辉.一种基于LSTM和CNN混合模型的文本分类方法[J].小型微型计算机系统，2020,41(6):1163-1168.WANG Haitao,SONG Wen,WANG Hui.Text classification method based on hybrid model of LSTM and CNN[J].Journal of Chinese Computer Systems,2020,41(6):1163-1168.
[27] 孙昭华，李奇，严鑫，等.洞庭湖区与城陵矶水位关联性的临界特征分析[J].水科学进展，2017,28(4):496-506.SUN Zhaohua,LI Qi,YAN Xin,et al.Analysis of the critical relationship between the water levels of Dongting Lake and Chenglingji station[J].Advances in Water Science,2017,28(4):496-506.
[28] 王鑫，吴际，刘超，等.基于LSTM循环神经网络的故障时间序列预测[J].北京航空航天大学学报，2018,44(4):772-784.WANG Xin,WU Ji,LIU Chao,et al.Exploring LSTM based recurrent neural network for failure time series prediction[J].Journal of Beijing University of Aeronautics and Astronautics,2018,44(4):772-784.
[29] KONG Weicong,DONG Zhaoyang,JIA Youwei,et al.Short-term residential load forecasting based on LSTM Recurrent Neural Network[J].IEEE Transactions on Smart Grid,2019,10(1):841-851.
[30] DAI Lingquan,JI Daobin,DAI Huichao,et al.Spatiotemporal variability in the effect of the water supply from the Three Gorges Reservoir on Dongting Lake during the dry season[J].Environmental Engineering Science,2020,37(12):815-825.
基本信息:
DOI：10.13928/j.cnki.wrahe.2022.02.010
中图分类号:TV697.1;P343.3
引用信息:
[1]张睿芝,戴凌全,戴会超,等.基于改进LSTM模型的三峡水库蓄水对洞庭湖水位影响的空间异质性分析[J],2022,53(02):98-108.DOI:10.13928/j.cnki.wrahe.2022.02.010.
基金信息:
国家自然科学基金青年基金项目(51809150);; 中国博士后科学基金特别资助(2019T120119);; 长江科学院开放基金(CKWV2019725/KY)