Prediction of discharge in a tidal river using the LSTM-based sequence-to-sequence models

The complexity of river-tide interaction poses a significant challenge in predicting discharge in tidal rivers.Long short-term memory(LSTM)networks excel in processing and predicting crucial events with extended intervals and time delays in time series data.Additionally,the sequence-to-sequence(Seq2Seq)model,known for handling temporal relationships,adapting to variable-length sequences,effectively capturing historical information,and accommodating various influencing factors,emerges as a robust and flexible tool in discharge forecasting.In this study,we introduce the application of LSTM-based Seq2Seq models for the first time in forecasting the discharge of a tidal reach of the Changjiang River(Yangtze River)Estuary.This study focuses on discharge forecasting using three key input characteristics:flow velocity,water level,and discharge,which means the structure of multiple input and single output is adopted.The experiment used the discharge data of the whole year of 2020,of which the first 80%is used as the training set,and the last 20%is used as the test set.This means that the data covers different tidal cycles,which helps to test the forecasting effect of different models in different tidal cycles and different runoff.The experimental results indicate that the proposed models demonstrate advantages in long-term,mid-term,and short-term discharge forecasting.The Seq2Seq models improved by 6%-60%and 5%-20%of the relative standard deviation compared to the harmonic analysis models and improved back propagation neural network models in discharge prediction,respectively.In addition,the relative accuracy of the Seq2Seq model is 1%to 3%higher than that of the LSTM model.Analytical assessment of the prediction errors shows that the Seq2Seq models are insensitive to the forecast lead time and they can capture characteristic values such as maximum flood tide flow and maximum ebb tide flow in the tidal cycle well.This indicates the significance of the Seq2Seq models.

Key Technologies for Design of Changqing Yellow River Super-long Bridge of Zhengzhou Ji'nan HSR

Changqing Yellow River Super-long Bridge of Zhengzhou-Ji'nan HSR is a partial cable-stayed bridge with concrete main girder and a unit length of 1,080 m.Studies are carried out on the key technologies of bridge design,and the main conclusions are as follows:The whole unit adopts the supporting system of tower pier consolidation and tower-beam separation,and each pier is provided with seismic mitigation and isolation bearing;shaped-steel reinforced concrete bridge tower is adopted to bring into full play the tensile performance of steel and the compressive performance of concrete,and avoid the construction challenges of setting up multi-layer and multi-stirrup reinforcement while improving the bearing capacity of section;a new type of double-side and bi-directional anti-skid anchorage device is adopted for the cable saddle of wire divider pipe in order to withstand the unbalanced cable force,and verify the reliability of the anti-skid anchorage device by solid model test;and large-segment cantilever pouring design is adopted for the main girder with a maximum segment length of 8 m to effectively shorten the construction period of the bridge.

Variation Characteristics of the Large-scale Frost in East Region of the Yellow River of Gansu in Recent 40 Years

[Objective]The research aimed to study variation characteristics of large-scale frost in the east region of the Yellow River of Gansu in recent 40 years.[Method]Based on daily minimum temperature data at 15 meteorological stations over the east region of the Yellow River of Gansu from 1969 to 2008,according to common climatic statistical index of the frost,variation characteristics of the large-scale frost and continuous frost in the east region of the Yellow River of Gansu in recent 40 years were studied.[Result]Since the 1990s,average last frost date in the east region of the Yellow River of Gansu obviously advanced,and first frost date started to obviously postpone.Advancing time of the last frost date was longer than postponing time of the first frost date.Average frost-free period also obviously prolonged.Extremely early first frost date and extremely late last frost date mainly happened in the 1970s and the 1980s.Extremely late first frost date and extremely early last frost date mainly happened after the middle period of the 1990s.Extremely long frost-free period gradually started to appear frequently.In recent 40 years,the continuous frost gradually decreased,and the intensity also declined.[Conclusion]The research was favorable for understanding change characteristics of the frost and climate in the east region of the Yellow River of Gansu,and had important guidance significance for improving prediction capability of the abnormal frost disaster,effectively preventing frost disaster and improving crop yield in the area.

Analysis of long-term dependence phenomenon in Benue River flow process and its hypothesis testing

In this paper, the long-term dependence phenomenon （the Hurst Effect） which characterizes hydrological and other geophysical times series is studied. The long-term memory is analysed for both daily and monthly streamflow series of the Benue River at Makurdi, Nigeria by using heuristic methods and testing specifically the null hypothesis of short-term memory in the monthly flow series. Results obtained by applying heuristic procedures indicated that there may be the presence of long-term memory component in mean daily flow series but there is no discernible reason to suspect the presence in both average monthly and maximum monthly flow series （extreme event）. Hypothesis testing was conducted by using original and modified versions of rescaled range statistic. When the modified rescaled range, which accounts for short-term memory in the series, is used, the null hypothesis is accepted for both the average monthly and maximum monthly flow series, indicating little or no probable presence of long-term memory in the series. An identical conclusion is also arrived at when second null hypothesis for independence of the monthly flow series is tested. Therefore, apart from the mean daily flow series, there is little evidence of long-term dependence in the Benue River streamflow series at Makurdi. However, considering the limited length of data used, the results are inconclusive.

Assessment of Salinization in the Main River Systems of Long An Province, Vietnam

Three provinces including Long An, Tien Giang and Dong Thap, located at Dong Thap Muoi area in the Mekong Delta, belong to the southern key economic region of Vietnam. These provinces run from west to east with a vast network of rivers. Among them, VCD （Vam Co Dong）, VCT （Vam Co Tay） and Soai Rap rivers are the three biggest rivers crossing the province. In recent years, the flows of rivers have changed a lot and saline intrusion has been very complicated. Salt water has intruded deeply into the field. In the 2000s, salt water only reached Tuyen Nhon （Thanh Hoa district）, but now it has reached Vinh Hung district, about 50 km far from Tuyen Nhon. In 2016, the maximum of salinity in Xuan Khanh （Duc Hoa district） in VCD river was 6.8‰ and Tuyen Nhon （Thanh Hoa district） in VCT river was 5.2‰. In the coming years, the situation of salt intrusion is going to happen more unpredictable under effects of global climate change and sea level rise.

Threats and Avoidance Measures of Frost Damage of‘Shushanggan’Apricot in the Ili River Valley

[Objectives]To explore the threat factors of frost damage to‘Shushanggan’Apricot in the Ili River Valley,and to provide measures for avoiding frost damage.[Methods]Based on the meteorological data of the Ili River Valley counties and cities during the 12 years from 2010 to 2021 and using the critical low temperature of‘Shushanggan’Apricot as the main factor,we comprehensively analyzed the threats of low temperature in winter in January and late frost in April in spring in the Ili River Valley.[Results]During the 12 years,there were 4 years of low temperature below the critical(-26—-28℃)of‘Shushanggan’Apricot in the Ili River Valley counties and cities in January,accounting for 33.3%,and a total of 59 d.The frequency of occurrence was:Nilka County>Qapqal County>Yining City>Gongliu County>Huocheng County>Khorgos City>Yining County>Tekes County>Xinyuan County.In April,there were 9 years with a low temperature below the critical temperature(-0.6℃)flowering and fruit setting of‘Shushanggan’Apricot,accounting for 75%,and a total of 134 d.The frequency of occurrence was:Nilka County>Tekes County>Gongliu County>Yining County>Huocheng County>Khorgos City>Xinyuan County>Yining City>Qapqal County.The low temperature threats of‘Shushanggan’Apricot suitable cultivation areas were ranked as follows:Nilka County>Gongliu County>Tekes County>Qapqal County>Huocheng County>Yining City>Yining County>Khorgos City>Xinyuan County.Combined with the observation and survey of frost damage on the spot,we comprehensively analyzed and evaluated the cultivation area of‘Shushanggan’Apricot in the Ili River Valley:three counties(Nilka County,Gongliu County,and Tekes County)in the eastern region,except Xinyuan County,suffered frequent late frost damage,are suitable areas for the cultivation of‘Shushanggan’Apricot;three counties and two cities in the western region(Qapqal County,Huocheng County,Yining City,Yining County,Khorgos City)and Xinyuan County in the eastern region are suitable areas for‘Shushanggan’Apricot.The inversion zone at an altitude of 820-1100 in the valley is the superior area for‘Shushanggan’Apricot.[Conclusions]We explored the suitable areas in the origin area of‘Shushanggan’Apricot,and came up with measures to avoid frost damage,to provide a reference for the development of‘Shushanggan’Apricot.

机器学习算法在降水和气温多模式集成中的应用

选取CMIP6中5种全球气候模式,利用算术平均、权重平均、多元线性回归、BP神经网络、长短期记忆(LSTM)神经网络和随机森林(RF)等6种多模式集成方法,基于黄河流域水源涵养区历史降水量和气温数据,评估不同集成方法的模拟效果,并选取模拟效果最好的多模式集成方法预估未来SSP1-2.6、SSP2-4.5和SSP5-8.53种情景下黄河流域水源涵养区的降水和气温变化趋势。结果表明:多模式集成能很好地再现基准期降水和气温变化,3种机器学习算法表现相对较好,其中LSTM神经网络最好;在未来3种情景下,多年平均降水量均有所增加,四季降水量变化各有差异;SSP1-2.6情景下年降水量峰值出现在各时段初期,SSP2-4.5和SSP5-8.5情景下的年降水量呈增长趋势,远期下降趋势较明显;3种情景下气温都呈上升趋势,但变化差异较大,增温幅度和速率由小到大为SSP1-2.6、SSP2-4.5、SSP5-8.5,秋季气温增幅最大,冬季最小;多模式集成方法对未来降水量和气温的预估存在较大的不确定性,均表现为中远期大于近期,降水量预估的不确定性比气温大,其中降水量秋冬季不确定性明显大于春夏季。

情境优化、结构重塑与行为纠偏:农村生态河道长效管护的路径突破

加强农村河湖管理是实施乡村振兴战略的重要任务。在自上而下的政策引导下,各级政府全面推进农村生态河道建设工作。然而,农村生态河道管护是一项长期任务和系统工程。反思既有研究从单一维度分析农村生态河道管护的局限性,在对P镇生态河道管护的运行逻辑及长效机制分析基础上发现农村生态河道要实现真正意义上的长效管护,必须构建集保障、组织、运行于一体的管护体系,并从情境优化、结构重塑、行动纠偏等三个层面提出相应的治理思路。

基于可解释机器学习的黄河源区径流分期组合预报

黄河源区是黄河流域重要的产流区和我国重要的清洁能源基地,提高黄河源区径流预报准确率可为流域水资源科学调配和水风光清洁能源高效利用提供重要支撑。以黄河源区唐乃亥和玛曲水文站为研究对象,基于不同月份径流组分的差异,考虑积雪覆盖率及融雪水当量变化,构建了中长期径流分期组合机器学习预报模型及其可解释性分析框架。研究结果表明:1)年内的径流预报时段可划分为融雪影响期(3—6月)和非融雪主导(以降雨和地下水补给为主)期(7月—次年2月);2)与传统不分期模型相比,唐乃亥站和玛曲站分期组合预报模型的纳什效率系数分别达0.897、0.835,确定系数(R2)分别达0.897、0.839,均方根误差分别降低了10%、17%,提高了径流预报准确率,通过分位数映射校正,唐乃亥站和玛曲站预报模型的R2分别进一步提升至0.926和0.850;3)基于SHAP机器学习可解释性分析框架,辨识了预报因子对径流预报结果的贡献程度,由高到低依次为降水、前一个月流量、蒸发、气温、相对湿度、融雪水当量等,发现了不同预报因子之间交互作用散点分布具有拖尾式或阶跃式的特征。

基于长短时记忆网络的山区中小流域降雨径流模拟

洪水预报是流域防洪减灾的重要非工程措施之一。目前我国中小河流暴雨洪水灾害频发,但应对短历时强降雨的洪水预报能力仍不强。以安徽省东部山区中小流域为研究对象,引入长短时记忆网络建立流域降雨径流模型,探讨其在山区中小流域的洪水模拟效果。结果表明,考虑降雨输入的空间差异可提升深度学习模型降雨径流模拟预测性能,且长短时记忆网络能够取得优于传统人工神经网络的精度;长短时记忆网络模型有效建立了流域降雨与径流间的复杂非线性关系,模型在所选流域内场次洪水的峰值模拟效果较好,训练、测试集场次洪水峰值合格率均在90%以上;长短时记忆网络内部结构特征与流域水文过程具有较好的相似性,对山区中小流域暴雨洪水非线性关系拟合效果突出。

水凝胶包裹的透水材料在寒区河道工程中应用及性能分析

随着人们生活水平的提高,对河道工程的完善提出了更高要求,提升寒区透水混凝土铺装使用时间的重要举措就是在防冻胀设计上做到有效性。将实际工程作为对象,利用现场试验方法,分别设定3种找平层材料和基层材料,分析不同方案下寒区河道工程的抗冻胀性能。试验结果显示,河道工程基层与找平层材料可使用透水水稳与中粗砂材料。建议使用相应的工程技术机制,对滞留基层的径流进行及时排空,最大限度弱化冻胀的破坏问题。

某高速铁路沿线天然场地季节冻胀特性监测与分析

某高速铁路沿线季节性冻土大面积分布,本文依据地形地貌、土性成分及水文地质条件等特征,选定14处天然场地。通过冻融周期的监测,得到沿线地基土的冻结融化速率、最大季节冻深及地表变形量等重要数据。通过分析积雪场地及裸露场地的冻结融化过程与地表变形情况,结合气温、降水量、地表雪盖等因素,分析了高速铁路沿线季节性冻土的冻融特性。研究结果表明:(1)沿线有雪条件下天然场地最大冻结深度19~90 cm,平均值为45 cm;(2)沿线季节冻土冻结速率0.27~1.20 cm/d,融化速率普遍大于冻结速率,为0.28~1.92 cm/d;(3)积雪对保持土体温度、抑制土体冻结具有良好效果,能有效缩短冻结、融化时间,使最大冻结深度减小18%~24%;(4)冻胀量呈由西向东增大趋势,其变化规律和气候条件吻合。研究成果可为类似季节性冻土地区高速铁路设计及冻害防治提供借鉴。

粗骨料最大粒径对水工混凝土变形性能及长期耐久性的影响

为研究不同粗骨料最大粒径对水工混凝土的变形影响,探明国家标准规定的骨料粒级能否满足水工混凝土的要求,依托引江济淮工程,开展了粗骨料最大粒径为31.5 mm和40.0 mm的混凝土变形试验,探究2种骨料粒级混凝土长期耐久性能演变规律。结果表明:采用最大粒径31.5 mm骨料的混凝土极限拉伸值和弹性模量略大,动态疲劳性能较好;采用最大粒径40 mm骨料的混凝土干燥收缩以及自生体积变形小;2种不同骨料最大粒径的混凝土徐变度相当;采用最大粒径31.5 mm骨料的混凝土长期抗渗性和长期抗冻性均优于采用最大粒径40 mm骨料的混凝土。因此,最大粒径31.5 mm骨料应用于水工混凝土时,能满足水工混凝土性能要求且动态疲劳寿命及耐久性更优,但需要注意混凝土开裂问题。

基于物联网和GCNN-LSTM的河流水文预测方法

针对河流水文存在预测精度不高的问题,利用物联网技术设计了分布式的降雨和水文信息自动采集系统,并提出了一种基于图卷积神经网络和长短期记忆网络模型对河流水位和径流量进行预测的方法;首先通过分析确定了影响河流水文的主要因素,将流域范围内的降雨量信息组成网格化的二维图形矩阵;然后提出了GCNN-LSTM预测模型,将含有降雨信息的二维图形矩阵作为网络模型的输入,获取该流域内降雨与水文变化的时空分布特征;最后采用所提出的GCNN-LSTM预测模型对河南省周口市段颍河的历史水文数据进行训练,再利用训练后的网络对测试集数据进行预测,得到了较高精度的径流量和水位结果,径流量预测结果的RMSE、MAPE和MAE分别仅为17.09 m^(3)/s、1.68%和8.57 m^(3)/s,水位预测结果的RMSE、MAPE和MAE分别仅为0.32 m、0.65%和0.29 m,与其他几种预测方法相比表现出了优越性,对科学合理利用水资源和防洪减灾具有重要意义。

舟曲虎家崖滑坡直接转化泥石流的特征、过程与机理研究

滑坡直接转化而成的泥石流通常比滑坡本身运移距离更远,波及范围更广,危害也更严重,但目前相关转化机制及其影响因素的研究较少。2020年10月20日位于白龙江左岸、甘肃省舟曲县的虎家崖暴发小型泥石流,侵入省道313,威胁毗邻村庄。通过现场调查、室内岩土试验、遥感解译和降雨资料分析,本文对虎家崖泥石流形成的地质环境条件、运移及堆积特征等进行了调查,研究了浅层滑坡失稳并直接转化为泥石流的主要特征、过程、影响因素及其形成机理,探讨其未来发展趋势。研究表明:(1)虎家崖上游狭长沟道堆积体受到长达19 d、累计98.5 mm的持续性降雨的激发而发生慢速滑坡,区域强震造成的坡体结构损伤对滑坡发育也有贡献。(2)滑坡后段与滑坡前缘的剪出部分在运动过程中直接转化为泥石流,但中段滑坡体则维持剪切滑动。(3)转化过程受滑坡前缘剪出土体饱水程度、沟道坡度与落差以及地表径流等因素的影响。(4)虎家崖狭长沟道后缘可能因毗邻泄流坡滑坡活动牵引而破坏,同时上游沟道残余滑坡体与灰岩崩塌体、出口右侧残余滑坡体可能受降雨、地震等激发而失稳,再次转化为泥石流,威胁沟道下方虎家崖村和省道S313。

计及短期运行特征的水风光互补中长期调度

传统基于旬/月尺度的中长期调度模型难以描述风、光出力随机波动性所诱发的运行风险,无法有效指导水风光互补系统的中长期运行。为此,提出计及短期运行特征的水风光互补中长期调度规则编制方法。首先,构建短期调度模型评估水风光联合运行中的弃电、弃水及失负荷风险;其次,基于短期调度长系列调度样本,提取水电出力与弃电、弃水及失负荷之间的响应关系;最后,将响应关系嵌套在水风光互补隐随机优化调度模型中提取调度函数。以黄河上游千万千瓦级水风光互补系统为研究对象进行验证,结果表明:相比于简化运行策略,所提出的方法能够增加系统发电量6.5%,同时降低失负荷率3.71%。在水风光互补中长期调度模型中嵌套短期运行风险能够降低系统的弃电与失负荷风险,从而有效指导互补系统中长期调度运行。

长江安庆江段存在溯河洄游型和淡水定居型刀鲚实证研究

为了探究长江十年禁渔后安庆江段刀鲚的生境履历,利用X射线电子探针微区分析技术研究安庆江段不同类型长颌刀鲚(Coilia nasus)和短颌刀鲚的耳石Sr和Ca微化学特征。结果显示,根据耳石Sr/Ca值的变化值将安庆江段的短颌刀鲚分为2类,一类是其比值为一直小于3.0的低值,表明其纯淡水的生境履历;另一类是其比值不仅有小于3.0的低值区,还有大于3.0的高值区(小于7.0),表明其不仅有淡水的生境履历,还有高盐度的河口半咸水生境履历。长颌刀鲚的耳石Sr/Ca值均具有小于3.0的低值区和大于3.0(甚至大于7.0)高值区的显著波动,表现为典型的淡水、河口半咸水及海水的溯河洄游型生境履历,Sr含量面分析图谱也可印证上述结果。本研究表明,长江安庆江段刀鲚群体组成较为复杂,同时存在溯河洄游型、淡水定居型短颌刀鲚和溯河洄游型长颌刀鲚3种生态表型个体。

基于相关性检验的VMD-LSTM耦合模型月径流模拟研究

近年来,极端强降雨和干旱事件频发,流域水文过程的不确定性变化加剧,使得流域中长期径流预测的难度增加。为提升LSTM(长短期记忆神经网络)模型对径流时序变化的捕捉及拟合能力,以博阳河流域为研究区域,选取月降雨、蒸发及流量数据,利用VMD(变分模态分解)和相关性检验,排除无关频率分量对LSTM模型规律学习的干扰,以达到模型输入优选的目的;此外,还考虑了VMD与LSTM模型的不同耦合方式对模型精度和稳定性的影响,最终优选出二者兼具的VMD-LSTM月径流耦合模式。结果表明:VMD-LSTM耦合模型可显著提升模拟精度,但在模型稳定性方面有所欠缺;而基于相关性检验的VMD-LSTM耦合模型不仅能够进一步提高模型精度,并且在模型的稳定性方面也有所改进。在基于相关性检验的VMD-LSTM耦合模型的不同耦合方式对比中,对输入、输出均进行VMD分解且对输入变量进行优选的D_(1)耦合方案的模拟效果最好,其60次模拟计算的NSE均为0.98以上且稳定性极佳;另外,在分析方案D_(1)的可解释性时发现历史径流对于LSTM模型的影响要比降雨和蒸发大。该研究结论可为流域水资源管理提供精准可信的中长期径流模拟成果。

基于长短期记忆网络的CO_(2)气层识别方法

CO_(2)监测是油气开采中的关键环节,传统的CO_(2)监测方法面临很多挑战,在人工智能逐渐兴起的当下,深度学习技术被广泛应用于地球物理测井。珠江口盆地恩平凹陷深层CO_(2)气藏发育,传统测井方法无法准确评价储层流体。构建了基于长短期记忆网络(LSTM)的CO_(2)气层识别模型,采用m×2正则化交叉验证优选CO_(2)敏感测井参数,并对模型进行训练。利用该模型对珠江口盆地恩平凹陷L2井CO_(2)气层进行识别,并与支持向量机和K近邻算法识别结果进行对比。结果表明,3种深度学习算法对CO_(2)气层的识别效果良好,其中LSTM算法对CO_(2)气层的识别效果最好,准确度达93.4%,为深层CO_(2)气层识别工作提供了新思路。

基于RF-LSTM模型的长江上游汛期日含沙量预测

含沙量预测对流域泥沙治理、水沙调控以及水质与水环境管理等具有重要意义。长江上游地区幅员广阔,支流众多,水沙来源复杂,对准确预测三峡入库含沙量过程构成了挑战。针对长江上游区间降雨和干支流来水来沙对寸滩站含沙量产生不同程度的影响,提出了一种基于随机森林(Random Forest,RF)与长短时记忆(Long Short Term Memory,LSTM)神经网络结合的日含沙量预测深度学习模型RF-LSTM。首先,该模型利用RF算法筛选出与寸滩站日含沙量相关性强的水沙因子,然后将这些因子作为LSTM神经网络的输入变量,进一步识别出优选水沙因子与寸滩含沙量之间的映射关系,最后以长江上游向家坝至寸滩区间为研究区域,应用该模型对不同预见期下的寸滩站汛期日含沙量进行了预测,结果表明:与LSTM模型相比,RF-LSTM模型能较好地考虑预测因子对含沙量影响的滞后效应,且有效捕获与寸滩站日含沙量相关性强的特征,四种预见期下其在预测精度和性能方面均有较好表现,其中无预见期和预见期1 d时两种模型预测精度均较高,验证期的纳什效率系数均大于0.82,无预见期下RF-LSTM模型的纳什效率系数可达到0.91,相应的均方根误差和平均绝对误差分别较LSTM模型降低了13%和8%,且两种预见期下RF-LSTM模型可以较为准确捕获沙峰及峰现时间;当预见期增加至2 d和3 d时两种模型精度均有明显下降,但RF-LSTM模型计算精度仍优于LSTM模型。研究结果可为长江上游含沙量预测提供参考。