LSTM神经网络和量纲分析法在弧形闸门过流计算中的对比  

Comparison between LSTM neural network and dimensional analysis method in discharge calculation of arc gates

作　　者：王艺霖 靳燕国[2] 陈晓楠[2] 段春青 张召[1] 雷晓辉[1] 常文娟[1] WANG Yilin;JIN Yanguo;CHEN Xiaonan;DUAN Chunqing;ZHANG Zhao;LEI Xiaohui;CHANG Wenjuan(State Key Laboratory of Simulation and Regulation of Water Cycle in River Basin,China Institute of Water Resources and Hydropower Research,Beijing 100038,China;Construction and Administration Bureau of South-to-North Water Transfer Middle Route Project,Beijing 100038,China;Beijing Water Affairs Center for Suburbs,Beijing 100071,China)

机构地区：[1]中国水利水电科学研究院流域水循环模拟与调控国家重点实验室,北京100038 [2]南水北调中线干线工程建设管理局,北京100038 [3]北京市水务局政务服务中心,北京100071

出　　处：《南水北调与水利科技（中英文）》2022年第3期590-599,共10页South-to-North Water Transfers and Water Science & Technology

基　　金：国家自然科学基金项目(51779268)。

摘　　要：为保障工程建筑物的合理设计、输水渠道的科学调控及调水系统的安全运行,需对弧形闸门过闸流量进行精确计算。以南水北调中线工程干渠59座节制闸为研究对象,建立基于量纲分析法和长短期记忆神经网络两种不同的弧形闸门流量计算模型,选取2018—2019年2 h时间尺度的原始观测数据作为模型输入,对比分析两种模型闸门流量计算的平均绝对误差、平均相对误差、均方根误差及纳什效率系数。结果表明:对于工程整体而言,长短期记忆神经网络方法的误差结果略优于量纲分析法,两者流量平均相对误差分别为2%~2.5%和3%~4%。另外,量纲分析法受水位波动影响较大,更适用于中线工程中下游(中、小流量)的闸门过流计算;长短期记忆神经网络方法受水位波动的影响相对略小,更适用于中线工程中上游(大、中流量)的闸门过流计算。该研究为南水北调中线工程闸门水力计算及调度运行提供了科学依据。The control gate is an important control hydraulic structure in open channel water transfer project,which controls the water level and discharge of the channel by adjusting the opening of the gate.Currently,the safety performance and service performance of the arc gate have been paid much attention.The arc gate has been widely used in various open channel water transfer projects because of its advantages such as light weight,small lifting force,stable water flow pattern,simple operation and maintenance.Therefore,the accurate calculation of the discharge of arc gate is of great significance to ensure the reasonable design of the engineering building,the scientific control of the water transmission channel and the safe operation of the water transfer system.In previous studies,the discharge of arc gate was mainly calculated by empirical formula.Due to the complex structure of the arc gate,it was difficult to calibrate the discharge coefficient,submergence coefficient and other parameters in the empirical formula.Besides,the coefficient changed with the change of flow state,so its applicable conditions had certain limitations.In addition,the discharge coefficient of arc gate of the empirical formula was a function of the gate opening degree and the upstream and downstream head difference,and the relationship was mostly nonlinear,which made parameter calibration process complicated and more erroneous.Based on this,the improved arc gate discharge calculation methods are put forward from two different levels of mechanism and data:dimensional analysis method and Long-Short Term Memory neural network.Since the beginning of water transmission of Middle Route of South-to-North Water Transfer Project,a long series of historical water situation data have been accumulated,and the amount of data fully meets the needs of model construction.Consequently,taking 59 control gates in the Middle Route of South-to-North Water Transfer Project as the research object,two different discharge calculation method of arc gates based on dimen

关 键 词：南水北调中线工程 弧形闸门 流量计算 量纲分析 LSTM神经网络 

分 类 号：TV68[水利工程—水利水电工程]