基于人工神经网络对侧堰流量系数的预测研究  

Prediction of side weir discharge coefficient based on artificial neural network

作　　者：沈桂莹 李国栋[1] 李珊珊[1] Gui-ying Shen;Guo-dong Li;Shan-shan Li(State Key Laboratory of Ecological Water Resources in Northwest Arid Area,Xi'an University of Technology,Xi’an 710048,China)

机构地区：[1]西安理工大学省部共建西北旱区生态水利国家重点实验室,西安710048

出　　处：《水动力学研究与进展（A辑）》2022年第1期125-131,共7页Chinese Journal of Hydrodynamics

摘　　要：流量系数是堰的设计重要参数之一,该文通过神经网络模型多层感知器(MLP)和广义神经网络(GRNN)对矩形尖顶侧堰流量系数(C_(d))进行建模和预测。通过MATLAB程序语言设计两种不同的神经网络模型,再以无量纲弗劳德数(F_(r))、堰长与堰宽之比(L/b)、堰长与上游水深之比(L/h_(1))和堰高与上游水深之比(P/h_(1))作为输入参数,流量系数(C_(d))作为输出参数。研究结果表明,验证阶段决定系数R^(2)=0.938、均方根误差RMSE=0.017、散度SI=0.036和平均绝对百分比误差MAPE=0.004%的GRNN模型优于MLP和GMDH模型,96%的预测数据误差在4%以下,具有较高的准确性,可为相关研究提供有价值的参考。Discharge coefficient is an important parameter for design and safe operation of weir.In this study,neural network model multi-layer perceptron(MLP)and generalized neural network(GRNN)are used to model and predict the flow coefficient(C_(d))of the rectangular sharp-crested side weir.Through two different machine learning models designed by the MATLAB programming language,the dimensionless Froude number(F_(r)),the ratio of weir length to weir width(L/B),the ratio of weir length to upstream water depth(L/h_(1))and the ratio of weir height to upstream water depth(P/h_(1))are regarded as input parameters,and the discharge coefficient(C_(d))as output parameter.The results show that the GRNN model with R^(2)=0.938,RMSE=0.017,SI=0.036,and MAPE=0.004%in the validation stage is superior to the MLP and GMDH model,the errors for96%of the forecasts are within 4%or less,which performs better and is more accurate.It can provide valuable reference for relevant researchers and engineers.

关 键 词：侧堰 流量系数 神经网络 MLP GRNN 

分 类 号：TV135.2[水利工程—水力学及河流动力学] TP183[自动化与计算机技术—控制理论与控制工程]