摘要
城市内河水位预报对城市内涝风险管理具有重要意义。传统数值模拟模型计算效率较低,且无法实时计算。针对以上问题,提出一种基于Gaussian函数改进BP神经网络的河道水位预报模型,解决了BP神经网络模型预报精度低、在误差平坦区收敛速度慢的问题。该方法利用Gaussian函数改进BP神经网络梯度下降算法,针对模型不同权重与阈值设定不同学习率,对各参数进行针对性优化,能够有效加速BP神经网络模型训练效率;针对模型在误差平坦区收敛速度慢的问题,通过Gaussian函数增大梯度下降算法在误差平坦区的学习率,控制梯度下降算法在误差较大时的学习率,能够有效加速BP神经网络模型在误差平坦区的收敛速度。以福州市晋安区6个河道水位测站为研究对象,构建GABP神经网络河道水位预报模型进行城市内河水位预报,并探讨不同降雨输入形式对河道水位预报精度的影响。结果表明:GA-BP神经网络能够有效提升BP神经网络在误差平坦区的收敛速度与模型预报精度,试验集预报纳什效率系数(NSE)均在0.8以上,能够将预报峰值水位相对误差控制在5%以内,其中降雨以小时降雨量形式输入能够将预报NSE提升至0.9以上。研究表明采用Gaussian函数改进BP神经网络模型能够有效提升模型预报精度,对提升城市河道水位预报具有重要意义。
The prediction of urban river water level is of great significance to the risk management of urban waterlogging.Traditional numeri⁃cal simulation models have low computational efficiency and are unable to perform real-time calculations.In response to the above issues,this paper proposes a channel water level prediction model based on Gaussian function improved BP neural network,which solves the prob⁃lems of low prediction accuracy and slow convergence speed in error flat areas of the BP neural network model.This method utilizes the Gaussian function to improve the gradient descent algorithm of the BP neural network,sets different learning rates for different weights and threshold values of the model,and optimizes each parameter accordingly,which can effectively accelerate the training efficiency of the BP neural network model.In response to the problem of slow convergence speed of the model in error flat areas,the paper uses the Gaussian function to increase the learning rate of the gradient descent algorithm in error flat areas,and to control the learning rate of the gradient de⁃scent algorithm when the error is serious,which can effectively accelerate the convergence speed of the BP neural network model in error flat areas.This paper takes 6 river water level measurement stations in Jin’an District,Fuzhou City as the research object,constructs a GA-BP neural network river water level prediction model for urban river water level prediction,and explores the impact of different rainfall input forms on the accuracy of river water level prediction.The results show that the GA-BP neural network can effectively improve the conver⁃gence speed and model prediction accuracy of the BP neural network in error flat areas.The Nash Efficiency Coefficient(NSE)of the experi⁃mental set prediction is above 0.8,and the relative error of the predicted peak water level can be controlled within 5%.The input of rainfall in the form of hourly rainfall can increase the predicted NSE to over 0.9.Research has shown that using Gaussian function to improve the BP neural network model can effectively improve the prediction accuracy of the model,which is of great significance to improving urban river wa⁃ter level prediction.
作者
蒋双林
王超
陈阳
廖卫红
JIANG Shuang-lin;WANG Chao;CHEN Yang;LIAO Wei-hong(Research Center of Fluid Machinery Engineering and Technology,Jiangsu University,Zhenjiang 212013,Jiangsu Province,China;China Institute of Water Resources and Hydropower Research,Beijing 100038,China;College of Resources and Civil Engineering,Northeastern University,Shenyang 110819,Liaoning Province,China)
出处
《中国农村水利水电》
北大核心
2024年第1期109-116,共8页
China Rural Water and Hydropower
基金
“十四五”国家重点研发计划课题(2022YFC3800102)。
作者简介
蒋双林(1999-),男,硕士研究生,主要从事城市洪涝预报与数值模拟仿真研究。E-mail:JiangShuanglin1999@163.com;通讯作者:王超(1989-),男,高级工程师,博士,主要从事流域水资源调度与智慧水利方面的研究。E-mail:wangchao@iwhr.com。
