东篱科研大数据发现系统（DRDS）

位置：成果数据库 > 期刊 > 期刊详情页

基于LMBP神经网络的土石坝渗流压力预测

ISSN号：1000-1379
期刊名称：《人民黄河》
时间：0
分类：TV698.1[水利工程—水利水电工程]
作者机构：[1]南京水利科学研究院,江苏南京210029, [2]水利部大坝安全管理中心,江苏南京210029, [3]江苏省水利厅,江苏南京210029
相关基金：水利部公益性行业科研专项（201501033）;国家重点研发计划项目（2016YFC0401600）;江苏省水利科技项目（2015010）;广西水利厅科技项目（201614）.

作者：吴云星[1,2], 周贵宝[3], 谷艳昌[1,2], 庞琼[1,2], 王宇[1]

关键词：土石坝, 渗流压力, LMBP神经网络, 预测精度, earth and rockfill dam, seepage pressure, LMBP neural network, prediction accuracy

中文摘要：

渗流压力是反映大坝工作状态的重要物理量,对渗流压力进行预测分析可以及时了解大坝渗流状况和趋势。为克服标准BP算法收敛速度慢、泛化能力弱和计算量大等不足,引入LM算法优化标准BP神经网络的权值和阈值,提高BP神经网络对土石坝渗流压力的预测效果。根据渗流分析,给出了渗流压力的统计模型,由统计模型选取上下游水位、降雨和时效作为神经网络输入层因子,以渗流压力作为输出层因子,建立了3层LMBP神经网络大坝渗流压力预测模型。利用MATLAB进行了多组仿真试验,确定了使本次渗流压力预测效果更好的训练样本数据量区间。以渗流压力实测数据及同期库水位和降雨资料作为训练样本,在选取适当数据量的训练样本的基础上,运用LM算法对BP网络进行训练,利用测试样本对训练好的神经网络进行测试。将同结构的LMBP神经网络和标准BP神经网络应用于某土石坝渗流压力的预测中,应用结果表明,LMBP神经网络收敛速度更快、拟合和预测精度更高,在土石坝渗流压力分析和预测应用方面是可行的。

英文摘要：

Seepage pressure is an important physical quantity to reflect the working condition of dam and the prediction and analysis of seep-age pressure can understand the situation and trend of dam seepage. In order to overcome the standard BP algorithm5 s defects,such as slow convergence rate,weak generalization ability and generous calculation and improve the prediction results of BP neural network of earth and rockfill dam seepage pressure,the introduction of LM （Leverberg Marquart） algorithm was proposed to optimize the weight and threshold of standard BP neural network. According to the seepage analysis, the statistical model of seepage pressure was given,selection of upstream and downstream water level,rainfall and aging as the neural network input factor,with seepage pressure as the output factor,a 3-layer LMBP neural network prediction model of dam seepage pressure was established. Multi group simulation experiments were carried out by MATLAB to deteiinine the training sample data quantity which made the seepage pressure prediction effect better. Seepage pressure measured data and the same period of reservoir water level and rainfall data as the training sample,on the basis of the selection of appropriate data amount of train-ing sample,using LM algorithm to train the network,then the testing samples were predicted by the trained neural network. The application results show that LMBP neural network model has a faster convergence rate and higher accuracy than that of standard BP neural network. It is feasible to analyze and predict the seepage pressure of earth and rockfill dam.

同期刊论文项目