基于BP神经网络-加权马尔科夫模型的泄水闸水平位移预测  

Prediction of horizontal displacement of sluice gate based on BP neural network-weighted Markov model

作　　者：丁倩 黄耀英[1] 谢同 李峰[1] 高磊 DING Qian;HUANG Yaoying;XIE Tong;LI Feng;GAO Lei(College of Hydraulic&Environmental Engineering,China Three Gorges University,Yichang 443002,China;Hubei Hanjiang Wangfuzhou Hydropower Co.,Ltd.,Xiangyang 430048,China)

机构地区：[1]三峡大学水利与环境学院,湖北宜昌443002 [2]湖北汉江王甫洲水力发电有限责任公司,湖北襄阳430048

出　　处：《水资源与水工程学报》2020年第6期187-193,共7页Journal of Water Resources and Water Engineering

基　　金：国家重点研发计划项目(2018YFC0407103);国家自然科学基金项目(51779130)。

摘　　要：针对传统变形统计模型和BP神经网络模型对水工建筑物变形预测精度欠佳的问题,建立了BP神经网络-加权马尔科夫模型。首先,采用均值-均方差法对BP神经网络模型拟合的相对误差序列进行状态分类,并检验状态序列的马氏性。然后计算各阶自相关系数和转移权重,利用加权和最大概率值预测未来的随机状态。最后以王甫洲水利枢纽泄水闸11#闸墩测点水平位移实测数据为例,分析比较逐步回归统计模型、BP神经网络模型和BP神经网络-加权马尔科夫模型的预测效果。结果表明:相比于逐步回归统计模型和BP神经网络模型,BP神经网络-加权马尔科夫模型的预测精度更高,说明BP神经网络-加权马尔科夫模型较为可靠。Aiming at the problem of poor prediction accuracy of traditional deformation statistical model and BP neural network model,this paper discussed the horizontal displacement prediction of BP neural network-weighted Markov model. First,the mean-mean square error method was used to classify the relative error sequence fitted by the BP neural network and to check the Markovity of the state sequence.Then,the autocorrelation coefficients and weights of each order were calculated,and the weighted and maximum probability values were used to predict the future random state. Finally,taking the measured horizontal displacement of the Gate Pier 11 of Wangfuzhou Water Control Project as an example,the prediction results of the stepwise regression statistical model,BP neural network model and BP neural network-weighted Markov model were compared. The results show that compared to the stepwise regression statistical model and BP neural network model,the BP neural network-weighted Markov model has higher prediction accuracy,which indicates that this model is more reliable.

关 键 词：水平位移预测 预测精度 BP神经网络 加权马尔科夫模型 马氏检验 

分 类 号：TV61[水利工程—水利水电工程]