摘要
大坝变形通常具有很强的波动性,传统方法往往不能很好地捕捉这种非线性关系,进而影响预测精度。为此,提出一种基于奇异值分解(SVD)和长短期记忆神经网络(LSTM)的大坝时序变形预测框架,旨在提升大坝变形的预测精度。首先,通过汉克尔矩阵的构建将原始变形序列分解为一系列较为规律的子序列;然后针对各分量建立相应的LSTM模型;最后将各模型的输出序列进行重构,从而得到最终的变形预测值。分析表明,SVD方法能够有效降低原始序列的非线性,同时,LSTM能够有效捕捉时间序列前后的非线性关系,得到令人满意的预测结果。与传统方法相比,SVD-LSTM的预测性能最优,为大坝安全系统的构建提供了新思路。
Dam deformation is usually highly volatile,and traditional methods often fail to capture nonlinear relationships well,which in turn affects prediction accuracy.This paper proposes a dam time-series deformation prediction framework based on Singular Value Decomposition(SVD)and Long Short-term Memory Neural Network(LSTM),aiming to improve the prediction accuracy of dam deformation.First,the original deformation sequence is decomposed into a series of more regular subsequences by the construction of Hankel matrix.Then the corresponding LSTM models are built for each component.Finally,the output sequences of each model are reconstructed to obtain the final deformation prediction values.Analysis shows that the SVD method can effectively reduce the nonlinearity of original sequence,while the LSTM can effectively capture the nonlinear relationship of time series and obtain satisfactory prediction results.Compared with traditional methods,the prediction performance of SVD-LSTM is optimal,which provides a new idea for the construction of dam safety system.
作者
李相如
苏超
袁荣耀
LI Xiangru;SU Chao;YUAN Rongyao(College of Water Conservancy and Hydropower Engineering,Hohai University,Nanjing 210024,Jiangsu,China)
出处
《水力发电》
CAS
2024年第6期67-71,共5页
Water Power
基金
国家自然科学基金资助项目(51579089)。
关键词
大坝变形预测
时间序列重构
奇异值分解
LSTM
非平稳
dam deformation prediction
time series reconstruction
singular value decomposition
LSTM
non-stationary