基于SSA-LSTM-GF的混凝土坝变形预测方法  

Deformation prediction method of concrete dam based on SSA-LSTM-GF

作　　者：周兰庭[1] 邓思源 柳志坤 龚云柱 ZHOU Lanting;DENG Siyuan;LIU Zhikun;GONG Yunzhu(College of Water Conservancy and Hydraulic Engineering,Hohai University,Nanjing 210098,China;Kinetic Energy Conversion Promotion Office,Qingdao Development and Reform Commission,Qingdao 266000,China;Qingdao Economic Development Research Institute,Qingdao 266000,China;Huaneng Lancang River Hydropower Inc.,Kunming 650214,China)

机构地区：[1]河海大学水利水电学院,江苏南京210098 [2]青岛市发展和改革委员会动能转换推进处,山东青岛266000 [3]青岛市经济发展研究院,山东青岛266000 [4]华能澜沧江水电股份有限公司,云南昆明650214

出　　处：《河海大学学报（自然科学版）》2023年第2期73-80,149,共9页Journal of Hohai University(Natural Sciences)

基　　金：国家自然科学基金(51209078,51739003)。

摘　　要：针对混凝土坝变形分析预测的复杂性,应用相空间重构思想和融合建模理念,提出了一种基于SSA-LSTM-GF的混凝土坝变形分析预测方法。SSA-LSTM-GF方法利用奇异谱分析法(SSA)将变形实测数据序列分解为趋势分量、周期分量和剩余分量,并将剩余分量视为噪声分量予以剔除;采用长短期记忆神经网络(LSTM)模型和高斯拟合(GF)算法分别进行周期分量和趋势分量的分析预测,并将二者结果进行叠加重构,得到最终预测结果。实例验证结果表明,SSA可以达到较好的数据分解和消噪效果,LSTM模型针对周期分量的预测性能优越,GF算法能够很好地实现趋势分量的拟合预测和部分信息的挖掘提取,LSTM模型和GF算法的成果重构效果良好,SSA-LSTM-GF方法具有一定的可行性和应用价值。In view of the complexity of deformation analysis and prediction for the concrete dam,a deformation analysis and prediction method of concrete dam based on SSA-LSTM-GF is proposed by using the idea of phase space reconstruction and fusion modeling.Firstly,the measured data sequence of deformation is decomposed into the trend component,the periodic component and the remaining component by singular spectrum analysis(SSA),and the remaining component is removed as noise;then,long short-term memory(LSTM)model and Gaussian fitting(GF)algorithm are used to analyze and predict the periodic component and the trend component respectively;finally,the two results are reconstructed to obtain the final prediction results.The example analysis shows that SSA can achieve good data decomposition and denoising effect;LSTM model has superior prediction performance for periodic components;GF algorithm can well realize the fitting and prediction of trend components and the mining and extraction of some information;the reconstruction effect of the results of LSTM model and GF algorithm is good.In conclusion,the deformation analysis and prediction method based on SSA-LSTM-GF has certain feasibility and research application value.

关 键 词：混凝土坝 变形预测 序列分解 奇异谱分析法 LSTM模型 GF算法 

分 类 号：TV698.1[水利工程—水利水电工程]