考虑子序列特征的大坝位移组合预测模型  

Multi-scale Dam Displacement Prediction Model Considering Subsequence Characteristics

作　　者：林宏恩 赵二峰[1,2] 刘峰 宋桂华 LIN Hongen;ZHAO Erfeng;LIU Feng;SONG Guihua(National Engineering Research Center of Water Resources Efficient Utilization and Engineering Safety,Hohai University,Nanjing 210024,Jiangsu,China;Yunnan Key Laboratory of Water Conservancy and Hydropower Engineering Safety,Kunming 650051,Yunnan,China;Shanghai Investigation,Design&Research Institute Co.,Ltd.,Shanghai 200335,China)

机构地区：[1]河海大学水资源高效利用与工程安全国家工程研究中心,江苏南京210024 [2]云南省水利水电工程安全重点实验室,云南昆明650051 [3]上海勘测设计研究院有限公司,上海200335

出　　处：《水力发电》2025年第3期113-118,共6页Water Power

基　　金：国家自然科学基金资助项目(52079046);云南省水利水电工程安全重点实验室开放课题基金(202302AN360003);上海勘测设计研究院有限公司科研项目(2021SD(8)-2009)。

摘　　要：针对多尺度分析模型处理子序列时,预测模式单一、泛化能力不足、易出现偏差等,提出了基于奇异谱分析与Bagging集成学习的大坝位移多尺度预测模型。首先,利用奇异值分解对实测数据进行分解,得到趋势项、周期项等子序列,考虑到周期项对时间依赖性较强,将其作为单时间序列进行预测以消除非关键因素的影响。其次,利用Bagging集成学习,结合支持向量机与随机森林模型,构建大坝位移组合预测模型。在此基础上,将趋势项和周期项的预测结果累加后得到大坝位移预测结果。应用实例表明,所建模型能够充分挖掘实测数据蕴含的趋势性和周期性变化物理机制,为大坝长效服役性态诊断提供了新思路。A multi-scale prediction model of dam displacement based on singular spectrum analysis and Bagging ensemble learning is proposed to solve the problem of single prediction mode,insufficient generalization ability and easy deviation when processing the subsequences of multi-scale analysis model.Firstly,the measured data is decomposed by singular value decomposition to obtain subsequences such as trend term and period term,and considering that the period term is highly dependent on time,the period term is forecasted as a single time series to eliminate the influence of non-critical factors.Secondly,the Bagging ensemble learning,support vector machine and random forest model are used to construct a combined prediction model of dam displacement.On this basis,the predicted results of the trend term and the period term are added up to get the predicted results of dam displacement.The application example shows that the model can fully explore the physical mechanism of trend and periodic change contained in measured data,and provides a new idea for the long-term service behavior diagnosis of dams.

关 键 词：多尺度 奇异谱分析 Bagging集成学习 大坝位移预测 

分 类 号：TV698.11[水利工程—水利水电工程]