基于改进聚类方法的大坝安全监测算法  

Improved Clustering-Based Algorithm for Dam Safety Monitoring

作　　者：李东明 聂一丹 晁阳 齐慧君[1] 林潮宁 LI Dongming;NIE Yidan;CHAO Yang;QI Huijun;LIN Chaoning(College of Water Conservancy and Hydropower Engineering,Hohai University,Nanjing 210024,Jiangsu,China;Sinohydro Bureau 11 Co.,Ltd.,Zhengzhou 450001,Henan,China)

机构地区：[1]河海大学水利水电工程学院,江苏南京210024 [2]中国水利水电第十一工程局有限公司,河南郑州450001

出　　处：《水力发电》2025年第4期97-103,110,共8页Water Power

基　　金：国家重点研发计划(2022YFC3005403);国家自然科学基金资助项目(52309151);水利部水库大坝安全重点实验室开放研究基金(YK323007)。

摘　　要：为提升大坝安全预测过程中离群值检验与风险预测的精度与效率,对K-Shape聚类方法进行改进,将空间距离纳入K-Shape聚类评价指标,改进了大坝监测时间序列聚类算法,并将聚类中心线与统计模型方法结合,构建基于聚类结果的离群值判断方法。基于聚类结果,结合长短期记忆网络算法(LSTM)对离群值以及缺失值的填补,实现了对大坝工作性态趋势的预测。将该方法应用于云南某大坝的安全监测中,对大坝的位移与温度监测数据时空序列开展了检验与预测。结果表明,空间距离对于聚类结果影响明显,改进的聚类算法在异常值识别性能上相较于传统算法有一定提升,多测点预测算法精度更高,证明通过空间距离纳入聚类算法具有一定的实用价值。To enhance the accuracy and efficiency of outlier detection and risk prediction in dam safety forecasting,the K-Shape clustering method is improved by incorporating spatial distance into the evaluation metrics,which refines the time series clustering algorithm used for dam monitoring,and additionally,the clustering center line is integrated with statistical modeling techniques to develop a method for outlier identification based on clustering results.By leveraging the clustering outcomes and utilizing Long Short-Term Memory(LSTM)network,the issues of outlier detection and missing value imputation are effectively addressed,and the prediction of trends in dam operational behavior is enabled.This method is applied to the safety monitoring of a dam in Yunnan Province for the tests and predictions on the spatiotemporal sequences of displacement and temperature monitoring data.The results indicate that the spatial distance significantly influences clustering outcomes,and the improved clustering algorithm demonstrates enhanced performance in outlier detection compared to traditional methods.Furthermore,the multi-point prediction algorithm exhibits higher accuracy,confirming the practical value of incorporating spatial distance into clustering algorithms.

关 键 词：大坝安全监测 精度 时空序列聚类 K-Shape聚类方法 长短期记忆网络(LSTM) 

分 类 号：TV698.1[水利工程—水利水电工程]