摘要
已有混凝土坝变形监控模型较少考虑多测点序列在时空特征上的动态关联,进而限制了变形的预测精度。鉴于此,构建了一种基于聚类分析、长短期记忆(LSTM)网络和注意力机制的混凝土坝空间变形动态预测模型,该模型通过合理引入动态关联因子,提升目标测点的变形预测精度。以某重力坝多个测点的水平位移监测数据为例,通过与其他3种基于LSTM的模型进行对比,验证了该模型的有效性以及空间动态关联因子的重要性;进一步针对不同测点,通过两种可解释性机器学习算法,将多重影响因素对模型输出结果的贡献程度进行了量化,其排序结果符合坝工先验知识。研究成果可为促进大坝变形智能监控模型的工程应用提供参考。
The existing deformation monitoring models of concrete dams rarely consider the dynamic spatiotemporal correlation of multiple measurement points,thus limiting the prediction accuracy of deformation.To overcome this shortcoming,this paper constructs a dynamic prediction model of spatial deformation of concrete dam based on cluster analysis,long short-term memory(LSTM)network and attention mechanism,which can improve the prediction accuracy of deformation of target measurement point by introducing dynamic correlation factors reasonably.Taking the horizontal displacement monitoring data of several measurement points of a gravity dam as an example,the effectiveness of the model and the importance of spatial dynamic correlation factors are verified by comparing with three other LSTM-based models.Furthermore,the contribution degree of multiple factors to the model output is quantified by two interpretable machine learning algorithms for different measurement points,and the ranking results are consistent with the prior knowledge of dam engineering.The research results can provide reference for promoting the engineering application of intelligent dam deformation monitoring model.
作者
艾星星
刘兴阳
仇建春
缪久兵
黄海燕
何海瑞
崔家浩
AI Xing-xing;LIU Xing-yang;QIU Jian-chun;MIAO Jiu-bing;HUANG Hai-yan;HE Hai-rui;CUI Jia-hao(College of Hydraulic Science and Engineering,Yangzhou University,Yangzhou 225100,Jiangsu Province,China;Nantong Haimen District Haimen Harbour New District Water Conservancy Service Station,Nantong 226155,Jiangsu Province,China;Yunnan Water Resources and Hydropower Vocational College,Kunming 650499,Yunnan Province,China;Ningbo Reservoir Management Centre,Ningbo 315000,Zhejiang Province,China)
出处
《中国农村水利水电》
北大核心
2025年第12期47-54,共8页
China Rural Water and Hydropower
基金
国家自然科学基金项目(52309173)
国家大坝安全工程技术研究中心开放基金(CX2023B01)
扬州市自然科学基金项目(YZ2024166)。
关键词
混凝土坝变形
动态时空关联
聚类分析
长短期记忆网络
注意力机制
可解释性分析
deformation of concrete dam
dynamic spatiotemporal correlation
cluster analysis
long short-term memory network
attention mechanism
interpretability analysis
