基于VMD-BSO-BiLSTM的混凝土坝变形智能预测模型  

Intelligent Prediction Model for Concrete Dam Deformation Based on VMD-BSO-BiLSTM

作　　者：王正新 高剑峰 杨振亚 周明明 WANG Zhengxin;GAO Jianfeng;YANG Zhenya;ZHOU Mingming(Nanjing Water Resources Planning and Design Institute Co.Ltd.,Nanjing 210000,China;School of Water Resources and Hydropower,Hohai University,Nanjing 210098,China;Jiangsu Provincial Water Resources Survey,Design and Research Institute Co.Ltd.,Yangzhou 225127,Jiangsu China)

机构地区：[1]南京市水利规划设计院股份有限公司,南京210000 [2]河海大学水利水电学院,南京210098 [3]江苏省水利勘测设计研究院有限公司,江苏扬州225127

出　　处：《河南科学》2025年第2期241-249,共9页Henan Science

基　　金：国家重点研发计划课题(2018YFC1508603);2021年江苏省水利科技项目(2021068);2022年江苏省水利科技项目(2022011)。

摘　　要：混凝土坝的变形对环境荷载的反馈存在一定的滞后性,从而导致混凝土坝的变形具有较强的时效性。为了模拟环境荷载对大坝变形的时间效应,采用了双向长短时记忆智能学习算法(BiLSTM)对大坝变形进行双向学习预测。同时为了提高BiLSTM算法的计算精度,采用了变分模态分解算法(VMD)对变形序列进行模态分解以得到规律性较好的变形分量。通过BiLSTM训练各分量的映射网络,以此计算得到了各变形分量的预测值,将各分量的预测值相加得到了大坝变形的预测值。为了加强预测模型的自适应学习能力和模型的鲁棒性,采用天牛群优化算法(BSO)对模型进行了全局优化,从而构建了基于BSO优化的VMD-BiLSTM混凝土坝变形智能预测模型。结合工程案例可知,该变形预测模型的平均绝对百分比误差(MAPE)为3.41%,其精度水平能够满足大坝变形安全监控的需要,并且较VMD-BSO-LSTM、BSO-BiLSTM和BiLSTM模型,其MAPE相应降低了1.35%、2.11%和4.02%,显著地提高了预测精度。Due to the lag in the feedback of the deformation of concrete dams on environmental loads,the deformation of concrete dams exhibits strong timeliness.The conventional prediction model lacks information extraction on the lag and timeliness of deformation,and cannot reflect the actual deformation behavior of the dam well.To simulate the time effect of environmental loads on dam deformation,this paper uses the bidirectional learning algorithm BiLSTM to predict dam deformation through bidirectional learning.Since a more stable numerical sequence can improve the computational accuracy of the BiLSTM algorithm,the VMD algorithm is used to perform modal decomposition on the deformation sequence to obtain more regular deformation components.By using BiLSTM to train mapping networks for each deformation mode component,the predicted values of each deformation component are calculated.The predicted values of each component are added together to obtain the predicted values of dam deformation.To enhance the adaptive learning ability and robustness of the prediction model,the BSO algorithm is used to globally optimize the model,and a VMD-BiLSTM concrete dam deformation prediction model based on BSO optimization is constructed.According to the engineering case,the mean absolute percentage error(MAPE)of this deformation prediction model is 3.41%,and its accuracy level can meet the needs of dam deformation safety monitoring.Compared with VMD-BSO-LSTM,BSO-BiLSTM and BiLSTM models,its MAPE is correspondingly reduced by 1.35%,2.11%and 4.02%,significantly improving the prediction accuracy.This model provides new ideas and methods for deformation prediction of similar dams.

关 键 词：混凝土大坝 变形预测 BiLSTM BSO优化算法 VMD算法 

分 类 号：TV32[水利工程—水工结构工程]