基于MLR-DE-LSTM的大坝变形串联组合预测模型  

Dam Deformation Serial Combined Prediction Model Based on MLR-DE-LSTM

作　　者：刘天翼 艾星星 张九丹 LIU Tian-yi;AI Xing-xing;ZHANG Jiu-dan(East Line Smart Water Of China South-to-North Water Diversion Corporation Limited,Beijing 100000,China;College of Hydraulic Science and Engineering,Yangzhou University,Yangzhou 225009,Jiangsu Province,China;China South-to-North Water Transfer Group Middle Route Co.,LTD.Tianjin branch,Tianjin 300380,China)

机构地区：[1]南水北调东线智能水务(北京)有限公司,北京100000 [2]扬州大学水利科学与工程学院,江苏扬州225100 [3]中国南水北调集团中线有限公司天津分公司,天津300380

出　　处：《中国农村水利水电》2025年第2期207-212,共6页China Rural Water and Hydropower

摘　　要：为了解决单一模型在大坝变形预测中可能带来的信息损失问题,将差分进化算法(DE)用于长短期记忆神经网络(LSTM)模型的参数优化,并结合多元线性回归(MLR)模型建立MLR-DE-LSTM串联组合模型。基于某重力坝的水平位移原型监测数据,对该模型进行了验证。结果表明,DE算法可以有效提高LSTM模型的预测精度,LSTM模型可以有效挖掘MLR模型尚未完全解释的信息。相较于单一模型,组合模型在预测位移数据时具有更高的准确度和稳定性,组合模型在充分利用数据信息方面具有更大优势。研究结果为提高大坝变形预测精度提供了参考价值。In order to solve the problem of information loss that may be brought by a single model in dam deformation prediction,differential evolutionary algorithm(DE)is used for the parameter optimisation of the long-and short-term memory neural network(LSTM)model and combined with the multiple linear regression(MLR)model to establish a tandem combination model of MLR-DE-LSTM.The model is validated based on the horizontal displacement prototype monitoring data of a gravity dam.The results show that the DE algorithm can effectively improve prediction accuracy of the LSTM model,and the LSTM model can effectively mine the information that has not been fully explained by the MLR model.Compared with a single model,the combined model has higher accuracy and stability in predicting displacement data,and the combined model has greater advantages in making full use of the data information.The results provide reference value for improving the accuracy of dam deformation prediction.

关 键 词：大坝变形 差分进化算法 长短期记忆神经网络 多元线性回归 组合模型 

分 类 号：TV698.1[水利工程—水利水电工程]