改进小波阈值与优化BiLSTM组合的大坝变形预测方法  

Improved wavelet thresholding combined with optimized BiLSTM for dam deformation prediction

作　　者：石佳晨 岳春芳[1,2] 朱明远 皮李浪 SHI Jiachen;YUE Chunfang;ZHU Mingyuan;PI Lilang(College of Water Conservancy and Civil Engineering,Xinjiang Agricultural University,Urumqi 830052,China;Xinjiang Key Laboratory of Hydraulic Engineering Security and Water Disasters Prevention,Urumqi 830052,China;Xinjiang Institute of Water Resources and Hydropower Research,Urumqi 830049,China;College of Water Conservancy and Hydraulic Engineering,Hohai University,Nanjing 210098,China)

机构地区：[1]新疆农业大学水利与土木工程学院,乌鲁木齐830052 [2]新疆水利工程安全与水灾害防治重点实验室,乌鲁木齐830052 [3]新疆水利水电科学研究院,乌鲁木齐830049 [4]河海大学水利水电学院,南京210098

出　　处：《水力发电学报》2024年第7期97-108,共12页Journal of Hydroelectric Engineering

基　　金：新疆维吾尔自治区重大科技专项项目(2022A02003-5);新疆农业大学研究生教育教学改革研究项目专业学位研究生课程案例库建设(xjaualk-yjs-2022010);新疆维吾尔自治区水利科技专项资助项目(XSKJ-2023-23)。

摘　　要：变形是反映大坝结构状态变化的重要指标。由于变形数据的非线性特点和其背后复杂的机理,提升变形的预测精度对大坝安全及结构控制具有重要意义。为此,基于融合建模理念提出了一种组合的大坝变形预测方法,该方法结合了改进小波阈值去噪与鹈鹕优化算法(POA)优化的双向长短期神经网络(BiLSTM)。首先,采用改进小波阈值去噪法对变形实测数据序列进行处理;其次,通过POA搜索最优超参数组合用于优化BiLSTM模型;最后,基于最优超参数下的BiLSTM模型进行大坝变形预测。工程实例表明,改进小波阈值法具有更好的去噪效果,POA-BiLSTM能够准确预测大坝变形。在最终测试集上平均MAE、MAPE、RMSE、R^(2)分别为0.244、0.041、0.301、0.906。相较于其他方法,表现出更高的预测准确性和稳健性,可为大坝变形监测提供参考。Deformation serves as a crucial indicator of the structural changes of dams.Enhancing the prediction accuracy of dam deformation is of paramount significance for the safety and structural control of dams,due to the nonlinear characteristics of deformation data and the underlying intricate mechanism.This paper develops a combined approach for dam deformation prediction based on the integrated modeling concept,integrating an improved wavelet threshold denoising and a Pelican Optimization Algorithm(POA)optimized Bidirectional Long Short-Term Memory(BiLSTM)network.First,the deformation measurement data sequence is processed using an improved wavelet threshold denoising method;then,POA is used to search for the optimal hyperparameter combination to optimize the BiLSTM model;finally,dam deformation prediction is conducted based on the BiLSTM with the optimal hyperparameters.Engineering case studies demonstrate that this improved wavelet threshold method produces superior denoising effects,and POA-BiLSTM gives a satisfactory accuracy for dam deformation prediction.And on the ultimate test set,it has achieved the average MAE,MAPE,RMSE,and R^(2) of 0.244,0.041,0.301,and 0.906,respectively.Compared to other methods,it exhibits higher predictive accuracy and robustness,offering valuable insight for dam deformation monitoring.

关 键 词：改进小波阈值 鹈鹕优化算法 双向长短期神经网络 去噪 变形预测 

分 类 号：TV698.1[水利工程—水利水电工程]