基于改进卷积神经网络的闸门结构响应数据修补方法  

Response Data Repair Method of Gate Structure Based on Improved Convolutional Neural Network

作　　者：谷奇丰 张钰奇 张文恒 陈栋[1] 李成[1] GU Qi-feng;ZHANG Yu-qi;ZHANG Wen-heng;CHEN Dong;LI Cheng(School of Mechanical and Power Engineering,Zhengzhou University,Zhengzhou 450001,China)

机构地区：[1]郑州大学机械与动力工程学院,河南郑州450001

出　　处：《水电能源科学》2024年第10期123-126,38,共5页Water Resources and Power

基　　金：国家自然科学基金项目(52175153);河南省水下智能装备重点实验室开放基金(ZT22064U)。

摘　　要：针对结构健康监测过程中数据缺失影响结构安全运行的问题,提出了一种基于改进卷积神经网络的数据补全方法,通过利用矩阵补全的方法实现了多通道数据的同时修补。首先将缺失信号经过傅里叶变换得到基础矩阵,并利用矩阵分解出的实部和虚部作为训练的输入,与L1-CNN神经网络作内积。采用基系数作为卷积运算的基础,并经过原始信号与修补信号进行对比更新,在更新卷积核同时应用L1正则化,得到修补后的信号。最后,以溢洪道水工闸门结构的加速度传感器信号为例进行数据修补,验证所提方法合理性,结果表明所提数据补全方法能够修复缺失率为30%以内的信号且不失真;修复缺失率为40%的信号,发现部分区域呈现失真状态。与其他方法对比分析表明,使用该方法补全的数据与原始数据具有最高拟合度,预测趋势接近完整数据。Aiming at the problem that the missing data affects the safe operation of structures in the process of structural health monitoring,a data completion method based on improved convolutional neural network is proposed.By using the matrix completion method,the multi-channel data is repaired simultaneously.Firstly,the missing signal is obtained by Fourier transform,and the real and imaginary parts of the matrix are used as the input of training,and the inner product is made with L1-CNN neural network.The base coefficient is used as the basis of the convolution operation,and the original signal and repair signal are compared and updated,and L1 regularization is applied to update the convolution kernel,and the repaired signal is obtained.Finally,the acceleration sensor signal of spillway hydraulic gate structure is taken as an example to verify the rationality and effectiveness of the proposed method.The results show that the proposed data completion method can repair the signal with a missing rate of less than 30%without distortion.The signal with 40%missing rate was repaired,and some areas were found to be distorted.Through comparison and analysis with other methods,the data completed by this method has the highest fitting degree with the original data,and the forecast trend is close to the complete data.

关 键 词：结构健康监测 深度学习 数据修补 神经网络 结构响应 

分 类 号：TV663.2[水利工程—水利水电工程]