摘要
结构健康监测系统中通常存在大量的异常监测数据,为保证数据的完整性和可用性,有必要对异常监测数据进行修复。大多数基于深度学习对异常数据进行修复的研究通常使用单输入维度和单向预测的方法搭建模型。提出一种基于门控循环(gated recurrent unit,GRU)神经网络的结构异常监测数据修复方法,该方法充分利用深度学习神经网络适合处理复杂非线性映射问题的优势,并对GRU神经网络进行了优化与重构。利用结构温度、时序先后相关性优化神经网络的输入和输出构造,并提出了利用异常数据前后时间段的信息进行双向序列预测的方法提升数据预测和修复精度。最后,利用某古城墙的应变、裂缝与温度监测数据进行方法验证,采用重构后的GRU神经网络模型对异常数据序列进行修复,并与长短时记忆(long and short-term memory,LSTM)神经网络和反向传播(back propagation,BP)神经网络的修复精度进行比较。结果表明,相比单输入维度、单向预测的网络模型,重构后的GRU神经网络的预测精度大幅提高,且显著优于LSTM神经网络和BP神经网络。异常数据序列修复后,应变和裂缝宽度等结构响应与结构温度的线性相关性大幅增强。该方法对具有温度相关性的结构监测数据具有良好的修复能力。
There are usually large amounts of abnormal monitoring data in structural health monitoring system.In order to ensure integrity and availability of data,it is necessary to restore abnormal monitoring data.Most studies based on deep learning to restore abnormal data usually use single-input dimension and one-way prediction methods to build models.Here,a structural abnormal monitoring data restoring method based on gated recurrent unit(GRU)neural network was proposed.The proposed method could make full use of advantages of deep learning neural network to deal with complex nonlinear mapping problems,and do optimization and reconstruction for GRU neural network.Input and output structures of GRU neural network were optimized by using structure temperature and sequence correlation,and the method with abnormal data information in previous time period and that in subsequent time period doing bidirectional sequence prediction was proposed to improve the accuracy of data prediction and restoration.Finally,the proposed method was verified by using strain,crack and temperature monitoring data of an ancient city wall,and the reconstructed GRU neural network model was used to repair abnormal data sequence,and its repair accuracy was compared with those of long-short term memory(LSTM)neural network and back propagation(BP)neural network.The results showed that compared with single-input dimension and one-way prediction network models,the prediction accuracy of the reconstructed GRU neural network is significantly improved,and it is also significantly better than those of LSTM neural network and BP one;after restoring abnormal data sequence,the linear correlation between structural responses,such as,strain and crack width and structural temperature is largely enhanced;the proposed method has good restoring ability for temperature dependent structural monitoring data.
作者
鞠翰文
邓扬
李爱群
JU Hanwen;DENG Yang;LI Aiqun(School of Civil&Transportation Engineering,Beijing University of Civil Engineering and Architecture,Beijing 100044,China;Beijing Energy Conservation&Sustainable Urban and Rural Development Provincial and Ministry Co-construction Collaboration Innovation Center,Beijing University of Civil Engineering and Architecture,Beijing 100044,China)
出处
《振动与冲击》
EI
CSCD
北大核心
2023年第9期328-338,共11页
Journal of Vibration and Shock
基金
国家自然科学基金(51878027)
北京市教委青年拔尖人才培育计划(CIT&TCD201904060)
北京建筑大学基本科研业务费(X20174,X21073)。
关键词
结构健康监测
数据修复
深度学习
神经网络
温度
structural health monitoring
data restoration
deep learning
neural network
temperature
