基于稀疏互信息的建筑振动信号特征增强  

Constructions Vibration Signal Anomaly Detection Based on Sparse Mutual Information

作　　者：孙永伟 SUN Yong-wei(Shandong Hi-Speed Laigang Green Construction Development Co.,Ltd.,Qingdao 266000,China)

机构地区：[1]山东高速莱钢绿建发展有限公司,青岛266000

出　　处：《价值工程》2024年第16期110-113,共4页Value Engineering

摘　　要：运行中的建筑振动监测记录不可避免地包含传感器故障、系统错误或环境影响引起的各种异常,导致振动信号特征微弱、数据不平衡,难以实现有效的异常检测。针对以上问题,本文提出基于稀疏互信息的建筑振动信号特征增强方法,以生成对抗网络为基础,首先以峭度作为阈值标签建立针对振动信号的稀疏向量,对向量构成的字典矩阵通过牛顿最优化方法求解最优特征向量,然后在输入层中以故障信息和峭度标签信息作为隐含编码构建隐变量信息空间,最后在全连接层中以互信息最大化为指标迭代生成故障特征,输出符合真实分布的生成信号特征。结果表明:该方法的信号特征拟合率达到97.07%,实现了对建筑振动信号数据的故障特征增强与有效异常检测。The construction vibration monitoring records in operation inevitably contain various anomalies caused by sensor failures,system errors,or environmental impacts,resulting in weak vibration signal characteristics and unbalanced data,making it difficult to achieve effective anomaly detection.To address the above problems,this paper proposes a sparse mutual information-based construction vibration signal anomaly detection method,based on the generative adversarial network,firstly,the sparse vectors of vibration signals are established with the crag as the threshold label,and the optimal feature vectors are solved by the Newton's optimization method for the dictionary matrix formed by the vectors,and then,in the input layer,the hidden variable information space is constructed by using the fault information and crag labeling information as the hidden coding.Then,in the input layer,the fault information and the cliff label information are used as the implicit coding to construct the hidden variable information space,and finally,the fault features are iteratively generated in the fully-connected layer with the maximization of mutual information as the index,and the generated signal features conforming to the real distribution are output.The results show that the signal feature fitting rate of this method reaches 97.07%,which realizes the fault feature enhancement and effective anomaly detection of building vibration signal data.

关 键 词：建筑 振动信号 生成对抗网络 异常检测 特征增强 

分 类 号：TU311.3[建筑科学—结构工程]