基于深度学习的桥梁健康监测数据异常诊断

文章针对桥梁健康监测系统中存在部分监测数据异常问题, 将时频分析和深度学习相结合, 用于监测数据的异常诊断。首先根据异常类型对数据进行划分和标记, 利用时频分析法将一维桥梁加速度数据转换为时频图像, 制备用于构建和训练深度神经网络模型的图像数据库; 然后利用深度学习框架搭建基于卷积神经网络的数据异常诊断模型, 标记过的数据图像组成的训练集和验证集将被随机选择并输入至模型中, 通过反向传播机制和 Adam 优化算法更新和优化模型权重参数, 使用批标准化、数据增强等方法提高模型准确率和泛化能力; 最后对模型诊断结果进行统计并绘制出监测数据异常分布图。结果表明, 文章提出的模型对监测数据异常诊断准确率为 96.69%, 实现了桥梁监测数据异常快速定位, 与传统深度学习模型相比具有更好的稳定性和识别性能。该研究成果可用于桥梁健康监测系统的设计之中。

Aiming at the problem of some abnormal monitoring data in the bridge health monitoring system, this paper combines time-frequency analysis and deep learning for anomaly diagnosis of monitoring data. Firstly, the data are divided and labelled according to the anomaly type, and the time-frequency analysis method is used to convert the one-dimensional bridge acceleration data into time-frequency images, and an image database is prepared for building and training deep neural network models. Then, the deep learning framework is used to build a data anomaly diagnosis model based on convolutional neural network (CNN), the training set and validation set composed of labelled data images are randomly selected and input into the model, the model weight parameters are updated and optimized through the backpropagation mechanism and Adam optimization algorithm, and batch normalization and data enhancement methods are used to improve the accuracy and generalization ability of the model. Finally, the model diagnosis results are statistically analyzed and the anomaly distribution map of the monitoring data is drawn. The results show that the proposed model has an accuracy rate of 96.69% for anomaly diagnosis of monitoring data, which realizes the rapid localization of abnormal monitoring data, and has better stability and recognition performance than the traditional deep learning model. The results of this paper can be used in the design of bridge health monitoring systems.