基于预测与重构的桥梁异常状态检测模型

doi:10.13229/j.cnki.jdxbgxb.20240810

摘要/Abstract

摘要：

提出了一种时间特征与空间特征融合的桥梁异常状态检测模型TSSF-BADM，通过图注意力网络（GAT）和长短期记忆网络（LSTM）提取时间和空间的数据特征，将不同维度的特征通过门控循环单元（GRU）进行融合并捕获时间序列的顺序模式；融合后的数据再通过预测和重构模型进行联合优化，分别使用堆叠LSTM网络和变分自编码器（VAE）进行预测和重构；最后，使用峰值超过阈值（POT）方法对模型的预测和重建误差进行分析，获得阈值并进行异常检测，将超过异常阈值的样本视为异常样本。实验对比结果表明：本文模型在真实桥梁Z24上取得了良好效果，识别的准确率达到了0.986 8，识别延迟仅有0.008，均优于LSTM_VAE、MAD_GAN、OmniAnomaly等其他对比模型，能够有效进行桥梁异常状态检测，可为桥梁安全检测、预防性维护等提供决策。

关键词: 桥梁异常状态检测, 时空特征融合, 重构模型, 预测模型

Abstract:

This paper proposes a bridge anomaly detection model that integrates temporal and spatial features TSSF-BADM. The model utilizes graph attention networks （GAT） and long short-term memory （LSTM） networks to extract temporal and spatial data features. These multi-dimensional features are then fused using gated recurrent units （GRU） to capture sequential patterns in the time series. The fused data undergo joint optimization through prediction and reconstruction models， employing stacked LSTM networks and variational autoencoders （VAE） for prediction and reconstruction， respectively. Finally， the prediction and reconstruction errors of the model are analysed using the peak over threshold （POT） method to obtain the threshold and perform anomaly detection， and the samples exceeding the anomaly threshold are considered as anomalous samples. The experimental comparison results show that the model in this paper achieves good performance on the real bridge Z24， the accuracy of recognition reaches 0.986 8， and the recognition delay is only 0.008， which are all better than other comparative models such as LSTM_VAE， MAD_GAN， OmniAnomaly， etc.， and are able to effectively carry out the detection of the abnormal state of the bridge. It provides decision-making for bridge safety detection， preventive maintenance， etc.

Key words: bridge anomaly detection, integration of spatial and temporal features, reconstructed model, prediction model

中图分类号:

TP399

火久元,窦瑞翔,常琛,陈峰,张耀南. 基于预测与重构的桥梁异常状态检测模型[J]. 吉林大学学报(工学版), 2026, 56(2): 497-508.

Jiu-yuan HUO,Rui-xiang DOU,Chen CHANG,Feng CHEN,Yao-nan ZHANG. Prediction and reconstruction based anomaly condition detection model for bridges[J]. Journal of Jilin University(Engineering and Technology Edition), 2026, 56(2): 497-508.

图/表 16

图1

图2

图3

图4

图5

表1

实验参数设置"

超参数	数值
批大小	64
学习率	0.001
轮次	100
$γ$	0.8
阈值	0.028 29

表1

图6

表2

图7

图8

图9

图10

表3

图11

表4

表5

不同γ值的对比结果"

$γ$	Precision	Recall	F₁
0.4	0.973 2	0.968 7	0.971 0
0.6	0.975 3	0.953 8	0.964 4
0.8	0.986 8	0.980 8	0.983 7
1.0	0.970 1	0.914 7	0.941 6

表5

参考文献 32

[1]	交通运输部. 2022年交通运输行业发展统计公报[EB/OL]. [2024-03-31]. .
[2]	Entezami A, Sarmadi H, Behkamal B. A novel double-hybrid learning method for modal frequency-based damage assessment of bridge structures under different environmental variation patterns[J]. Mechanical Systems and Signal Processing, 2023, 201:No. 110676.
[3]	Domenico D. A review of vibration-based methods for damage detection in structures[J]. Journal of Structural Health Monitoring, 2018, 17: 3-35.
[4]	Gao T, Wang X, Li Z. Sensor-based methods for structural health monitoring of bridges: a review[J]. Measurement, 2018, 115: 243-261.
[5]	Abdeljaber O, Avci O, Kiranyaz M S, et al. 1-D CNNs for structural damage detection: verification on a structural health monitoring benchmark data[J]. Neurocomputing, 2018, 275: 1308-1317.
[6]	Cha Y J, Wang Z. Unsupervised novelty detection–based structural damage localization using a density peaks-based fast clustering algorithm[J]. Structural Health Monitoring, 2018, 17(2): 313-324.
[7]	Rafiei M H, Adeli H. A novel unsupervised deep learning model for global and local health condition assessment of structures[J]. Engineering Structures, 2018, 156: 598-607.
[8]	Silva M F, Santos A, Santos R, et al. Damage‐sensitive feature extraction with stacked autoencoders for unsupervised damage detection[J]. Structural Control and Health Monitoring, 2021, 28(5):No. e2714.
[9]	Soleimani B M H, Sepasdar R, Nasrollahzadeh K, et al. Toward a general unsupervised novelty detection framework in structural health monitoring[J]. Computer‐Aided Civil and Infrastructure Engineering, 2022, 37(9): 1128-1145.
[10]	Sony S, Gamage S, Sadhu A, et al. Vibration-based mul ticlass damage detection and localization using long short-term memory networks[J]. Structures Elsevier, 2022, 35: 436-451.
[11]	Li L, Morgantini M, Betti R. Structural damage assessment through a new generalized autoencoder with features in the quefrency domain[J]. Mechanical Systems and Signal Processing, 2023, 184: No.109713.
[12]	Fan J, Zhang K, Huang Y, et al. Parallel spatio-temporal attention-based TCN for multivariate time series prediction[J]. Neural Computing and Applications, 2023, 35(18): 13109-13118.
[13]	Hinton G E, Salakhutdinov R R. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(5786): 504-507.
[14]	Chen N, Tu H, Duan X, et al. Semisupervised anomaly detection of multivariate time series based on a variational autoencoder[J]. Applied Intelligence, 2023, 53(5): 6074-6098.
[15]	Dubey A K, Kumar A, García D V, et al. Study and analysis of SARIMA and LSTM in forecasting time series data[J]. Sustainable Energy Technologies and Assessments, 2021, 47: No.101474.
[16]	Ding C, Sun S, Zhao J. MST-GAT: a multimodal spatial-temporal graph attention network for time series anomaly detection[J]. Information Fusion, 2023, 89: 527-536.
[17]	Tang C, Xu L, Yang B, et al. GRU-based interpretable multivariate time series anomaly detection in industrial control system[J]. Computers & Security, 2023, 127: No.103094.
[18]	Lei T, Gong C, Chen G, et al. A novel unsupervised framework for time series data anomaly detection via spectrum decomposition[J]. Knowledge-Based Systems, 2023, 280:No. 111002.
[19]	Shan D, Yao K, Zhang X. Sequential learning network with residual blocks: incorporating temporal convolutional information into recurrent neural networks[J]. IEEE Transactions on Cognitive and Developmental Systems, 2023, 16(1): 396-401.
[20]	Lin M, Wu J, Meng J, et al. State of health estimation with attentional long short-term memory network for lithium-ion batteries[J]. Energy, 2023, 268: No.126706.
[21]	刘嫣然, 孟庆瑜, 郭洪艳, 等. 图注意力模式下融合高精地图的周车轨迹预测[J]. 吉林大学学报: 工学版, 2023, 53(3): 792-801.
	Liu Yan-ran, Meng Qing-yu, Guo Hong-yan, et al. Vehicle trajectory prediction combined with high definition map in graph attention mode[J].Journal of Jilin University(Engineering and Technology Edition), 2023, 53(3): 792-801.
[22]	Siffer A, Fouque P A, Termier A, et al. Anomaly detection in streams with extreme value theory[C]∥ Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Halifax, Canada, 2017: 1067-1075.
[23]	Peeters B, Roeck D. One year monitoring of the z24-bridge: environmental influences versus damage events[J]. Proceedings of the International Modal Analysis Conference, 2000, 2: 1570-1576.
[24]	Maeck J, Roeck D. Description of Z24 benchmark[J]. Mechanical Systems and Signal Processing, 2003, 17(1): 127-131.
[25]	Reynders E, Roeck D. Vibration-based damage identification: the Z24 benchmark[J]. Living Reference Work Entry, 2015, 1: 1-8.
[26]	Liu F T, Ting K M, Zhou Z H. Isolation forest[C]∥ Eighth IEEE International Conference on Data Mining, Pisa, Italy, 2008: 413-422.
[27]	Borghesi A, Bartolini A, Lombardi M, et al. Anomaly detection using autoencoders in high performance computing systems[C]∥ Proceedings of the AAAI Conference on Artificial Intelligence, 2019, 33(1): 9428-9433.
[28]	Hundman K, Constantinou V, Laporte C, et al. Detecting spacecraft anomalies using lstms and nonparametric dynamic thresholding[C]∥ Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, London, UK, 2018: 387-395.
[29]	Park D, Hoshi Y, Kemp C C. A multimodal anomaly detector for robot-assisted feeding using an lstm-based variational autoencoder[J]. IEEE Robotics and Automation Letters, 2018, 3(3): 1544-1551.
[30]	Zong B, Song Q, Min M R, et al. Deep autoencoding gaussian mixture model for unsupervised anomaly detection[C]∥ International Conference on Learning Representations, Vancouver, Canada, 2018: 1-19.
[31]	Li D, Chen D, Jin B, et al. MAD-GAN: multivariate anomaly detection for time series data with generative adversarial networks[C]∥ International Conference on Artificial Neural Networks, Budapest, Hungary, 2019: 703-716.
[32]	Su Y, Zhao Y, Niu C, et al. Robust anomaly detection for multivariate time series through stochastic recurrent neural network[C]∥ Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, Anchorage, USA, 2019: 2828-2837.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

损伤情况	时间	描述	损伤情况	时间	描述
D0	8月4日	健康状态	D8	8月27日	桥台滑坡1 m
D1	8月10日	Koppigen桥墩降低20 mm	D9	8月31日	混凝土铰链失效
D2	8月12日	Koppigen桥墩降低40 mm	D10	9月2日	2个锚头失效
D3	8月17日	Koppigen桥墩降低80 mm	D11	9月3日	4个锚头失效
D4	8月18日	Koppigen桥墩降低95 mm	D12	9月7日	2根肌腱断裂
D5	8月19日	桥墩提升，基础倾斜	D13	9月8日	4根肌腱断裂
D6	8月25日	底板混凝土剥落12 m²	D14	9月9日	6根肌腱断裂
D7	8月26日	底板混凝土剥落24 m²

Models	ACC	Precision	Recall	F₁
iForest	0.596 2	0.649 8	0.805 6	0.719 4
PCA	0.535 0	0.551 0	0.828 4	0.662 0
AutoEncoder	0.572 0	0.591 1	0.556 2	0.573 1
LSTM	0.618 2	0.652 6	0.968 1	0.779 6
LSTM_VAE	0.651 3	0.625 7	0.854 6	0.722 4
DAGMM	0.623 2	0.656 9	0.985 8	0.788 4
MAD_GAN	0.793 7	0.851 7	0.899 1	0.874 8
OmniAnomaly	0.637 4	0.656 9	0.963 4	0.781 1
TSSF-BADM（Ours）	0.972 1	0.986 8	0.980 8	0.983 7

模型变体	Precision	Recall	F₁
V1	0.938 8	0.980 8	0.959 3
V2	0.927 2	0.680 1	0.784 6
V3	0.902 9	0.797 3	0.846 8
V4	0.932 0	0.786 2	0.852 9
TSSF-BADM（Ours）	0.986 8	0.980 8	0.983 7