Research On Bridge Data Anomaly Detection Based On OPTICS-Transformer

• Published in: 2024 36th Chinese Control and Decision Conference (CCDC) pp. 5886 - 5891
• Main Authors: Shu, Junjie, Li, Funian, Yu, Hong, Yan, Junfeng, Guo, Jiezhen
• Format: Conference Proceeding
• Language: English
• Published: IEEE 25.05.2024
• Subjects: Accuracy; Anomaly Detection; Bridges; Clustering algorithms; Data models; Optics; Predictive models; Transformer; Transformers
• ISSN: 1948-9447
• DOI: 10.1109/CCDC62350.2024.10588284

From time to time in bridge engineering, abnormal data are collected by sensors due to abnormal factors, and the traditional method using LSTM usually has high time cost, in order to better meet the needs of the bridge monitoring system, this paper constructs an anomaly detection method based on the Transformer model and OPTICS clustering algorithm, which is used in the bridge anomaly early warning system. This paper takes Wuhan Gaoxin Avenue cable-stayed bridge as the background, processes the bridge dynamic strain sensor data, and in order to improve the accuracy of anomaly data detection, establishes a time-sequence anomaly detection model based on OPTICS-Transformer. In this paper, OPTICS clustering is utilized to obtain the main operating states of dynamic strain sensors, and the Transformer model is utilized for anomaly data detection. The experimental results show that the RMSE of the Transformer prediction model is 0.1159, which meets the demand of data prediction, and the accuracy of the anomaly detection based on the predicted data to be detected is 83.25%, which is higher than that of the anomaly detection using only the Transformer model with an accuracy of 79.50%, which meets the demand of the early warning system.