A two‐stage data cleansing method for bridge global positioning system monitoring data based on bi‐direction long and short term memory anomaly identification and conditional generative adversarial networks data repair. Issue 9 (11th May 2022)

• Record Type: Journal Article
• Authors: Yang, Kang; Ding, Youliang; Jiang, Huachen; Zhao, Hanwei; Luo, Gan
• Is Part Of: Structural control and health monitoring. Volume 29:Issue 9(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-05-11
• Subjects: anomaly detection -- bidirectional long‐ and short‐term memory neural network -- conditional generation adversarial network -- data cleansing -- data repair; Structural engineering -- Periodicals; Structural control (Engineering) -- Periodicals; Automatic data collection systems -- Periodicals; Detectors -- Periodicals; 624.17
• DOI: 10.1002/stc.2993
• Languages: English
• ISSNs: 1545-2255

Abstract

Summary: Data cleansing is an essential approach for improving data quality. Therefore, it is the key to avoiding the false alarm of the monitoring system due to the anomaly of the data itself. Data cleansing consists of two parts: anomaly identification and anomaly repair. However, current research on data cleansing has mainly focused on anomaly identification and lacks efficient data repair methods. The key to data repair lies in sensor correlation models based on mapping relationships between sensors. To obtain a good inter‐sensor relationship model, it is first necessary to exclude anomalous data from the training data set used for modeling. Therefore, a two‐stage data cleansing framework for collaborative multi‐sensor repair is proposed. First, based on the analysis of anomalous features of GPS data, a bidirectional long‐ and short‐term memory (Bi‐LSTM) neural network model is adopted for data anomalies classification and localization. As a result, the data segment to be repaired is determined. Then, on the basis of all sensor data in the time range of the day before the target repair data segment, the data set for data repair is constructed by excluding the anomaly data segments in the data set with the help of the above anomaly identification results. Then, a conditional generation adversarial network (CGAN) is proposed to achieve data repair. Experimental validation shows that the two‐stage data cleansing method of identification followed by repair can accurately identify and repair GPS anomalies. Finally, several factors affecting the repair effect are discussed.