Prediction of the failure point settlement in rockfill dams based on spatial-temporal data and multiple-monitoring-point models

• Published in: Engineering structures 2021-09, Vol.243, p.112658, Article 112658
• Main Authors: Li, Yanlong, Min, Kaiyi, Zhang, Ye, Wen, Lifeng
• Format: Article
• Language: English
• Subjects: Aging; Cluster analysis; Clustering; Dam construction; Dam failure; Dam safety monitoring; Dams; Data analysis; Data collection; Data recovery; Deformation; Failure point; Lager; Malfunction sensors; Measurement methods; Missing data; Missing time series; Model accuracy; Monitoring; Multiple monitoring points model; Position sensing; Predictions; Rockfill dams; Sensors; Settlement analysis; Spatiotemporal data; Spatiotemporal data clustering; Training
• ISSN: 0141-0296; 1873-7323
• DOI: 10.1016/j.engstruct.2021.112658

•Failure point data missing is general among all stages of dam.•New model is more suitable for the long-term prediction of missing data than the single point model.•Spatiotemporal data clustering analysis is helpful to improve the prediction accuracy.•Performance of new model improves during the increase of monitoring data. The integrity of monitoring data is important in the study of the deformation law of rockfill dams. Environment, construction, aging, and other factors result in dam monitoring sensors malfunction at the initial stage of operation. Data collection discontinue leads to insufficient monitoring data. The huge amount of missing data is lager than traditional model training samples and increases the difficulty of data recovery in failure points. In this study, the multiple-monitoring-point (MMP) model is established to extend the number of training set samples. MMP model integrates spatiotemporal information to make predictions of long-term missing data in malfunctioning settlement sensors according to the corresponding relationship among the coordinate position, environment values, and settlement. Additionally, this paper presents the study of monitoring point selection and the clustering time period division in the MMP model. The spatiotemporal data clustering analysis is used as the measurement method to determine the settlement similarity to screen the appropriate data. Experiments on large-scale real dam deformation data demonstrate that the MMP model is suitable for the long-term data prediction of failures in rockfill dam settlement monitoring. After the spatiotemporal panel data clustering analysis, the model prediction accuracy is significantly improved. This model provides a new method for dam settlement prediction and analysis.