Multisource monitoring data-driven slope stability prediction using ensemble learning techniques

Abstract

Field monitoring data from multiple monitoring points often contain redundant and/or dirty data, which provide less value for slope safety evaluation. Existing slope safety evaluation methods suffer from dealing with large amounts of monitoring data, resulting in a significant waste of monitoring effort. Moreover, these methods cannot evaluate the value of information (VoI) of the monitoring data for slope performance prediction and may abuse dirty data. To address these issues, this study proposes a multisource monitoring data-driven slope stability prediction method based on ensemble learning techniques. The complex mapping relationship between slope stability and monitoring data is established using the ensemble learning algorithm, based on which the VoI of the monitoring data is evaluated and the optimization of the monitoring layout is analyzed. An undrained slope example with multiple monitoring points arranged at various locations on the slope is analyzed to demonstrate the effectiveness of the proposed method. The results show that the proposed method can effectively and efficiently utilize the massive monitoring data to predict the slope stability. It also can provide a versatile tool for optimizing the layout of monitoring points and saving the monitoring costs.