A Prediction Model and Factor Importance Analysis of Multiple Measuring Points for Concrete Face Rockfill Dam during the Operation Period

Abstract

Dam settlement monitoring is a crucial project in the safety management of concrete face rockfill dams (CFRD) over their whole life cycle. With the development of an automatic monitoring system, a large amount of settlement data was collected. To precisely predict the structural health of dams, a combined multiple monitoring points (MMP) model and a machine learning model has been developed. In this paper, based on the physical factors of the CFRD, we comprehensively analyzed the influence of water level load transfer, rockfill rheology and soil properties on the settlement during the impoundment operation period. Then, we established a space-time distribution model of the CFRD during its operation period under multiple factors. An extreme gradient boosting (XGBoost) model was used for fitting prediction, and the model was evaluated using various performance indicators. The results show that spatial parameters such as the upper filling height, rockfill thickness, panel-point distance and soil material correlate to the deformation characteristics of the rockfill dam. Taking the monitoring data of the settlement of the Liyuan CFRD as an example, the new MMP model was evaluated and used to predict the settlement of the full-section points with higher accuracy, which has certain application and popularization value for related projects. Then, to evaluate the contribution of the components of the new MMP model, the SHapley Additive explanation (SHAP) methods are used to evaluate the importance of the selected factors, and the reasonability of these factors is verified.