Ensemble Learning-Aided Hydrothermal Coupling Investigation on Dam Impoundment Impact: A Case Study of the Xiluodu Project in China

Abstract

The monitoring and analysis of the temperature field in the reservoir area is gaining importance as it provides insights regarding structural safety and environmental impact. Most studies focus on earthen dams and the seasonal fluctuation of the thermal field induced by the ambient temperature. Few studies analyse the long-term trend of the bedrock temperature after high-arch dam impoundment as the involved geological elements are likely to be unexpected and unknown. In this paper, a unique geothermal evolution model in the downstream bedrock of a 285.5 m high-arch dam is reported based on continuous field observations during the past nine years, which cannot be explained by available hydrogeological data. To clarify the phenomenon, a research framework that comprises field measurements, machine learning interpretation, and hydrothermal coupling simulation is proposed. Borehole surveys and water chemistry analyses are performed to gather more data from critical areas. The proposed machine learning pipelines attempt to interpret the field test data. These are integrated with a double-loop grid-search process and can achieve hyperparameter searching and model evaluation simultaneously. It proves to be robust to local minima and more applicable to groundwater source discrimination when compared to other models. The contribution of the newly discovered features is further discussed through comparative and sensitivity analysis based on the hydrothermal coupling theory. With the proposed hybrid analysis framework, the shallow buried limestones and the confined hot groundwater have been identified as the causes of the unique phenomenon. These factors should be of concern to researchers facing similar situation in the future.