Author(s): Yuchen Li; Xiang Ji; Qirui Ma
Linked Author(s):
Keywords: Temperature control; RCC dam; Digital twins; Deep learning
Abstract: With the effective decision support provided by Digital twins (DT) technology, the construction and management of hydraulic engineering safety are characterized by digitization and intelligence. The characteristics of data in different phases, limited in the construction phase while abundant in the operational phase, determine that DT technology has a variety of manifestations in different phases. This study focused on intelligent temperature control of dams and proposed an entire-lifecycle DT model. A Physical Mechanisms and Deep Learning Coupled Model is introduced to address the data dependency issue in the construction phase. This model trained an ensemble surrogate model (ESM) based on the data obtained by high-accuracy finite element analysis (FEA), which is combined with artificial neural network (ANN), extreme gradient boosting (XGBoost), and support vector regression (SVR), to grasp the underlying mapping relationship between material parameters and structure responses. Subsequently, a multi-objective optimization algorithm Non-dominated Sorted Genetic Algorithm-III (NSGA-III) is used to eliminate errors caused by inaccurate material parameter measurements, using fusion information from multiple monitoring points. Then, a deep learning long short-term memory (LSTM) model is utilized to manage abundant data during the operational phase. The LSTM model explores the potential relationships between environmental factors and the dam’s health from long-term sequences. The proposed entire-lifecycle DT model is applied to a constructed high roller compacted concrete gravity dam, where the results show that the DT model achieves high forecasting capability and rapid response. The DT model achieves precise control of the dam, ensures its safety throughout its entire lifecycle, and improves the digitalization and intelligence of hydraulic engineering.
Year: 2024