Author(s): Jiawei Lin; Haozhi Xu; Saiyu Yuan
Linked Author(s): Saiyu Yuan
Keywords: Yangtze River; Poyang Lake; Deep learning; Hydrodynamic prediction; Ecological scheduling
Abstract: The Yangtze River, the longest river in China, has a basin rich in rainfall and abundant hydropower resources. As a result, large-scale cascade hydropower projects, including six major hydropower stations like the renowned Three Gorges Dam, have been constructed along its upper reaches. These hydropower stations brought significant benefits in electricity generation, flood control, navigation, and water resource security, but have altered hydrological characteristics, including flood pulse patterns and sediment processes, impacting ecosystems in the middle and lower reaches, such as the entire life cycle of migratory fish. Poyang Lake, the largest freshwater lake in China, is a tributary-connected lake of the Lower Yangtze River, playing a vital supporter of biodiversity within the Yangtze River–Poyang Lake system. In recent years, Poyang Lake has experienced continuously declining water levels along with earlier and prolonged dry seasons. This has reduced the lake’s eco-environmental carrying capacity, resulting in a sharp decline in aquatic organisms and plants. The accurate simulation and prediction of hydrodynamic influenced by cascade hydropower projects are important for ecological scheduling within the Yangtze River-Poyang Lake system. Physical models typically have high accuracy in hydrodynamic simulation; however, establishing such models for large systems requires a significant amount of time and must account for boundary conditions, including topography in the upper Yangtze River. With advancements in computational facilities and statistical algorithms, machine learning-based nonlinear models have also been used for streamflow series prediction. Recently, Physics-Informed Neural Network (PINN) have emerged as a promising approach in hydrodynamic simulation. Incorporating physical information, PINN can reduce the required training data and produce results with physical significance. However, the construction of PINN for the entire basin has high costs and is unnecessary, considering the differences in data volume and accuracy requirements within the basin. Therefore, the main goal of this study is to develop a new approach for hydrodynamic prediction of the Yangtze River-Poyang Lake system based on deep learning.
Year: 2025