Flood Peak Prediction of Three Gorges Reservoir in China: Evidence from Artificial Intelligence and Observation Data

Abstract

The Three Gorges Project in China is the world’s largest water conservancy hub, located in the middle reaches of the Yangtze River in central China. After the storage of water in the Three Gorges Reservoir, the artificial lakes formed between the Three Gorges Reservoir area change frequently with the fluctuation of the reservoir water level. The size of the flood peak seriously affects the safety and regulation of the Three Gorges Dam. In this paper, based on the Three Gorges Reservoir interval observation data, a method of adaptive segmentation of rainfall-runoff events in time periods is proposed, and an artificial intelligence prediction model for the Three Gorges Reservoir interarea flooding is established on the basis of hydrological principles using data mining and machine learning methods. Nonlinear partial least squares regression and random forest are two models used and supported by later observations. Comparing the simulation results of the model with the hydrological observation records of the Three Gorges Reservoir, it is found that the model has its own advantages for different prediction objects. The results show that the nonlinear partial least squares regression model is suitable for flood prediction with high rainfall, while the random forest model can better predict the flood prediction process with low rainfall and a complex flood process.