Simulation of Water Level and Flow of Catastrophic Flood Based on the CNN-LSTM Coupling Network

The occurrence of catastrophic floods will increase the uncertainty hydrological forecasting at downstream stations. In order to solve problems unclear propagation law in middle and lower reaches Yangtze River inadaptability traditional methods, this paper uses M-K trend test method analyze annual average flow water level Yichang Hankou For conventional floods, Random Forest (RF), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), CNN-LSTM neural networks are used simulate level/flow station. simulation results analyzed by Nash–Sutcliffe Efficiency Coefficient (NSE), Kling–Gupta efficiency coefficient (KGE), Root Mean Square Error (RMSE), Symmetric Absolute Percentage (SMAPE). show that station a downward shows an upward trend. By comparing four indicators NSE, KGE, RMSE, SMAPE, coupling model was determined be best fitting model, with NSE KGE greater than 0.995 RMSE SMAPE less 0.200. proposed can provide technical support for flood control optimization, scheduling, emergency rescue, scheduling impact analysis Three Gorges Power Station.