Abstract: Using experimental tests on the Wudongde hydro-elastic model, this paper studies the effects of different operation modes and gate openings of its flood outlets on the vibrations of the downstream slope and bedrock, and develops two numerical models for vibration prediction using the BP neural network. Flood flows are divided into three levels of discharge, and optimized schemes for flood control and vibration reduction at each level are obtained. The results show that (1) in the mode of only one surface outlet opened, bedrock vibration is intensified substantially with the increasing gate opening; while in the mode of all surface outlets fully opened, bedrock vibration is reduced and slope vibration is always positively correlated with gate opening. In the joint operation of the surface outlets and deep outlets, bedrock vibration is the weakest at a surface outlet opening of 4 m, but slope vibration is more complicated. (2) The two numerical models are good in prediction accuracy, and constructing a specific prediction model for the joint mode can improve accuracy. (3) To release flood at very small flow rates, multiple surface outlets can be used with a small gate opening; at normal flow rates, the middle outlets should be put into full use, combining, if necessary, with a few surface outlets opened at 4 m opening; at large flow rates, the middle outlets must be fully opened and multiple surface outlets working at small openings.

Key words: hydraulic model test, Wudongde hydro-elastic model, prediction model, BP neural network, flood discharge and vibration reduction