Abstract:

The average velocity of a debris flow is one of indispensable parameters in control design of debris flow, so how to forcast accurately the average velocity is very important. The BP neural network model is suggested to forecast the average velocity of a debris flow. The average grain size, the debris flow depth, the gradient of the channel and the debris flow density are taken as the input units. The BP neural network model is established by training and forecasting the observation data of Jiangjiagou debris flows in Dongchuan, Yunnan Province. Comparing the forecasting results with the computing results by Dongchuan equation and modified Manning equation, the maximum errors are respectively 27% and 7.3% computing by modified equation and Dongchuan equation, and maximum error of the BP neural network is only 3.2%. The accuracy of the BP neural network is the highest. The method proposed in this paper is feasible and can forecast the average velocity of a debris flow accurately. This method is used to predict the average velocities of debris flows close to Wudongde hydropower station and the result can offer references for the design of debris flow control.

Key words: BP neural network, debris, average velocity, forecasting