基于BP神经网络的泥石流平均流速预测

Forcast for Average Velocity of Debris Flow Based on BP Neural Network

泥石流平均流速是泥石流防治工程中不可缺少的重要参数,准确地预测泥石流平均流速对于泥石流防治工程的设计是至关重要的。将BP神经网络应用于泥石流平均流速的预测:将泥石流平均流速的影响因素——泥沙平均粒径、泥深、沟床比降和泥石流密度作为BP神经网络的输入单元,通过对云南东川蒋家沟泥石流观测数据的训练与预测建立了泥石流平均流速的BP神经网络预测模型。将预测结果与东川公式和曼宁修正公式的计算结果进行对比:曼宁修正公式和东川公式预测结果最大误差分别为27%和7.3%,BP神经网络的预测结果最大误差仅为3.2%,BP神经网络的预测精度是最高的,可见此方法对泥石流平均流速预测具有适用性和准确性。最后应用此方法预测了乌东德水电站近坝库区内的3条泥石流的平均流速分别为12.8m/s、11.3m/s和13.0m/s,为库区泥石流防治工程提供了可靠的参考数据。

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.