滑坡变形的多因素小波神经网络预测模型

Multiple Factors Model for Landslide Deformation Prediction Based on Wavelet Neural Network

小波神经网络模型结合了小波变换良好的时频局域化性质及传统神经网络具有的自学习功能,具有良好的逼近与容错能力。以某水电工程区的典型滑坡为例,在对滑坡的基本特征、滑坡变形与主要影响因素相关关系进行分析的基础上,选择滑坡位移速率和对滑坡位移起控制作用的降雨2个因素,建立了滑坡变形的小波神经网络预测模型,并与其他多因素小波神经网络模型进行了比较。结果表明,所建的滑坡多因素小波神经网络模型的预测精度总体均比较高,其中以位移速率和降雨量建立的2因素小波神经网络模型的预测精度最高,优于其他小波神经网络模型。

Wavelet neural network has better approximation and fault-tolerance for combining the time-frequency localization of wavelet transform and self-study function of traditional neural network.We took some typical landslides in hydropower engineering region as an example and built three wavelet neural network models of multiple factors for landslide deformation prediction,on the basis of analyzing basic characteristics and the relationships between landslide deformation and main influencing factors of the landslide.By analyzing and comparing the results of the models,we found that the wavelet neural network model including the two factors（displacement rate and rainfall） has the highest prediction accuracy in the three models.