基于小波去噪及优化BP神经网络的滑坡变形预测研究

Study on Landslide Deformation Prediction Based on Wavelet Denoising and Optimized BP Neural Network

为提高滑坡变形预测精度,以小波去噪和优化BP神经网络为基础,构建了滑坡变形预测模型,即先利用小波去噪剔除滑坡变形序列中的误差信息,再利用BP神经网络实现滑坡变形预测,且为保证其预测精度,利用试算筛选和混沌理论优化其模型参数,以实现滑坡变形的优化预测。实例研究表明:小波函数、阈值选取方法和小波分解层数对去噪效果的影响较大,sym8小波函数、软阈值及12层分解层数组合在实例中的去噪效果相对最优;同时,隐层节点数优化和节点阈值优化能有效提高BP神经网络的预测精度,在初步预测效果评价中,SH1号监测点的相对误差均小于2%,平均相对误差仅为1.65%,并在可靠性验证中,SHZ2号和SHZ3号监测点预测结果的平均相对误差分别为1.54%和1.51%,说明该模型不仅具有较高的预测精度,还具有较好的稳定性,适用于滑坡变形预测。

In order to improve the prediction accuracy of landslide deformation,based on wavelet denoising and optimized BP neural network,this paper constructs the landslide deformation prediction model,which first uses wavelet denoising to eliminate the error information in landslide deformation sequence,and then uses BP neural network to realize landslide deformation prediction.And to ensure the prediction accuracy,it optimizes the model parameters by trial calculation filter and chaos theory to achieve optimal prediction of landslide deformation.The case study shows that the wavelet function,threshold selection method and wavelet decomposition level have great influence on the denoising effect.The combination of sym8 wavelet function,soft threshold and 12 decomposition level is relatively optimal in this case.At the same time,the hidden node number optimization and node threshold optimization can effectively improve the prediction accuracy of BP neural network.In the evaluation of preliminary prediction effect,the relative error of SH1 monitoring point is less than 2%,and the average relative error is only 1.65%.In the reliability verification,the average relative error of the prediction results of SHZ2 and SHZ3 monitoring points are 1.54% and 1.51%,respectively,which indicates that the model has not only high prediction accuracy,but also good stability,so it is suitable for landslide deformation prediction.