基于趋势项分离的隧道变形组合预测研究

Prediction Combination of Tunnel Deformation Based on Trend Term Separation

以小波变换分离隧道变形数据的趋势项和误差项,采用回归模型对趋势项进行单项预测,并进一步对其进行组合预测;同时,利用BP神经网络对误差项及原始数据项进行预测,最后对比本文组合预测与传统BP神经网络的预测结果。结果表明:sym5小波函数对本文监测数据的分离效果最好,而非线性组合预测的精度要优于线性组合预测的精度,且综合对比本文组合预测和BP神经网络的结果,得出本文的组合预测很大程度上提高了预测精度,为隧道的变形预测提供了一种新的思路。

In this paper, the trend and the error terms of the deformation data of the tunnel are separated by wavelet transform, the authors use the regression model to forecast the trend, and then make a combination forecast. At the same time, the authors use BP neural network to predict the error term and the original data, and then compare the forecasting results between combination forecast and the traditional BP neural network. The results show that sym5 wavelet function is the best for the separation of the monitoring data in this paper. The accuracy of nonlinear combination forecasting is better than that of linear combination forecasting, through comprehensive comparison of the results of the combined forecasting and BP neural network, the combination forecasting of this paper can greatly improve the prediction accuracy, and provide a new idea for the tunnel deformation prediction.