基于多尺度组合核极限学习机模型的隧道围岩变形预测及应用

Deformation Prediction for a Tunnel Rock Mass Based on the Multi-Scale Combination Kernel Extreme Learning Machine Model

隧道围岩变形序列具有高度非线性,采用常规方法很难得到满意的预测精度。为了提高隧道围岩变形的预测精度,基于实测变形数据,提出一种隧道围岩变形的多尺度组合核极限学习机预测模型。首先,通过集合经验模态分解技术将实测变形数据分解为多个不同的尺度序列,然后通过组合核极限学习机对各分量序列进行建模预测,最后将预测得到的各分量结果进行组合获得最终的预测值。改进模型中通过径向基函数和多项式核函数线性加权而成组合核函数,运用粒子群算法对核参数和加权系数进行优选,并通过马尔可夫链对模型的预测结果进行了讨论,可以较好地提高隧道围岩变形的预测精度。文章通过大相岭隧道围岩变形预测实例表明:提出的改进模型在单步预测和连续多步预测隧道围岩变形时,都能取得较高精度,对比可得优于贝叶斯正则化BP神经网络,与工程实例监测变形相比处于可接受范围内,具有一定的应用价值。

Tunnel rock deformation sequences are nonlinear and it is therefore difficult to obtain satisfactory precision in predictions by conventional methods. To improve the prediction accuracy of tunnel surrounding rock deformation, a model of a multi-scale extreme learning machine with a combination kernel is proposed based on measured deformation data. The measured deformation data is divided into different scale sequences using the empirical mode decomposition technique, a sequence of each component is predicted by the combination of the extreme learning machine and the final forecast value is obtained by combining the results of each component. For the improved model, a compound kernel parameter is obtained by the linear weighting of the radial basis function and poly- nomial kernel function, kernel parameters and weighting coefficients are optimized by particle swarm optimization, the prediction results of the model are discussed using Markov Chain and the prediction accuracy for tunnel surrounding rock deformation is improved. The predicted surrounding rock deformation of the Daxiangling tunnel shows that higher accuracy can be achieved in both the one-step prediction and multi-step prediction of tunnel surrounding rock deformation using the improved model, the proposed model is better than the Bayesian regularization BP neural network and the deformation is acceptable compared to the measured deformation.