基于递进复合式组合预测模型的软岩隧道大变形预测研究

Study of Large Deformation Prediction of Soft Rock Tunnel Based on Progressive Composite Combined Prediction Model

为实现对隧道大变形发展趋势的判断,达到优化现场施工,避免出现施工安全问题的目的,采用LS-SVM和优化GM(1,1)模型对隧道变形进行预测,并以误差平方和为指标,将两者的预测结果进行组合,再进一步利用BP神经网络对前者的预测误差进行修正,以实现综合预测。通过实例检验,得到最小二乘法对支持向量机的优化效果要优于对灰色模型的优化效果,且误差修正模型能进一步有效地提高预测精度,使预测值与实测值更为接近;同时,通过本文的预测结果,得到后4个周期的变形仍具有持续变形的趋势,应采取有效措施,避免工程事故的发生。本文预测模型具有较好的预测精度及适用性,对隧道大变形研究具有一定的参考意义。

In order to realize the judgment of the development trend of large deformation of tunnel, optimize the site construction, and avoid construction safety risk, the LS-SVM and optimized GM （1, 1） models are adopted to predict the deformation of tunnel. And then the quadratic sum of error is taken as an index, and the prediction results of the two models are combined. Finally, the BP neural network is used to modify the prediction error so as to realize comprehensive prediction. The case test results show that the optimization effect on support vector machine by least squares method is superior to that by grey model; and the error modification model can further effectively improve the prediction accuracy, which make the predicted and measured values closer. Meanwhile, the prediction results show that the deformation of later 4 cycles are still with continuous deformation trend, which indicates that effective measures should be adopted to avoid accidents. The above-mentioned prediction model has better prediction accuracy and applicability, which has a certain significance for the study of large deformation of tunnel.