Abstract

Under the conditions of unfavorable topography and geology, the construction of the mountain tunnel may pose a significant risk of tunnel collapse and large deformation accidents, which will cause casualties and great economic losses. Then how to predict the occurrence of tunnel collapse is particularly important. We usually predict tunnel collapse based on the tunnel deformation. In the process of tunnel construction, there are many factors related to the tunnel deformation, such as the rock mass classification, tunnel dimensions, groundwater, ground stress, construction methods and process, climate, management methods, etc. And they will affect each other as well. Therefore, some soft science technologies often are adopted to predict the tunnel deformation, such as artificial neural network, support vector machine (SVM), grey theory, etc. Support vector machine (SVM) has shown a remarkable generalization performance. But there is no determinate theory or method to select the internal indexes of the SVM model for tunnel deformation prediction at present. This paper establishes several tunnel collapse prediction models using support vector machine (SVM) according to the 150 cases we collected. Among them, 100 cases are used as train set to establish the SVM prediction models, and the rest are used as test set. Through a series of tests, several indexes which can improve the prediction accuracy are selected among the all indexes mentioned above. Combined with these selected indexes, this paper establishes an improved SVM deformation prediction model and tests the accuracy of the model with the data from Yima mountain tunnel. In order to assess the influence of those factors which are difficult to quantify, this paper takes the Hurst exponent into consideration in the support vector machine deformation models. And it indicates that the deformation prediction accuracy can be improved by analyzing the Hurst exponent of the tunnel deformation sequence. Furthermore, the accuracy of the improved deformation prediction model can satisfy the need of actual projects.

This content is only available via PDF.
You can access this article if you purchase or spend a download.