Design of tunnel shotcrete-bolting support based on a support vector machine approach

Design of tunnel shotcrete-bolting support based on a support vector machine approach

ARTICLE IN PRESS International Journal of Rock Mechanics & Mining Sciences 41 (2004) 510–511 SINOROCK2004 Paper 3A 09 Design of tunnel shotcrete-bo...

137KB Sizes 3 Downloads 171 Views

ARTICLE IN PRESS

International Journal of Rock Mechanics & Mining Sciences 41 (2004) 510–511

SINOROCK2004 Paper 3A 09

Design of tunnel shotcrete-bolting support based on a support vector machine approach K.Y. Liu*, C.S. Qiao, S.F. Tian School of Civil Engineering and Architecture, Beijing Jiaotong University, Beijing, China

Abstract Based on the minimization of structural risk, the support vector machine (SVM) approach is usually less vulnerable to overfitting than the artificial neural network (ANN) approach based on minimization of empirical risk. The solution calculated by using the SVM algorithm must be the global optimum—because the training problem in SVM is reducible to solving a convex quadratic programming (QP) problem. Design in the New Austrian tunnelling method (NATM) is still mainly in the qualitative phase, because the mechanical model and input parameters are difficult to select correctly; this causes the design of shotcrete-bolting support to still rely on constructive experiences. The support vector regression (SVR) algorithm is introduced to construct the non-linear relation between the parameters of tunnel shotcrete-bolting support and the influencing factors. Since the classical SVR algorithm can only solve a single output variable problem, an improved SVR algorithm (Fig. 1) is presented to resolve the multi-output-variable problem, and a computer code is developed in MATLAB. Sixteen factors influencing the tunnel shotcrete-bolting support are chosen as the input variables and six support parameters as the output variables of the SVM network. Sixty-four out of 94 tunnelling cases are employed as the learning sample and the remainder as the testing sample. Normalizing the sample data and adopting the trial-and-error method, a 16-24-6 BP network is selected to train and forecast for the same set of learning and testing samples. The e-insensitive loss function and the RBF kernel function are employed in virtue of their superiority to others. One method to find the optimal parameters of the SVM network is put forward. The essence of this method is to gradually determine the optimal parameters of the SVM network to achieve the effect of group optimization. From the results calculated by the two algorithms, it can be seen that the improved SVR algorithm is more effective than ANN and the design result is applicable to engineering, thus providing a new method for tunnel shotcrete-bolting design. Keywords: Tunnel; Support vector regression (SVR); Shotcrete-bolting support; Improved SVR algorithm; BP neural network; Parameter design

*Corresponding author. E-mail addresses: [email protected] (K.Y. Liu), [email protected] (C.S. Qiao). For full length paper see CD-ROM attached. doi:10.1016/j.ijrmms.2003.12.041

ARTICLE IN PRESS K.Y. Liu et al. / International Journal of Rock Mechanics & Mining Sciences 41 (2004) 510–511

START

START

j=0

i=1

X,Y=(y 1,...,y i) of learning samples

X,Y ¨ @ =(y 1¨ @ ,...,y j ¨ @ ) of testing samples

training and fitting

extrapolation and prediction j=j+1

i=i+1

y

y* i

i=N

NO

NO

YES output Y ¨ @

output Y*

END

¨ @

j+1=N

YES

(a)

j+1

(b)

END

Fig. 1. Flow chart of the improved SVR algorithm.

511