ABSTRACT

One improved SVR algorithm was introduced into the field of elasto-plastic displacement back analysis in this paper. The strong nonlinear and uncertain relation between calculation parameters of rock mass and displacement of surrounding rock was described by the improved SVR algorithm. In order to find the optimal parameters of this improved SVR model during samples training course, the Genetic Algorithm (GA) was combined with it to form the improved GA-SVR algorithm. After the optimal non-linear mapping between the elasto-plastic mechanical parameters and displacement had been established, GA was used to identify these mechanical parameters within their search interval. By dint of MATLAB toolbox, GA also was integrated with BP neural network to form the GABP algorithm. Compared the back analysis results of the same elasto-plastic model parameters in BAOZHEN long and large railway tunnel by the two different algorithms, it can be concluded that the improved GA-SVR algorithm can obtain a more high inversion precision and calculation efficiency than that of GA-BP algorithm, so the improved GA-SVR can be applied in similar geotechnical engineering.

1 INTRODUCTION

Artificial neural network (ANN) is characterized by self learning, self adapting and parallel computation, so it can be used to solve strong nonlinear, discontinuity and uncertain relation. The BP neural network was employed to estimate the mechanical parameters of rock mass of Three Gorges permanent shiplock(Feng,2000). However, ANN is a large sample learning machine and has the disadvantages in overfitting and local optimization, which set a bottleneck constraint in application of geotechnical engineering (Zhao, 2003).

Support vector regression (SVR) algorithm has many merits such as small sample and global optimization etc. Compared with ANN, SVR algorithm based on the structure risk minimization principle can avoid the overfitting (Vapnik, 1995). So it is more appropriate to be applied into underground engineering than ANN algorithm (Zhao 2003, Liu 2004). During the process of BAOZHEN tunnel construction in Yichang-Wanzhou railway, SVR algorithm was introduced into the displacement back analysis in this paper.

2 BRIEF INTRODUCTION OF BAOZHEN LONG AND LARGE TUNNEL

BAOZHEN tunnel, located between the HEJIAPING and LANGPING Town of CHANGYANG County, HUBEI province, is one long and large tunnel of railway from YICHANG to WANZHOU which is the national key engineering. The distance between the left and right line is 30M. The total length of left line is 11563M, and the other is 11595M. According to the initial design, the right line is a parallel pilot which severs as assistant function during left line construction period. The length of IV and V grade surrounding rock is about 60 percents of whole tunnel. The embedded depth of this tunnel is large, and local section reaches 630M. The section of left line from DK72+834 to DK79+887 and the section of right line from YDK72+248 to YDK79+995 belong to extreme high stress district and inclined to generate large deformation during construction process.

The section from DK73+430 to DK73+480 of left line is the study part in this paper which possess high crustal stress and soft surrounding rock.

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