Abstract:

A displacement back analysis method is proposed by combining uniform design method, the least squares support vector machine (LSSVM)and particle swarm optimization (PSO). The learning samples are produced by uniform design and FLAC3D. The parameters of LSSVM are determined in global optimal by PSO. Thus, the LSSVM with optimal parameters are used to describe the nonlinear relationship between the back analysis parameters and displacements. The PSO is used again to search for the optimal rock mechanical parameters in global ranges. The displacement back analysis method combines the advantages of three algorithms. The PSO has merits such as easy operation, highly active; the LSSVM has merits such as small sample, good generalization and the uniform design method can produce small sample.The model of PSO-LSSVM is used to make back-analysis of in-situ stress field of the underground powerhouse area of Longtan Hydropower Station. By comparison of measured and calculated displacements and in-situ stresses of rock masses, it is shown that the obtained results are satisfactory. The results also indicate that the model can be well applied to the fast displacement back analysis in geotechnical engineering.

Key words: displacement back analysis, least squares support vector machines, particle swarm optimization, uniform design method, fast back analysis