Abstract:

For slopes that has failed or deformed significantly, the shear strength of rock and soil mass is frequently inversely estimated based on a factor of safety assumed. For the slope with a sliding surface passing through multi-layer rock and soil mass, it is unreasonable to achieve this goal by trial and error. To solve this issue, back propagation (BP) neural network is constructed using shear strength of multi-layer rock and soil mass as the input and the factor of safety of the slope, and the entry and exit positions of the sliding surface obtained by GeoSlope as the outputs. Then, based on the assumed factor of safety and the entry and exit positions measured in site, the shear strength is acquired by carrying out the “reverse back analysis-error check-sample correction” procedure repeatedly. The result of a case study verifies that the shear strength obtained by this method is reasonable and can be used as a reference when designing prevention measures for small-scale slopes. BP neural network usually considers the known information as the input, and the information to be determined as the output, which will induce a mathematical underdetermined problem when solving this issue. The proposed method avoids this demerit successfully, and has a lower requirement on the number of samples in the library and a higher precision compared to the classical BP neural network.

Key words: slope prevention, neural network, parameter inversion, reverse iteration, underdetermined problems