Abstract: Neural network has been used to inverse the shear strength parameters of rock mass. When safety factor is used as network input, the number of input parameters less than the number of output parameters will result in big errors as the mapping between input and output could not be established. In view of this, a new method of neural network back analysis suitable for the shear strength parameters of rock mass is presented. Firstly, neural network back analysis is employed to determine the range of inversion parameter, and then normal analysis is used for simulation in the above range to predict the safety factor. Finally, by using method of optimization, the absolute difference value between predicted and inversed values of safety factor is taken as objective function to find out the combination of cohesion and internal friction angle in correspondence with the safety factor which has minimum difference with the safety factor in the inversion conditions. This combination is thus regarded as the final inversed shear strength parameters of rock mass. Engineering examples show that the safety factor obtained by the method is equal to the safety factor in inversion conditions, indicating that this method is effective and feasible to solve the above problems.

Key words: neural network, shear strength parameters, safety factor, back analysis, optimization