Abstract: A basic back analysis model is proposed in consideration of the weakening effect of loose zone parameters for Dagangshan underground powerhouse caverns excavated in layers. The learning and testing samples of neural network were acquired by introducing GA BP algorithm based on orthogonal experimental design and FLAC 3D differential procedures. Hence a highly nonlinear mapping relation between the rock’s mechanical parameters and the displacements was established. According to field measured displacement and network mapping displacement, the objective function was obtained. By using genetic algorithm, the optimum rockmass parameters were obtained by searching parameter combinations which can make the network mapping displacement and the field measured displacement satisfy the optimal solution of the objective function within the range of empirical values. Finally, forward analysis on the obtained rockmass parameters and backward error detection on the displacements were carried out. Results show that the method is suitable for the design and construction of underground works.

Key words: rock mechanics, rockmass parameters, loose zone, underground caverns, parameter’s weakening effect, GA BP algorithm, Dagangshan hydropower station