Abstract: Back analysis of the mechanical parameters of surrounding rocks using a method without sufficient geotechnical tests, has been at the forefront in geotechnical engineering research. To obtain more reasonable and accurate surrounding rock parameters for a water diversion tunnel, an intelligent inversion model integrating multiple machine learning algorithms is developed, and the influence of the parameters on tunnel displacement is examined via sensitivity analysis. Parameters of 25 groups are designed using orthogonal experiment, and the displacements at the monitoring points are calculated through FLAC3D simulations. Then, based on these data, different algorithms are selected to construct an intelligent fusion model for calculations of elastic modulus, Poisson's ratio, cohesion, and internal friction angle. Finally, through a case study of the Yindajihuang water diversion tunnel in Qinghai, the influence of mechanical parameters of surrounding rocks on its displacements is analyzed. By back analysis using this model and FLAC3D forward calculations of the parameters, the settlements at the different positions are obtained with the errors of 5.01%, 3.21%, 3.87% and 4.17% in calculations of the crown settlement, bottom heave, and left and right spandrel displacements respectively relative to on-site measurements. These relative errors, smaller than those of the single models, indicate our intelligent inversion fusion model and analysis method are a significant improvement on surrounding rock parameter calculations.

Key words: diversion tunnel, mechanical parameters of surrounding rock mass, parameter inversion, intelligent fusion model, numerical simulation, orthogonal experiment