Abstract: In rock engineering, the inversion of mechanical parameters has always been a key issue, and its accuracy is an essential factor in the design reliability and construction safety of a project. Traditional geomechanics physico-mechanical measurements are often biased by geomechanics interference, equipment and staff operation, so the results obtained are usually discontinuous. Aimed at such issues, this paper adopts a new method or PSO-XGB to invert and examine rock parameters－elastic modulus, internal friction angle, and cohesion of the surrounding rock of the regulating chamber tunnel. We tune the Extreme Gradient Boosting (XGBoost) model with the help of a bio-heuristic algorithm or Particle Swarm Optimization (PSO), and make a multi-model comparison. The findings indicate the PSO-XGB method is superior in efficiency and accuracy. The difference between the calculated and in-situ observed rock displacements is less than 5%, indicating a good agreement. This parameter inversion method is verified to be practically applicable and accurate, thus it is a new useful tool to determine and calibrate the rock mechanical parameters required for future design and construction.

Key words: PSO-XGB, surge chamber tunnel, parameter inversion, numerical simulation, model comparison