Back analysis of mechanical parameters based on GPSO-BP neural network and its application

Rock mass mechanical parameters are essential for the design and construction of underground engineering projects, but parameters obtained through traditional methods are often unsuitable for direct use in numerical simulations. The back analysis method, based on displacement monitoring, has emerged as a new approach for determining rock mass parameters. In this study, an experimental scheme was developed using orthogonal and uniform experimental designs to obtain training samples for the neural network. A GA-PSO-BP neural network model (GPSO-BP) was proposed, combining the fast convergence of the particle swarm optimization (PSO) algorithm and the global optimization capability of the genetic algorithm (GA). This model was applied to invert the rock mass parameters E, mu, phi, and c for deep-buried tunnels. The results indicate that the GPSO-BP neural network model outperforms the BP, GA-BP, and PSO-BP neural network models in terms of faster convergence and higher accuracy. It also shows superior performance in handling small datasets and complex problems, achieving better data fitting and the highest score in rank analysis. The DDR curve further confirms the GPSO-BP model's computational efficiency. When the rock mass parameters derived from this model are applied to forward numerical simulations, the average error across four monitoring projects is only 4.34%, outperforming the other three models. Thus, this study provides an effective method for improving the accuracy of rock mass parameter inversion in underground engineering.