Integrating knowledge-data-driven method to predict load-displacement curve on a trapdoor

Abstract

Accurately predicting the overlying earth pressure of underground structures is a key point in the design and construction of underground structures. On the basis of the existing composite function model describing the load–displacement curve (LDC) and experimental data, a intergrating knowledge-data-driven method to predict LDC is proposed in this paper. The buried depth (H), trapdoor width (B), soil weight (γ) and internal friction angle (φ) are selected as multi-characteristic input variables. The initial modulus of arching a, the minimum soil arching ratio ρ_min, the minimum soil arching ratio displacement ξ_min and the ultimate soil arching ratio ρ_ult are output variables. Two swarm intelligence optimization algorithms (i.e., sparrow search algorithm (SSA) and particle swarm optimization (PSO)) are used to optimize the parameters of the established generalized regression neural network (GRNN), and the knowledge-data cooperatively driven prediction of the LDC is realized. The results show that the GRNN model optimized by swarm intelligence algorithm has better prediction performance than the GRNN model. The LDC obtained from the output parameters of three GRNN models are compared with the results of the trapdoor experiments. The comparison results show that the LDC obtained by the GRNN model optimized by swarm intelligence algorithm are more consistent with the experimental results than that those obtained by GRNN model, and the prediction performance of the SSA-GRNN is slightly better than that of the PSO-GRNN.