Multi-index classification model for loess deposits based on rough set and BP neural network

Classifying loess deposits is an important process for selecting support form and construction methods for tunnels. An accurate evaluation of loess deposits is a necessary prerequisite to control deformation, save cost, and improve construction efficiency. In this paper, a neural network model with an evaluation system consisting of physical and mechanical indices of loess is proposed to realize intelligent classification of loess deposits for tunneling. The influence of water content, natural density, cohesion, internal friction angle, elastic modulus, and Poisson ratio on stability level of loess is analyzed by rough set theory based on statistical data of borehole samples. Results show that the affect of natural density is negligible. Then other indicators such as input nodes and the BP neural network model are formed after learning statistical samples and being applied to the project for testing. Finally, the output of the model is consistent with the actual. This study provides a multi-index model for evaluating loess deposits surrounding tunnels and provides a reference for future research.