Composite interpretability optimization ensemble learning inversion surrounding rock mechanical parameters and support optimization in soft rock tunnels

The mechanical parameters of the surrounding rock of a tunnel are the premise and foundation of the supporting design in soft rock tunnel engineering. To obtain the mechanical parameters of the surrounding rock accurately and quickly, a composite inversion model is proposed, and different models are used to invert different mechanical parameters. In this paper, first, the Latin hypercube and Monte Carlo sampling methods are used to construct the mechanical parameter samples from the surrounding rocks. The numerical simulation of the water diversion project in central Yunnan is carried out by using FLAC3D, and the corresponding displacement data are obtained, three sets of effective inversion datasets are formed along with the mechanical parameter samples. Second, the adjusted coefficient of determination and the symmetrical mean absolute percentage error are used as the performance evaluation indices. The Bayesian 10-fold cross-validation iteration is used to optimize the five regression model hyperparameters for the three sets of inversion parameters. The base model of the best performance is selected for inverting the friction angle (phi) and cohesion (c). For the deformation modulus (Em) inversion parameters, LIME interpretability is used to optimize the weights and integrate the five models. Finally, by developing a calculation tool for the elastic modulus of tunnel support, an orthogonal experimental design is applied based on the verified engineering numerical model. The results show that the average error of the inversion results of the mechanical parameters of the surrounding rock by the proposed composite model is 9.32%, the minimum error is only 0.81%, and the error is within 15%. The inversion trend is similar to the actual displacement height, and the optimized support scheme provides a reference for the engineering site.