Optimal Selection and Combination of Seismic Intensity Measures for Shallow Shield Tunnels Considering Multicollinearity

Multivariate regression models incorporating multiple seismic intensity measures (IMs) can effectively represent tunnel seismic damage. However, multicollinearity among the IMs hinders their rational selection and the accurate establishment of the model. This study analyzed the distribution pattern of tunnel damage from both probabilistic and mechanical perspectives. A methodology system, termed LTP (Lasso-TOPSIS screening with Partial least squares regression), for IMs selection and combination optimization in shallow-buried soft-soil shield tunnels was proposed while addressing multicollinearity. The research further constructed a corresponding fragility analysis methodology incorporating multiple IMs, thereby extending the fragility analysis from a single parameter to multiple parameters. Results indicate that under transverse seismic loading, the shoulder and foot regions of the tunnel are most susceptible to severe damage. As the damage state escalates, the distribution pattern of the most severe damage locations shifts from concentrated to dispersed. Multivariate models established using the LTP methodology demonstrate superior fitting performance compared to conventional univariate linear models, achieving optimal fit with 6–8 IMs. Fragility analysis reveals that even low-intensity seismic motions can induce slight tunnel damage due to the presence of initial stresses. Multivariate-based fragility curves demonstrate higher predictive accuracy and damage probabilities than univariate fragility curves. The influence of seismic intensity uncertainty diminished with increasing damage severity. The proposed methodology was developed and validated with shallow shield tunnels. Its applicability to other burial depths, geological conditions, and tunnel types requires further validation with more case studies. This work provides a new methodology and analytical framework for IMs selection and fragility assessment.