<p>Deep tunnels excavated in creep-prone rock formations exhibit pronounced time-dependent deformations, which introduce significant uncertainty into the long-term support performance. In this study, an integrated probabilistic framework was proposed to address this challenge by combining Bayesian inference with reliability-based design optimization. The approach first characterized the probabilistic properties of creep rock parameters using time-dependent tunnel convergence data. These uncertainties were then incorporated into the optimization of a tunnel support system composed of multiple liners (including an inner concrete layer covered by a compressible layer). The use of Artificial Neural Networks as surrogate models significantly enhances computational efficiency while maintaining high prediction accuracy for tunnel convergence and liner stress responses over time. Numerical investigations confirmed that the proposed approach reliably captures the uncertainty associated with creep rock parameters and its impact on long-term tunnel behavior. The proposed framework was validated through numerical applications, demonstrating its effectiveness in identifying optimal support designs that met prescribed safety and serviceability criteria (i.e., allowable stress and displacement of the liner).</p>