The parallel construction of new railways adjacent to existing lines poses critical challenges in balancing infrastructure safety and land-use efficiency. Traditional finite element analysis (FEA), while accurate, faces prohibitive computational demands for system-wide reliability assessments. This chapter proposes a Bayesian Neural Network (BNN)-based surrogate modeling framework that integrates two-dimensional FEA with probabilistic reliability theory. Key contributions include: (1) An automated script-driven FEA automation framework workflow generating 33,000 simulation datasets spanning diverse soil parameters and construction phases; (2) A BNN surrogate model that predicts minimum safe distances while quantifying prediction uncertainty through Monte Carlo dropout; (3) A dual-limit state reliability model compliant with China’s code. Validated through the Shijiazhuang-Jinan High-Speed Railway case, the framework achieved safety boundary predictions consistent with as-built measurements, demonstrating significant computational efficiency gains over conventional 3D FEA. This methodology provides engineers with an uncertainty-aware decision tool for urban railway projects under spatial constraints.