Compartment Fire Dynamics and Surrogate-Assisted Optimization in Confined Electrical Infrastructure

This paper presents a surrogate-assisted optimization framework for performance-based fire protection design of confined electrical infrastructure, demonstrated on 65 Fire Dynamics Simulator (FDS) simulations of a distribution room and a cable trench. A cell-by-cell geometry validator revealed that FDS silently discards obstructions, vents, and fire surfaces falling outside the mesh or on solid cells; all 65 inputs were corrected. First, mechanical exhaust in the distribution room correlates negatively with available safe egress time (ASET; r=−0.45, p<0.01): increasing exhaust to 15 ACH reduces mean ASET by 36% (61 to 39 s) via destratification. Second, cable-trench tenability is governed by natural ventilation: manhole size (r=+0.56) and end-opening ratio (r=+0.48) dominate, while fire-barrier spacing is negligible (r=+0.15). A Gaussian-process surrogate predicts Case-2 ASET at R2=0.68; NSGA-II maps the safety-margin–cost trade-off at ∼ 10 5 speedup over CFD-in-the-loop optimization.