Instability-Aware Digital Twins for Robust Co-Optimization of Operational Carbon Emissions, Thermal Comfort, and Structural Serviceability Reliability under Uncertainty in Smart Buildings

Abstract

Modern smart buildings operate under continuously varying climatic, occupancy, equipment, grid-carbon, material, degradation, and sensing conditions, making nominal optimization insufficient for reliable operational decision-making. This study develops and numerically evaluates an instability-aware digital twin for robust co-optimization of operational carbon emissions, occupied-zone thermal discomfort, and component-level structural serviceability reliability under uncertainty. The framework combines a multi-rate thermal–energy–moisture–IAQ–structural model, carbon-aware CO₂-based ventilation control, extended Kalman filtering, Monte Carlo uncertainty propagation, FORM and design-point importance sampling, Gaussian-process surrogate modeling, NSGA-II, and dimensionless objective- and decision-instability metrics. A controlled six-zone benchmark represents a six-story office building with a gross floor area of 10,200 m² over a one-year operational horizon and a 20-year reliability horizon. Re-evaluation of the nominally selected P0 policy under the full modeled uncertainty set increased carbon emissions from 42.6 to 50.4 kgCO₂e m⁻² year⁻¹, occupied zone-time discomfort from 115 to 128 K h year⁻¹, and the serviceability limit-state exceedance probability from 0.021 to 0.031. The corresponding nominal-input underestimations were 15.5% for carbon and 32.3% for the exceedance probability. The selected robust policy produced 48.8 kgCO₂e m⁻² year⁻¹, 98 K h year⁻¹, and PSLS = 0.012, with a reliability index of 2.257. Relative to the uncertainty-aware P0 evaluation, it reduced carbon by 3.2%, discomfort by 23.4%, PSLS by 61.3%, and decision instability by 65.7%. The robust policy maintained a mean occupied zone-time CO₂ concentration of 684 ppm, a 99th percentile of 821 ppm, and zero occupied zone-hours above 1,000 ppm. Within the controlled numerical benchmark, the results show that Pareto efficiency alone does not establish operational robustness and that explicit instability quantification materially changes policy selection.