From Energy Efficiency to Thermal Resilience and Passive Survivability During Summer Power Outages: AI-Supported Prediction and Physics-Based Genetic Algorithm Optimization of Building Envelopes

Climate change and extreme weather events are intensifying the vulnerability of existing buildings to energy system failures, especially during prolonged heat waves. Power outages under such conditions compromise occupant health, and impose economic burdens. While resilience research has traditionally focused on energy infrastructure, the role of building envelopes in maintaining habitable indoor conditions during outages has been underexplored. This study presents a BIM-based simulation framework that couples AI prediction with multi-objective genetic algorithm optimization for evaluating and optimizing building envelope energy retrofits to enhance thermal resilience in residential buildings. Four representative housing typologies including detached, semi-detached, attached, and mid-rise apartments, were modeled across 14 ASHRAE climate zones. A dataset of 16,856 EnergyPlus simulations was generated using Latin Hypercube Sampling, capturing a wide spectrum of building envelope energy retrofits for walls, roofs, floors, and windows. Thermal resilience was evaluated through a suite of indicators, including Passive Survivability Index (PSI), Time to Thermal Failure, Temperature Deviation, Humidity Exposure Risk, Recovery Time, Post-outage Recovery Load, and Weighted Unmet Temperature Penalty ( WUMT P t ). XGBoost was trained on this dataset to predict thermal resilience metrics with high accuracy, enabling scalable and automated assessment. Finally, multi-objective optimization using NSGA-II identified Pareto-optimal retrofit solutions balancing survivability, thermal comfort, and energy recovery needs across climates. Results indicate that envelope retrofit strategies increased the average PSI during multi-day summer power outages by 14.9 percentage points across the evaluated climate zones and building typologies, corresponding to a 72.1% relative improvement. At the individual-case level, PSI improved by up to 54.7 percentage points (298%), thermal failure was delayed by up to 25 hours (261%), WUMT P t decreased by up to 145.6 points (63%), and peak indoor temperatures decreased by up to 8.3°C relative to baseline configurations. AI-based prediction achieved an overall R² above 0.99 for time-series outputs and 0.81 for static resilience metrics, with individual static-metric R² values ranging from 0.66 to 0.95, enabling rapid screening of retrofit options and supporting informed decision-making. Multi-objective optimization revealed clear trade-offs between passive survivability and energy performance, identifying Pareto-optimal retrofit packages tailored to climate and housing type. This study demonstrates that resilience-oriented envelope retrofits can improve occupant safety during outages, providing actionable guidance for climate-adaptive residential energy retrofit planning.