Scholarlink

<p>Retrofitting housing in hot–humid regions must balance energy, comfort, costs, and carbon emissions while remaining robust to climate change. This study develops a decision-support workflow that links physics-based simulation (TRNSYS), an interpretable machine-learning surrogate (Light Gradient Boosting Machine (LightGBM) with SHapley Additive exPlanations (SHAP)), multi-objective optimization (Non-dominated Sorting Genetic Algorithm II (NSGA-II)), and Multi-Criteria Decision-Making (Elimination and Choice Expressing the Reality III (ELECTRE III)). The framework is applied to a neighborhood of 34 dwellings in Chitré, Panama (tropical savanna climate), under current weather conditions and a mid-century Shared Socioeconomic Pathway 2–4.5 (SSP2–4.5) climate scenario. The surrogate, trained on a Latin hypercube of design alternatives, reproduces simulation outputs with low error and enables computationally efficient exploration of thousands of retrofit combinations and trade-offs among primary energy supply for cooling, life-cycle cost, operational and embodied CO₂e, and thermal discomfort (hours with Predicted Percentage of Dissatisfied (PPD) &gt; 10%). Across alternative decision-maker preference weightings, Pareto-optimal solutions reveal distinct retrofit pathways reflecting investment–performance trade-offs, yet “moderate” envelope upgrades (walls ≈ 0.5&#xa0;W/m<sup>2</sup>K; roof ≈ 0.25–0.50&#xa0;W/m<sup>2</sup>K) combined with glazing having a low solar heat gain coefficient (SHGC) (≈ 0.20–0.30) and thermostat set-points of 26&#xa0;°C consistently emerge near the Pareto front. Relative to the baseline, recommended packages reduce cooling energy by roughly one-third in the present climate and remain effective under both SSP2–4.5 and supplementary SSP5–8.5 future climate assessments, although with attenuated performance under stronger warming conditions. The resulting cooling load levels remain compatible with the capacities of residential mini-split air-conditioning (AC) systems commonly used in Panamanian dwellings. Sensitivity analyses of preference weights and pseudo-criteria thresholds confirm stability of the rankings. The results provide decision-ready retrofit bundles for tropical neighborhoods and a transparent, transferable methodology that unifies simulation, interpretable Artificial Intelligence (AI), optimization, and ranking for policy and practice.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hybrid AI-driven decision framework for climate-resilient neighborhood retrofits in tropical savanna climate

  • 期刊论文

  • 发表日期：

    2026年6月26日

查看全文
  • Fatima Harkouss,
  • Farouk Fardoun

摘要

Retrofitting housing in hot–humid regions must balance energy, comfort, costs, and carbon emissions while remaining robust to climate change. This study develops a decision-support workflow that links physics-based simulation (TRNSYS), an interpretable machine-learning surrogate (Light Gradient Boosting Machine (LightGBM) with SHapley Additive exPlanations (SHAP)), multi-objective optimization (Non-dominated Sorting Genetic Algorithm II (NSGA-II)), and Multi-Criteria Decision-Making (Elimination and Choice Expressing the Reality III (ELECTRE III)). The framework is applied to a neighborhood of 34 dwellings in Chitré, Panama (tropical savanna climate), under current weather conditions and a mid-century Shared Socioeconomic Pathway 2–4.5 (SSP2–4.5) climate scenario. The surrogate, trained on a Latin hypercube of design alternatives, reproduces simulation outputs with low error and enables computationally efficient exploration of thousands of retrofit combinations and trade-offs among primary energy supply for cooling, life-cycle cost, operational and embodied CO₂e, and thermal discomfort (hours with Predicted Percentage of Dissatisfied (PPD) > 10%). Across alternative decision-maker preference weightings, Pareto-optimal solutions reveal distinct retrofit pathways reflecting investment–performance trade-offs, yet “moderate” envelope upgrades (walls ≈ 0.5 W/m2K; roof ≈ 0.25–0.50 W/m2K) combined with glazing having a low solar heat gain coefficient (SHGC) (≈ 0.20–0.30) and thermostat set-points of 26 °C consistently emerge near the Pareto front. Relative to the baseline, recommended packages reduce cooling energy by roughly one-third in the present climate and remain effective under both SSP2–4.5 and supplementary SSP5–8.5 future climate assessments, although with attenuated performance under stronger warming conditions. The resulting cooling load levels remain compatible with the capacities of residential mini-split air-conditioning (AC) systems commonly used in Panamanian dwellings. Sensitivity analyses of preference weights and pseudo-criteria thresholds confirm stability of the rankings. The results provide decision-ready retrofit bundles for tropical neighborhoods and a transparent, transferable methodology that unifies simulation, interpretable Artificial Intelligence (AI), optimization, and ranking for policy and practice.