Machine learning-based predictive model for optimal insulation thickness of external walls and life-cycle benefits in intermittently air-conditioned buildings integrated with global sensitivity analysis

Determining the optimal exterior wall insulation thickness is crucial for balancing air conditioning (AC) energy consumption and construction costs. Existing optimization approaches for intermittently conditioned buildings, however, are often limited to specific parameters derived from individual models, which restricts their applicability to other building contexts. Moreover, the optimal insulation thickness is inherently uncertain, arising from the complex interrelationships among building physical characteristics, AC operational patterns, and economic factors. To address the challenges of applicability and uncertainty, this study employs two sensitivity analysis methods: the Morris method for rapid identification of dominant parameters and the Sobol method for comprehensive uncertainty quantification, to identify the key factors influencing insulation optimization in intermittently conditioned buildings. Then, a predictive model using an artificial neural network is developed to assess the optimal insulation thickness and associated life-cycle benefits. Results demonstrate that the optimal insulation thickness is predominantly concentrated within the range of 0.04–0.07 m. Sensitivity analysis identifies insulation material lifespan (contributing 30% of variation), AC setpoint temperature (23%), and daily AC runtime (20%) as primary determinants, with secondary influences from the discount rate, inflation rate, and material costs. The derived six-parameter integrated model enables reliable predictions of optimal insulation thickness, life-cycle savings, and payback periods, with R2 values of 0.9718, 0.9851, and 0.9724 respectively. When applied across diverse building scenarios, the model-determined optimal insulation achieves an energy saving rate of approximately 17–60%, alongside a CO2 emission reduction of about 5–29 kg CO2/m2, thereby establishing a quantitative decision-making tool for building energy retrofits.