An AI-driven multi-objective framework for optimizing window dimensions considering energy demand and thermal comfort

Designing energy-efficient windows requires balancing energy demand and occupant thermal comfort, particularly during early-stage design. This study presents an artificial intelligence-driven multi-objective framework for optimizing window dimensions in a simplified office building under Tehran's hot-dry climate. A parametric simulation model was developed using Grasshopper, Ladybug Tools, and EnergyPlus to generate a dataset covering variations in window width and height. For each configuration, annual cooling energy demand (kWh), annual heating energy demand (kWh), and adaptive thermal comfort, defined as the percentage of occupied hours within the ASHRAE 55 adaptive comfort range, were evaluated. An artificial neural network surrogate model was trained using decorrelated geometric inputs obtained through principal component analysis, achieving high predictive accuracy. The surrogate was embedded within a multi-objective evolutionary optimization process to explore trade-offs between energy demand and thermal comfort. The resulting Pareto-optimal solutions indicate that intermediate window sizes provide a balanced compromise between reduced energy demand and improved comfort. The proposed framework supports performance-informed window design decisions at the early design stage while remaining extensible to more complex design scenarios.