GenFusion: Crafting future urban building layouts via diffusion model and genetic algorithm

With urbanization accelerating, building energy consumption now accounts for 35% of global energy use, making energy efficiency, solar utilization, and pedestrian wind comfort critical factors in urban planning. In high density cities, optimizing building layouts to reduce energy demand, enhance solar access, and improve wind conditions has become crucial. This paper proposes a framework called GenFusion, integrating generative AI with multi-objective optimization to efficiently generate and refine urban building layouts. By learning from real-world configurations, a diffusion model incorporates planning constraints to produce diverse and controllable layout designs. To accelerate energy consumption predictions, the framework employs an XGBoost model, enhancing computational efficiency. Additionally, the diffusion model accounts for wind speed and direction to regulate wind comfort conditions effectively. Multi-objective optimization is achieved using the NSGA-II algorithm. It has been demonstrated in the experimental results that GenFusion has an outstanding performance on urban building layouts design with diversity and suitability, and the evaluation shows that the layout generated achieves an SSIM value of approximately 90% SSIM compared with ground truth, while exhibiting an R 2 value exceeding 80% in terms of energy consumption. Overall, the proposed framework offers an innovative approach for sustainable urban design and serves as an important reference for future research combining AI-generated design with optimization.