This study investigates the development of a hybrid model integrating metaheuristic optimization (Genetic Algorithm – GA) and deep generative models (Generative Adversarial Networks – GANs) to address the apartment layout planning challenge, with a focus on optimizing the inter-room distances under architectural constraints.
The proposed framework operates in two phases. First, a Genetic Algorithm (GA) is employed to optimize the spatial positions of rooms by minimizing weighted distances based on functional relationships. Then, a conditional Generative Adversarial Network (Pix2Pix-based cGAN) is used to generate detailed and topologically valid apartment layouts, conditioned on the room positions provided by the GA. This hybrid approach leverages the optimization capabilities of GA and the generative capacity of GANs to produce realistic and functionally coherent floor plans.
The experimental results demonstrate successful integration of spatial optimization (via GA) and generative design (via GANs), establishing a novel methodology for data-driven functional layout planning. This hybrid framework effectively combines the strengths of both optimization and generative approaches, achieving simultaneous improvements in spatial efficiency and design feasibility.
This study contributes to computational design literature by unifying evolutionary optimization and generative AI, thereby addressing the long-standing challenge of balancing functional efficiency and design synthesis. The proposed method addresses the limitations of earlier works that treated optimization and generation as separate processes, establishing instead an integrated workflow that achieves both functional optimization and creative design.
