High-Density Block Generation and ENVI-met Microclimate Evaluation Based on Pix2Pix GAN

Abstract

Rapid urbanization has precipitated the development of high-density urban blocks, particularly in metropolitan regions facing land scarcity. While high-density living offers economic efficiency, it often exacerbates the Urban Heat Island effect and compromises outdoor thermal comfort. Traditional urban design methodologies, which rely heavily on manual iteration and intuition, struggle to balance the complex trade-offs between floor area maximization and microclimatic performance. This research proposes an automated generative design framework integrating a Conditional Generative Adversarial Network, specifically the Pix2Pix architecture, with ENVI-met microclimate simulation. By training the neural network on a dataset of existing high-density urban morphologies and their corresponding layout constraints, the model learns to generate realistic block typologies. Subsequently, these generated forms are subjected to Computational Fluid Dynamics simulations via ENVI-met to evaluate their thermodynamic and aerodynamic performance. The study demonstrates that the Pix2Pix model can successfully synthesize urban textures that adhere to spatial constraints while simulation results reveal specific morphological traits that enhance wind permeability and reduce thermal stress. This interdisciplinary approach bridges the gap between artificial intelligence and sustainable urban planning, offering a robust decision-support tool for architects and planners.