Scholarlink

<p>The objective of this study is to develop and evaluate a hybrid generative design framework that integrates Post-Occupancy Evaluation (POE) data into a conditional Generative Adversarial Network (GAN), with the aim of producing elementary school layouts that are both user-centered and energy-efficient. This responds to the challenge of balancing environmental performance with user experience in early-stage school design. Empirical data were collected from 384 users across several Tehran public schools. POE analysis identified ventilation (<i>r</i> = 0.586), lighting quality, and furniture/materials (<i>r</i> = 0.701) as the strongest predictors of satisfaction. Factor analysis revealed three latent dimensions: circulation clarity, sensory comfort, and spatial quality. These latent factors, combined with six spatial zoning masks (e.g., classrooms, offices, workshops), were used to condition the GAN. The model was trained for 500 epochs, and a feedback loop incorporating Structural Similarity Index (SSIM) and Mean Squared Error (MSE) was applied to refine outputs. The GAN produced layouts in which classrooms occupied up to 34% of floor area, emerging as the dominant zone. EnergyPlus simulations showed an average 13.5% annual energy savings compared to baseline school designs, with several high-performing layouts achieving improved user satisfaction, particularly in heating, cooling, and lighting performance. Design features such as adjacency of service zones, window-to-wall ratio, and classroom orientation had significant impacts on energy outcomes. The results demonstrate that integrating POE-derived user satisfaction metrics into a GAN framework extends prior cGAN-based layout work by embedding empirical measures of comfort and spatial quality into generative design. While building upon existing generative frameworks, this approach incrementally contributes a hybrid evaluation mechanism that links user feedback with energy performance, offering a scalable tool for early-stage school design that balances environmental objectives with user-centered needs.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Integrating post occupancy evaluation and energy performance metrics using conditional generative adversarial networks for intelligent school design

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月28日

查看全文
  • Marjan Ilbeigi,
  • Mohammad Ghomeishi,
  • Ali Asgharzadeh,
  • Nima Amani

摘要

The objective of this study is to develop and evaluate a hybrid generative design framework that integrates Post-Occupancy Evaluation (POE) data into a conditional Generative Adversarial Network (GAN), with the aim of producing elementary school layouts that are both user-centered and energy-efficient. This responds to the challenge of balancing environmental performance with user experience in early-stage school design. Empirical data were collected from 384 users across several Tehran public schools. POE analysis identified ventilation (r = 0.586), lighting quality, and furniture/materials (r = 0.701) as the strongest predictors of satisfaction. Factor analysis revealed three latent dimensions: circulation clarity, sensory comfort, and spatial quality. These latent factors, combined with six spatial zoning masks (e.g., classrooms, offices, workshops), were used to condition the GAN. The model was trained for 500 epochs, and a feedback loop incorporating Structural Similarity Index (SSIM) and Mean Squared Error (MSE) was applied to refine outputs. The GAN produced layouts in which classrooms occupied up to 34% of floor area, emerging as the dominant zone. EnergyPlus simulations showed an average 13.5% annual energy savings compared to baseline school designs, with several high-performing layouts achieving improved user satisfaction, particularly in heating, cooling, and lighting performance. Design features such as adjacency of service zones, window-to-wall ratio, and classroom orientation had significant impacts on energy outcomes. The results demonstrate that integrating POE-derived user satisfaction metrics into a GAN framework extends prior cGAN-based layout work by embedding empirical measures of comfort and spatial quality into generative design. While building upon existing generative frameworks, this approach incrementally contributes a hybrid evaluation mechanism that links user feedback with energy performance, offering a scalable tool for early-stage school design that balances environmental objectives with user-centered needs.