座席移動を伴う個人差空調システムに関する研究：制御ロジックが執務者の満足度に与える影響

神谷 洋介, 谷口 景一朗, 村上 弘晃, 佐藤 大輔, 松井 伸樹, 赤司 泰義
日本建築学会環境系論文集, 2025, 90 巻, 838 号, p. 617-626
https://doi.org/10.3130/aije.90.617

Abstract
We proposed a system that provides seat recommendations or air conditioning feedback based on user reports such “hot” or “cold.” This system was validated through experiments and simulations. As a result, we confirmed that control logic ensuring temperature variation, minimizing seat changes, and considering thermal dynamics significantly improves satisfaction. Additionally, we found that using large language models (LLMs) is crucial for considering various qualitative factors. Based on these insights, we developed a system integrating LLMs and deep reinforcement learning (DRL), which improved overall satisfaction by 31–35% compared to steady-state operation.