Physics-Informed Transformer Framework for Indoor Acoustic Comfort Optimization in Buildings

Improving indoor acoustic comfort in buildings remains challenging for conventional acoustic design methods because they often rely on empirical formulas, static databases, and limited adaptability to dynamic environmental changes. To address these limitations, this study proposes a physics-informed Transformer framework for indoor acoustic comfort optimization in buildings. The proposed framework integrates measured, simulated, and publicly available acoustic data to construct an adaptive dataset for indoor acoustic modeling. A hierarchical time–frequency attention mechanism is introduced to capture complex acoustic features, while a physics-informed module is incorporated to improve the physical consistency of the predicted acoustic responses. In addition, a dynamic optimization strategy is developed to adjust acoustic parameters under changing indoor conditions while considering both acoustic performance and energy efficiency. Experimental results show that the proposed framework achieves a Mel-spectral distortion of 6.12 dB and a speech intelligibility score of 89.3%, while reducing the prediction errors of reverberation time and early decay time to 0.11 s and 0.09 s, respectively. In practical deployment tests, the proposed method increased the early sound energy ratio by 46.4%, improved reverberation time control accuracy by 41.2%, reduced total energy consumption by 29.7%, and increased the user acoustic comfort score by 38.1%, with an average response time of 1.8 s. These results demonstrate that the proposed framework provides an effective data-driven and physics-informed approach for improving indoor acoustic comfort and energy-efficient acoustic management in buildings.