Physics-Informed Modeling and Deep Learning Prediction of Railway Noise and Vibration Propagation

Lvbai Hang^*

• China Construction Civil Construction Co., LTD, Beijing, China

Environmental noise and vibration are critical challenges in modern urban settings, particularly in transportation systems where their propagation impacts public health and infrastructure durability. Traditional approaches to modeling these phenomena rely heavily on analytical solutions or empirical formulations, which often oversimplify complex propagation dynamics and fail to integrate site-specific conditions. To address these limitations, we introduce a physicsinformed framework that integrates numerical simulations with deep learning for precise noise and vibration prediction. Our method combines domain knowledge in wave mechanics with advanced neural networks to model multi-scale interactions and capture environmental and structural variability. By incorporating physics-based constraints into the learning process, the framework enhances interpretability and ensures physically plausible predictions. Experimental evaluations demonstrate superior accuracy over baseline models, particularly in scenarios involving heterogeneous materials and irregular geometries. This integration of physics-informed modeling and machine learning offers a robust tool for sustainable urban design and proactive mitigation strategies.