Hybrid Computational Fluid Dynamics (CFD) and Machine Learning for Airflow Simulation Over Coastal Terrain

Abstract

In recent years, Physics Informed Neural Networks (PINNs) have emerged as a powerful tool for solving complex partial differential equations (PDEs) governing physical phenomena. While numerous studies have explored the theoretical aspects of PINNs using benchmark problems, their application to real-world data and simulations remains limited. In this paper, we employ the PINNs method to simulate fluid flow over coastal dunes in the Netherlands, addressing a real-world problem. We aim to test the usability and performance of Physics Informed Loss by comparing it to similar models without this loss. Our results demonstrate that models incorporating Physics Informed Loss do not improve performance within the training data bounds but do show enhanced generalization to unseen data.