2024년 5월 13일 진행

개요:

Our research begins with the question of whether artificial intelligence can understand the mechanism of traffic noise through data generated by a physical model and predict the results. The typical noise prediction model consists of a noise generation function that takes inputs such as traffic volume, speed, and vehicle type, and an energy attenuation function due to structural surfaces, transmission medium like air, etc. We construct a Convolutional Neural Network (CNN) regression model that learns noise predictions from multi-channel images representing the information used as inputs in traditional noise prediction models. Applying this CNN regression model to predict the average noise levels of peak-over time in Gwangju, South Korea, we observed significant improvement in generalized predictive performance compared to existing machine learning models that summarize input values on a grid basis. Through the artificial neural networks, we successfully reproduced the physical model of noise prediction. and efficiently propagated noise prediction values. Additionally, we investigate model structures and database construction methods using CNN to enhance noise prediction performance.