Deep learning based prediction of urban air mobility noise propagation in urban environment

Abstract

A deep learning based method is proposed to predict the urban air mobility (UAM) noise propagation in the urban environment. This method aims to efficiently estimate the noise impact of UAM flights on the complex urban area. The noise hemisphere was created via the comprehensive multirotor noise assessment framework to determine the noise level of UAM. The noise propagation to a randomly generated three-dimensional (3D) urban area was then calculated using the ray tracing method, including atmospheric attenuation and multiple reflections. 45 000 two-dimensional noise maps were used to train and evaluate the modified convolutional neural network. The results demonstrated high accuracy, with a root mean square error of only 2.56 dB compared to the ray tracing method, while reducing computation time by more than 1800 times. This model was applied to analyze the noise impact of various UAM flight conditions and landing scenarios at a vertiport. This deep learning approach is a fast method with adequate accuracy for predicting UAM noise impact in 3D urban environments. Also, it can inform the development of noise based strategies for UAM operations.