A framework for community noise modelling using machine learning methods

A framework for community noise modelling is proposed, where noise sources are treated as their equivalent noise sources defined in a multidimensional space. The noise levels at measurement locations under different settings of noise sources in the space are predicted using a noise propagation simulator and the method of Design of Computer Experiments (DoEs). A surrogate model is built using a Gaussian process for machine learning where the predictions are used as training data. Sound power levels of the equivalent noise sources are obtained by means of measured noise levels and the surrogate model. As a case study, six LA(eq-ihr) models from 2:30 pm to 8:30 pm of an outdoor food court and shopping area are reported to demonstrate the effectiveness and flexibility of the framework in noise prediction, noise source modelling and its verification. (C) 2019 Elsevier Ltd. All rights reserved.