Avaliação de modelos de previsão do ruído sonoro urbano com redes neurais artificiais

Abstract

The urban traffic noise is one of the biggest sources of noise pollution being composed by the different types of vehicles that circulate on the roads. Few studies have been carried out in search of a methodology capable of predicting urban noise based on the volume of traffic and several other variables present in the environment. This work aims to evaluate through comparisons the smallest noise prediction errors in urban traffic using the MLP, ELM and RBF Artificial Neural Networks models against the linear Multiple Linear Regression model. The data used in this work are the result of a literature review on the subject and databases collected in the cities of Goiânia-GO and Maringá-PR. The results found by the networks are compared in terms of their average errors to evaluate the best mappers of the equivalent sound level Leq, which is the output of the models and represents the noise produced by traffic. The neural models in this study proved to be better Leq mappers when compared to the linear model.