Prediction of annoyance evaluations of electric vehicle noise by using artificial neural networks

Abstract

Road traffic noise is the most annoying form of environmental noise pollution. The enforcement of using artificially generated noise for electrically powered vehicles is currently on the rise. Regarding to that, it is important to generate sounds regarding to the regulations which are not annoying. While many annoyance models are available around the world, these models cannot be simply generalized for these new sounds and while it is very time consuming to measure the annoyance for each newly generated noise with an listening test, the idea came up to use artificial neural networks instead. The aim of this work is to determine the annoyance of different electric vehicle sounds for a constant speed, single car pass-by situation. For this purpose, the differences in annoyance are investigated with perception studies. The correlation between physical-psychoacoustical parameters and annoyance estimations obtained from jury testing is also investigated in this study. Moreover, an artificial neural network (ANN) is also used as a prediction tool of annoyance estimations for further evaluations of different possible stimuli. Overall, a total of 150 ANN models with different hidden layers were undertaken in this research. The best-performing models were compared with linear regression models based on psychoacoustic parameter. Lastly, advantages and shortcomings of using ANNs for detectability estimations are also discussed.