@article {Välisuo:2017:0736-2935:5667,
title = "Automated wind turbine noise analysis by machine learning",
journal = "INTER-NOISE and NOISE-CON Congress and Conference Proceedings",
parent_itemid = "infobike://ince/incecp",
publishercode ="ince",
year = "2017",
volume = "255",
number = "2",
publication date ="2017-12-07T00:00:00",
pages = "5667-5678",
itemtype = "ARTICLE",
issn = "0736-2935",
url = "https://ince.publisher.ingentaconnect.com/content/ince/incecp/2017/00000255/00000002/art00083",
author = "V{\"a}lisuo, Petri Olavi",
abstract = "Noise from the wind turbines has become increasingly a hindrance for land based wind power development. The trust of the public to the given sound power levels of the turbines and the propagation modelling is not always very good. Therefore, more measurements are needed, both to understand
the sound source and propagation better and to find out the real sound levels near dwellings. The sound measurements need to be carried out over a long period to cover representative set of different weather conditions, and often the recorded sound is contaminated by the sound from other sources
than wind turbines. For these reasons, the sound recording has been an expensive and tedious process, including manual sound analysis. The purpose of this study is to find out if the sound analysis can be automatized by using sound classification methods. The classification begins with, by
the selection of sound features, which can be for example sound pressure level (SPL), octave band SPL, amplitude modulation index, tonality index and such. Having selected the features, the machine learning algorithms are trained to sort the samples in different categories, based on the classification
rules. After classification, the sound pressure levels, and other parameters can be calculated separately for each class. The aim is that noise from different sources is in different classes and the wind turbine noise can be analysed separately from other noise sources. In this study, the
focus is in optimizing the features needed for classification, to make the processing faster and to allow the classification system to generalize the classification rules for varying conditions. The method is applied to a long term wind turbine noise measurement data set recorded over 12 months
in 2016 and 2017 in Finland.",
}