Acoustical Noise Analysis and Prediction by means of Multiple Seasonality Time Series Model

Physical polluting agents are a relevant problem in urban areas. The need for monitoring and prediction of their time evolution is very useful to assess the impact to human health and activities. Considering their effects on health, the most hazardous agents to be considered are air pollution, acoustical noise and electromagnetic fields. Regarding acoustical noise, the complexity of predicting its slope is strongly correlated to its intrinsic randomness, related to the great variability of the sources. Sometimes, in some special areas, the predominant sources are stationary or have a periodic behaviour. In these cases, a time series analysis approach can be adopted, considering that a general trend and a local periodicity can be highlighted and used to build a predictive model. In particular, in this paper, the model is built composing three parts: the trend, that is the long term behaviour, the seasonality, that is the periodic component, and the irregularity, that includes the random variations. Applying such a model to a traffic noise levels dataset, obtained from a site in the city of Messina, Italy, a multiple seasonality is evidenced, resulting in two seasonal coefficients introduction (low frequency and high frequency). The validation of the presented model will be performed on a 44 days dataset, not used in the calibration. Results will be encouraging and will show a very good prediction performances of the model, especially in terms of difference between observed and simulated values (error). The error distributions will be analyzed and discussed by means of statistical indexes, plots and tests.