Model to predict the level and the temporal and spectral composition of the sound pressure in urban soundscapes based on neural networks

Abstract

Environmental noise is a factor with a significant impact on personal well-being. One of the principal challenges of urban planning is the creation of soundscapes capable of providing inhabitants with a high quality of life. In the framework of the European legislation, the urban planners need tools that allow approaching the final goal of the design, planning and assessment of the soundscapes to adapt them to the needs required by the population. Nowadays, competent authorities have models of LAeq prediction to undertake this task, nevertheless, is not enough only with the analysis of the A-weighted equivalent sound pressure level (LAeq), but it is necessary to analyze the temporal and spectral composition of the sound pressure of the considered soundscape. For this reason, in this work a model for the prediction of the A-weighted (LAeq) and no weighted (Leq) equivalent sound pressure level, the difference between both descriptors, the temporal sound level variability, the impulsiveness of the sound level, as well as the sound level for each of the 1/3-octave bands between 20-20000 Hz, has been developed. For that purpose, due to the great urban complexity, with a wide range of variability in the great amount of relevant acoustic variables, together with the great amount of input and output variables in the model we have based our model in a backpropagation neural network, with which a prediction with an average mistake of 1.62±1.63 dB and an average factor r2 0.87 has been obtained.