Propuesta de un modelo para la predicción de ruido por tráfico rodado a partir de aprendizaje automático que ofrezca información potencialmente útil en los procesos de gestión ambiental del ruido en entornos urbanos de Bogotá
Fecha
Autores
Autor corporativo
Título de la revista
ISSN de la revista
Título del volumen
Editor
Compartir
Altmetric
Resumen
This research proposed a model based on machine learning for predicting road traffic noise for the city of Bogota. The model considers conditions typical of vehicular traffic in the city. The input variables of the model were: vehicle capacity, speed, type of flow and number of lanes. Measurement campaigns involving audio and video recordings were carried out to obtain the input data. The audio recordings allowed the calculation of the noise levels through software processing, since they were taken with a measuring microphone calibrated at a height of 4 meters. On the other hand, the video data were used to count and classify the number of vehicles in four categories: motorcycles, light, medium and heavy vehicles. This process was done using a classifier trained with images of vehicles taken in the field and from free databases. Similarly, a processing algorithm based on an image classifier was used to estimate the speed of the vehicles from the video data. Then, the analysis of the measurements was carried out for some measuring points characterizing the noise emission of vehicle categories, arterial and secondary roads, traffic situations and pavements. Then, through exploratory data analysis, correlations were found, and a regression study was performed between noise levels and predictor variables. To determine the machine learning algorithm to be used, five models were compared, configured with their respective hyperparameters obtained through mesh search. The results showed that the Multilayer Perceptron (MLP) regression had the best fit with MAE = 0,86 dB for the test dataset. Finally, the proposed MLP regressor was compared with classical statistical models for traffic noise prediction. In conclusion, the MLP regressor obtained the best error and fit indicators with respect to statistical models.