Vehicular traffic noise prediction and propagation modelling using artificial neural network

Noise is a sound of variable frequencies considered as one of the leading causes of environmental challenges faced in many cities due to high traffic volume and has a harmful effect on the population. Discomforting issues such as interference with communication, speech, effects on attention, people’s health and well-being, psychological and cardiovascular alterations are some of the major disturbances caused to our environment. This thesis presents a Neural Network (NN) model developed to predict and simulate the propagation of vehicular traffic noise in a dense residential area at the New Klang Valley Expressway (NKVE) in Shah Alam Seksyen 13, Malaysia. The proposed model comprises of two main simulation steps: i) the prediction of the vehicular traffic noise using NN in order to obtain the final noise maps for weekends and weekdays; ii) The simulation of the propagation of the traffic noise emission in the study area using a mathematical model to define the propagation of the study area. By utilizing the Chi-square statistical analysis, the former model was developed with six selected noise predictors. These predictors include the number of motorbikes, the sum of vehicles, car ratio, large vehicles ratio (truck, lorry, and bus), highway density, and a LiDAR derived Digital Surface Model-DSM. The neural network and its hyperparameters were optimized through a systematic optimization procedure based on a grid search approach. In contrast, the noise propagation model was developed based on principle concepts of traffic noise. This model was based on road geometry, barriers, distance, the interaction of air particles, and weather parameters which are applied to Geographic Information System (GIS). The noise measurement was carried out continuously at 15-min intervals and the data were analyzed by taking the minimum, maximum, and averages of every data set recorded during the day. The measurement was carried out four times a day (morning, afternoon, evening, and midnight) all through two-days of the week (Sunday and Monday). The optimal radial basis function NN model was used which comprised of 17 hidden layers with a back-propagation algorithm. The learning rate of 0.05 and a momentum of 0.9 were used in this experiment. The results showed that the proposed NN model achieves a validation accuracy of 78.4% and an error in noise prediction with less than 4.02 dB. The model also outperforms the Multilayer Perceptron (MLP) model by almost 5% of validation accuracy and 0.3 dB in noise level prediction. In addition, the three most influential parameters on traffic noise were car ratio, the sum of vehicle, and large vehicle ratio. Overall, the proposed models were found to be promising tools for traffic noise assessment in dense urban area of the study area.

Preview Text FK 2018 63 IR.pdf Download (2MB) | Preview

Actions (login required)

View Item