Vehicular traffic noise modelling of urban area-a contouring and artificial neural network based approach.

Road traffic vehicular noise is one of the main sources of environmental pollution in urban areas of India. Also, steadily increasing urbanization, industrialization, infrastructures around city condition causes health risks among the urban populations. In this study, we have explored noise descriptors (L, L, L, LNI, TNI, NC), contour plotting and find the suitability of artificial neural networks (ANN) for the prediction of traffic noise all around the Dhanbad township in 15 monitoring stations. In order to develop the prediction model, measuring noise levels of five different hours, speed of vehicles, and traffic volume in every monitoring point have been studied and analyzed. Traffic volume, percent of heavy vehicles, speed, traffic flow, road gradient, pavement, road side carriageway distance factors were taken as input parameter, whereas L as output parameter for formation of neural network architecture. As traffic flow is heterogenous which mainly contains 59%, two wheelers and different vehicle specifications with varying speeds also affect driving and honking behavior which constantly changing noise characteristics. From radial noise diagrams shown that average noise levels of all the stations beyond permissible limit and the highest noise levels were found at the speed of 50-55 km/h in both peak and non-peak hours. Noise descriptors clearly indicate high annoyance level in the study area. Artificial neural network with 7-7-5 formation has been developed and found as optimum due to its sum of square and overall relative error 0.858 and .029 in training and 0.458 and 0.862 in testing phase respectively. Comparative analysis between observed and predicted noise level shows very less deviation up to ± 0.6 dB(A) and the R linear values are more than 0.9 in all five noise hours indicating the accuracy of model. Also, it can be concluded that ANN approach is much superior in prediction of traffic noise level to any other statistical method.