Please use this identifier to cite or link to this item: http://hdl.handle.net/10397/7467
Title: Prediction of noise generated from road traffic in urban environment using neural networks
Authors: Chu, Yiu Chung
Keywords: Traffic noise.
Noise control.
Hong Kong Polytechnic University -- Dissertations
Issue Date: 2014
Publisher: The Hong Kong Polytechnic University
Abstract: Noise pollution casts a serious threat to human health in modern society, and traffic noise is one of the major contributors to urban noise. By monitoring and predicting the traffic noise level in our environment, we can ensure that citizens are not subject to unhealthy noise hazards when building roads or performing urban planning. Traffic noise prediction provides information about the unknown noise levels due to a new establishment. In this study, the performances of prediction methods on traffic noise level are analyzed. Traditionally, the Calculation of Road Traffic Noise (CRTN) method is used for predicting traffic noises. However, a new artificial intelligent method using artificial neural networks (ANN) is used in this study. The purpose of this investigation is to study the use of neural networks in the traffic noise prediction by investigating what would be the best network configuration for the prediction of road traffic noise, and by comparing the predictions from the traditional CRTN method and neural network method. A sensitivity analysis of using neural network on the prediction of traffic noise is also performed. With ANN, input parameters are used to predict the output. In this study, traffic flow parameters are used as inputs to predict traffic noise levels (as outputs). The performances of a number of commonly adopted net configurations in the prediction of traffic noise are tested. In this research, noise level data measured at various Hong Kong heavy road traffic sites are used. The 56 data sets used in this study are provided by Professor S.K. Tang of Department of Building Service Engineering of The Hong Kong Polytechnic University.
These data are pre-processed and part of them is used for the artificial neural network learning and testing. The predicted values are then compared with the measured noise data not involved in the network training to find the mean absolute percentage error (MAPE). These MAPEs are then used to determine the best net configuration, and then based on the best net configuration, a sensitivity analysis on the effect of input data on the ANN prediction is carried out. It is found that the best net configuration for L₁₀ and Leq is that with 1 hidden layer and 2 nodes using Radial Basis Function (RBF) activation net method and the best net configuration for L₅₀ and L₉₀ are that with 3 hidden layers and 2 nodes using the RBF activation net. For sensitive analysis, the more sensitive parameters for L₁₀, Leq, L₅₀ and L₉₀ are as follows. For L₁₀, they are receiver height [m], distance from kerb [m]; for Leq, the mean vehicle speed (kph) and percentage of heavy vehicles [%]; for L₅₀, the mean vehicle speed (kph), percentage of heavy vehicles [%], receiver height [m], distance from kerb [m] and traffic flow [Veh/hr]; and for L90, the traffic flow [Veh/hr], mean vehicle speed (kph), distance from kerb [m] and receiver height [m]. Another main purpose of this thesis is also to develop an optimized scheme to develop the best ANN configurations. It is by first find out the optimized number of layers, optimized number of nodes and the best activation functions then use ANN matrix to find the best ANN configurations.
Description: xiii, 109 leaves : illustrations ; 30 cm
PolyU Library Call No.: [THS] LG51 .H577M BSE 2014 Chu
URI: http://hdl.handle.net/10397/7467
Rights: All rights reserved.
Appears in Collections:Thesis

Files in This Item:
File Description SizeFormat 
b27804999_link.htmFor PolyU Users 203 BHTMLView/Open
b27804999_ir.pdfFor All Users (Non-printable)8.48 MBAdobe PDFView/Open
Show full item record

Page view(s)

118
Last Week
3
Last month
Checked on Apr 23, 2017

Download(s)

44
Checked on Apr 23, 2017

Google ScholarTM

Check



Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.