Skip to main content
Sayed Omid
ُُDissertation Title
Short Term Traffic Forecasting in Urban Highways by Hybrid Wavelet Transform-Artificial Neural Network
Case Study
Dr. Meisam Akbarzadeh


In urban traffic management is required to predict short term traffic condition in urban highways. If we forecast near future traffic flow (a few minutes later) is possible to guidance and manage traffic flow with Variable Message Signs that installed along highways. Traffic flow condition influenced by different factor and constantly is changing, therefore usually it is not possible to forecast intuitive. This study aimed to develop a method to forecast short-term traffic flow by using the past traffic flow data. The method is proposed in this study is based on data volume in one of the United States urban highway in Los Angeles in California.

In this study by combination of Wavelet Transform (WT) and Artificial Neural Network (ANN), a hybrid method is provided with good accuracy to predict traffic this hybrid method the WT as a tool of signal processing, decomposes signal before ANN. This method extract the main traffic flow signal frequencies and bypass subsidiary frequencies. Main frequencies input to the ANN and trained by training algorithm and provide the output result. In this study we are using a step by step method to achieve to the optimal flow forecasting in the context of hybrid approach. Based on the result of the optimum model ten time delay inputs are selected and the WT decomposed them in two level by dmey mother wavelet. The structure of ANN is formed by two layer with four and 1 neurons. Purelin transfer function set between layers and the Levenberg-Marquardt training function set as a training algorithm. Result show hybrid WT-ANN method can estimate the signal traffic with good accrate.this method is able to predict the five minutes later traffic flow by 2.5 percent in the term of MAPE. Without using WT aside of   ANN the MAPE decrease into 13.7 percent.

Among statistics methods the least MAPE is dedicated to moving average method with an MAPE of 11.56. Developing the results of hybrid method for predicting vaster range shows a decrease in prediction accuracy, so that calculated MAPE with hybrid method reaches 26 in result of predicting for next 45 minutes.



General Classification
Type of person