Loading...

Prediction of Link Traffic Volume in Urban Road Networks Using Cell Phone Data

Alian, Ramin | 2020

509 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 53653 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Shafahi, Yousef
  7. Abstract:
  8. In the development of intelligent transportation systems for customer use and resource management, accurate prediction of traffic volume on links across urban road network is critical. Since road traffic is the result of a complex interplay between traffic demand and traffic supply features, traffic volume prediction is a complex non-linear operation that has been the subject of much research over the past few decades. The two popular methods to trace people’s location during a day are the Global Positioning System (GPS) technology and commuters voluntary surveys. The two approaches bring significant benefits, yet hold limitations in their applications. In this research, a new method is proposed and tested for the use of available cellular data to predict traffic volume on links across urban road networks. The ability to collect big data without the need for any active participation of the users can be considered the main advantage of this practice over the other methods. Additionally, the rapid velocity of data collection in this method brings the opportunity to produce short-term predictive models. This research examines the spatio-temporal data gathered from the cellular network in the city of Shiraz, Iran. The data captures the location of antennas to which 300,000 random cellular network users are connected and collected every 5 minutes in a time span of 44 hours in two consecutive days. In this case study, the value of observed traffic volume in every 15-minute interval in some links in the Shiraz road network was used as the ground truth for the target value. After reviewing the application of the conventional methods for predicting the traffic volume on links across road networks, the data-driven method and neural network tool were selected as the proposed method. In the proposed method, for each network link and its known target value, useful features are first selected and prepared from the cell phone data for use in the model input. The neural network model is then trained with the prepared data. Lastly, the model will be able to perform a short-term prediction of traffic volume for each link of the network using real-time cell phone data. More precisely, in our model, the traffic state of the urban road network (link traffic volume) is predicted according to the location of the users. For the case study, the weighted absolute percentage error (WAPE) in the best-trained model was 22.27 percent
  9. Keywords:
  10. Neural Networks ; Cellular Network Data ; Spatiotemporal Dataset ; Link Traffic Volume Prediction ; Cellular Network Features ; Data Driven Modeling

 Digital Object List

 Bookmark

No TOC