본문 바로가기
  • Home

Accurate prediction of lane speeds by using neural network

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2023, 28(5), pp.9-15
  • DOI : 10.9708/jksci.2023.28.05.009
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : January 6, 2023
  • Accepted : May 24, 2023
  • Published : May 31, 2023

Dong hyun Pyun 1 Changwoo Pyo 1

1홍익대학교

Accredited

ABSTRACT

In this paper, we propose a method predicting the speed of each lane from the link speed using a neural network. We took three measures for configuring learning data to increase prediction accuracy. The first one is to expand the spatial range of the data source by including 14 links connected to the beginning and end points of the link. We also increased the time interval from 07:00 to 22:00 and included the data generation time in the feature data. Finally, we marked weekdays and holidays. Results of experiments showed that the speed error was reduced by 21.9% from 6.4 km/h to 5.0 km/h for straight lane, by 12.9% from 8.5 km/h to 7.4 km/h for right turns, and by 5.7% from 8.7 km/h to 8.2 km/h for left-turns. As a secondary result, we confirmed that the prediction accuracy of each lane was high for city roads when the traffic flow was congested. The feature of the proposed method is that it predicts traffic conditions for each lane improving the accuracy of prediction.

Citation status

* References for papers published after 2023 are currently being built.