Cell-Speed Prediction Neural Network (CPNN): A Deep Learning Approach for Trip-Based Speed Prediction

Travel speed prediction is a valuable tool used to help manage traffic operations as well as to help travelers make decisions regarding their trips. Traditionally, travel speed prediction is based on the road segment level or a larger network level. This paper presents a novel trip-based perspective for travel speed prediction. The traditional approaches are useful for transportation managers, but this new approach is also potentially useful for travelers. A Modular Plug-in Deep Neural Network, based on the combination of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) recurrent neural network are used to achieve speed predictions at the trip level, for any given point in space and time. The network created is called the Cell-Speed Prediction Neural Network and includes a unique Feature Extraction Module that is used to capture features of each trip cell. The study uses data provided by Didi GIYA and the results are promising, with the mean absolute error for each trip cell being about 2.56 meters per second. This novel model also performs better than other state of the art approaches to trip-based speed prediction.Keywords: deep learning, speed prediction, spatial temporal prediction, trip cell, long short-term memory neural network

  • Supplemental Notes:
    • This paper was sponsored by TRB committee ABJ70 Standing Committee on Artificial Intelligence and Advanced Computing Applications. Alternate title: Cell-Speed Prediction Neural-Network (CPNN): A Deep-learning Approach for Trip Cell Speed Prediction in Metropolitan Road Networks.
  • Corporate Authors:

    Transportation Research Board

  • Authors:
    • Yang, Hao
    • Liu, Chenxi
    • Gottsacker, Christopher
    • Ban, Xuegang
    • Zhang, Chao
    • Wang, Yinhai
  • Conference:
  • Date: 2019


  • English

Media Info

  • Media Type: Digital/other
  • Features: Figures; References;
  • Pagination: 6p

Subject/Index Terms

Filing Info

  • Accession Number: 01697450
  • Record Type: Publication
  • Report/Paper Numbers: 19-02492
  • Files: TRIS, TRB, ATRI
  • Created Date: Dec 7 2018 9:27AM