Deep learning methods for long term traffic flow forecasting

Effective long-term forecasting of traffic flow has many applications for Intelligent Transportation Systems. With more historical measurements being available, data-driven methods are becoming promising for conducting this forecasting task. Existing research focuses on prediction for up to 1 day, yet there is an emerging demand for a longer forecasting horizon. This paper first reviews representative data-driven methods for traffic forecasting. Subsequently, the promising Sequence to Sequence (Seq2Seq) deep-learning methods are evaluated on the long-term forecasting up to 14 days ahead. Their performances on 1 day, 7 days and 14 days ahead traffic flow forecasting are compared and discussed. Then, the impacts of the shrinking size of training data and holiday traffic are investigated. Based on a real-world and large dataset from Melbourne, Australia, test results indicate that a state-of-the-art Transformer-based method, Informer, generates superior results than Seq2Seq RNN, LSTM, and GRU, especially when the forecasting span is extended. Besides, although the accuracy of Informer slightly degrades with the decreasing size of training data, 2 months of training data seem enough for it to produce a decent performance. In addition, the investigation of holiday traffic reveals its evident impact on forecasting accuracy.

• Record URL:
  https://australasiantransportresearchforum.org.au/papers/2022
• Authors:
  • Zheng, X
  • Sarvi, M
  • Bagloee, S A
• Conference:
  • Australasian Transport Research Forum (ATRF), 43rd, 2022, Adelaide, South Australia, Australia
• Publication Date: 2022-9

Media Info

• Pagination: 15p
• Monograph Title: Australasian Transport Research Forum, 28-30 September, Adelaide, South Australia

Subject/Index Terms

• TRT Terms: Accuracy; Data analysis; Data collection; Holidays; Mathematical models; Traffic flow; Traffic surveillance
• Uncontrolled Terms: Traffic fluctuation
• Geographic Terms: Melbourne, Victoria
• ATRI Terms: Accuracy; Data analysis; Holiday period; Modelling; Traffic flow; Traffic fluctuation; Traffic monitoring
• ITRD Terms: 6471: Analysis (math); 8735: Intelligent transport system; 673: Traffic concentration
• Subject Areas: I71: Traffic Theory;

Filing Info

• Accession Number: 01895129
• Record Type: Publication
• Source Agency: ARRB Group Limited
• Files: ITRD, ATRI
• Created Date: Oct 2 2023 11:29AM