Citation

Hsueh YL, Yang YR. Int. J. Intell. Transp. Syst. Res. 2021; 19(3): 510-524.

Copyright

(Copyright © 2021, Holtzbrinck Springer Nature Publishing Group)

DOI

10.1007/s13177-021-00260-7

PMID

unavailable

Abstract

To successfully deploy an intelligent transportation system, it is essential to construct an effective method of traffic speed prediction. Recently, due to the advancements in sensor technology, traffic data have experienced explosive growth. It is therefore a challenge to construct an efficient model with highly accurate predictions. To improve the accuracy and the efficiency of short-term traffic predictions, we propose a prediction model based on deep learning approaches. We use a long short-term memory (LSTM) network to analyze sequential sensor data to predict the car speed of the next time interval on the freeway. Unlike the traditional model that only considers the changes in traffic speed which is used to derive the temporal and spatial features from the prediction road section, we mainly consider the features of the number of the most representative car types and the traffic speed variation of the front road segment that is ahead of the prediction road segment in addition to the number of cars, the road occupancy, and the traffic speed latency to successfully learn and capture the hidden patterns from the sensor data so as to improve the prediction accuracy. To the best of our knowledge, very few investigations have been conducted to consider the correlation between car speed and car type for a prediction model. Moreover, our extensive experiments demonstrate that the proposed method for traffic speed prediction has achieved high accuracy.

Language: en