TY - GEN
T1 - Long Term Traffic Prediction in Highway Using Parallel CNN
AU - Lim, Donghyun
AU - Lee, Minhyeok
AU - Seok, Junhee
N1 - Funding Information:
ACKNOWLEDGMENT This work was supported by grants from Hyundai Next Generation Vehicle and the National Research Foundation of Korea (NRF-2019R1A2C1084778).
Publisher Copyright:
© 2020 IEEE.
PY - 2020/9
Y1 - 2020/9
N2 - For navigation system, predicting future traffic conditions is crucial. To predict the traffic condition, statistical methods and neural network models have been studied. However, conventional methods have three limitations in which only the temporal properties are used, only narrow sections or time steps are predicted and not general road sections such as all section of highway but specific sections are used as test results. This paper proposes a parallel Convolutional Neural Network (CNN) that uses spatiotemporal properties and predicts for the next five hours and up to 400 km ranges in Korea's representative highway. Using a highway dataset, the proposed parallel CNN is trained and evaluated. As a result, the result of our model is improved by 10.6%, in terms of Root Mean Square Error (RMSE), compared to the conventional method. Moreover, in terms of the average of Average Speed Difference (ASD), the result of our model is improved by 63.5%.
AB - For navigation system, predicting future traffic conditions is crucial. To predict the traffic condition, statistical methods and neural network models have been studied. However, conventional methods have three limitations in which only the temporal properties are used, only narrow sections or time steps are predicted and not general road sections such as all section of highway but specific sections are used as test results. This paper proposes a parallel Convolutional Neural Network (CNN) that uses spatiotemporal properties and predicts for the next five hours and up to 400 km ranges in Korea's representative highway. Using a highway dataset, the proposed parallel CNN is trained and evaluated. As a result, the result of our model is improved by 10.6%, in terms of Root Mean Square Error (RMSE), compared to the conventional method. Moreover, in terms of the average of Average Speed Difference (ASD), the result of our model is improved by 63.5%.
KW - deep learning
KW - forecast up to five hours ahead
KW - highway
KW - spatialoral properties
KW - traffic prediction
UR - http://www.scopus.com/inward/record.url?scp=85097234337&partnerID=8YFLogxK
U2 - 10.1109/ICITE50838.2020.9231436
DO - 10.1109/ICITE50838.2020.9231436
M3 - Conference contribution
AN - SCOPUS:85097234337
T3 - 2020 IEEE 5th International Conference on Intelligent Transportation Engineering, ICITE 2020
SP - 107
EP - 110
BT - 2020 IEEE 5th International Conference on Intelligent Transportation Engineering, ICITE 2020
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 5th IEEE International Conference on Intelligent Transportation Engineering, ICITE 2020
Y2 - 11 September 2020 through 13 September 2020
ER -