Car-following behavior models based on conventional mathematical models cannot adequately reproduce traffic phenomena, such as traffic breakdown, capacity drop, and oscillations, and require parameter setting. Therefore, this study aims to construct a highly accurate car-following behavior model using deep learning. We evaluated the influence of variables in the dataset using a random forest. Furthermore, we constructed models to predict the acceleration in one second using the deep learning methods, deep neural network, long short-term memory, one-dimensional convolution neural network (1DCNN), and 2DCNN models. The models were evaluated using root mean square error, MAE, yyplot, and loss plot. The results showed that spatiotemporally structuring the data increased the accuracy of the predictions.
Data-driven Modeling of Car-Following Behavior on Freeways Considering Spatio-Time Effects: A Comparison of Different Neural Network Structures
Int. J. ITS Res.
International Journal of Intelligent Transportation Systems Research ; 21 , 1 ; 86-98
2023-04-01
13 pages
Article (Journal)
Electronic Resource
English
Car-following , Deep learning , Data-driven , Vehicle trajectory data , Freeways Psychology and Cognitive Sciences , Psychology , Engineering , Electrical Engineering , Automotive Engineering , Robotics and Automation , Computer Imaging, Vision, Pattern Recognition and Graphics , Civil Engineering , User Interfaces and Human Computer Interaction
Spatio-Temporal Evolution of Traffic Congestions on Urban Freeways
British Library Conference Proceedings | 2009
|Engineering Index Backfile | 1960