Traffic Flow Prediction Based on Cooperative Vehicle Infrastructure for Cloud Control Platform Wuhan University of Technology, China

Format:

Book

Conference/Event

Author/Creator:

Chen, Zhijun, author.

Contributor:

Chen, Chen

Chen, Qiushi

Dong, Yuhuan

Yang, Shuai

Zhang, Jingming

Zhang, Yishi

Conference Name:

3rd International Forum on Connected Automated Vehicle Highway System through the China Highway & Transportation Society (2020-10-29 : Jinan, China)

Language:

English

Physical Description:

1 online resource cm

Place of Publication:

Warrendale, PA SAE International 2020

Summary:

Under the cloud control platform in the Cooperative Vehicle Infrastructure System (CVIS), the traffic flow prediction of a single time step is not sufficient for the traffic control and traffic induction needs of nowadays. Accurate prediction of traffic flow in multiple time steps can provide more information for traffic guidance and travel route planning. So, it is necessary to explore effective methods of multi-time-step traffic flow prediction. In addition, traffic flow data has low dimensionality. There are potential correlations among the features of input data, which may be difficult to mine if the data are fed directly into the prediction model. To address these issues, a hybrid model of Autoencoder and LSTM-based Sequence-to-Sequence (Seq2Seq) model is proposed in this paper, which named as AE-Seq2Seq. AE-Seq2Seq excels at the task of traffic flow prediction in multiple time steps. The autoencoder in the proposed hybrid model can expand the dimensions of low-dimensional traffic flow data to expose more potential information hidden in the input features. Meanwhile, the LSTM-based Seq2Seq model can capture the long-term dependence of traffic data and the sequential relationship between the output data, thus effectively predicting the traffic flow with multiple time steps. Two deep learning models (the Multi-Layer Perceptron and LSTM) and four machine learning methods (Support Vector Regression, Random Forest, XGBoost, and Linear Regression) are employed in our comparison experiment to demonstrate the superiority of the proposed method. The experimental results show that the proposed method obtains a lower error in the prediction of each time step; the performance of AE-Seq2Seq is not significantly degraded for longer time step predictions. Therefore, the superiority of the proposed model in the multi-time-step prediction tasks has been verified

Notes:

Vendor supplied data

Publisher Number:

2020-01-5182

Access Restriction:

Restricted for use by site license