Short-term traffic speed prediction is an important technique for advanced traffic management systems. In this paper, a deep architecture combining convolutional neural networks (CNNs) and restricted Boltzmann machines (RBMs) with Gaussian units (GRBM-CNN) is proposed to predict short-term traffic speed; it combines the advantages of RBMs and CNNs and achieves a powerful capability to learn complex features of traffic data. Furthermore, this paper uses a graphics processing unit (GPU) to accelerate the learning process. Experimental results show that the proposed deep architecture has a lot of advantages over state-of-arts on the short-term traffic speed prediction task.
A Deep Architecture Combining CNNS and GRBMS for Traffic Speed Prediction
17th COTA International Conference of Transportation Professionals ; 2017 ; Shanghai, China
CICTP 2017 ; 310-319
2018-01-18
Conference paper
Electronic Resource
English