Transfer Probability Prediction for Traffic Flow with Bike Sharing Data: A Deep Learning Approach

Abstract

As the fourth generation sharing bike, the dockless sharing bike is equipped with an electronic lock on the rear wheel. Since the dockless sharing bike system does not require the construction of specific infrastructures, it has become one of the important travel modes for residents. The distribution pattern of the use of shared bikes can capably represent the travel demands of the residents. However, since the number of sharing bikes is very large, the trajectory data that contains a large number of sparsely distributed data and noise. It results in high computational complexity and low computational accuracy. To address this problem, a novel deep learning algorithm is proposed for predicting the transfer probability of traffic flow of Shared Bikes. A stacked Restricted Boltzmann Machine (RBM)-Support Vector Regression (SVR) deep learning algorithm is proposed; a heuristic and hybrid optimization algorithm is utilized to optimize the parameters in this deep learning algorithm. In the experimental case, the real shared bikes data was used to confirm the performance of the proposed algorithm. By making comparisons, it revealed that the stacked RBM-SVR algorithm, with the help of the hybrid optimization algorithm, outperformed the SVR algorithm and the stacked RBM-SVR algorithm.