Bi-BLS: A Bidirectional Connection Broad Learning System Model for Traffic Flow Prediction

Real-time and accurate short-term traffic prediction enables efficient operation of road traffic via real-time vehicle status information. However, the traffic prediction model based on deep learning or machine learning is generally not efficient due to the long training time. Thus, we propose a Bidirectional Connection Broad Learning System (Bi-BLS) model to characterize the relationship between road flow and vehicle speed, so as to perform real-time and short-term traffic prediction. The Bi-BLS consists of two parts, namely the mapped feature nodes layer and the enhancement nodes layer. The mapped nodes layer employs cascade connection to extract hierarchical features from the dataset by edge computing servers. The enhancement nodes layer uses bidirectional connection to extract hierarchical features from the mapped nodes layer. Moreover, it can efficiently add enhancement nodes or mapped feature nodes. To validate the effectiveness of our proposed model, the simulations are conducted based on the England Freeway Dataset. Numerical results show that our proposed model has better prediction performance in short-term traffic prediction than other benchmark models.