PROJECT TITLE :

Short-Term Traffic Flow Forecasting Method With M-B-LSTM Hybrid Network

ABSTRACT:

Recently, good results in short-term traffic forecasting have been achieved through the use of Deep Learning. Nevertheless, the traffic flow is characterized primarily by its stochasticity and distribution imbalance; both of these aspects will bring uncertainty and cause a problem with network overfitting during the process of deep learning. In order to address the issues, the authors of this paper propose an innovative end-to-end hybrid deep learning network model for short-term traffic flow forecasting. The model is given the name M-B-LSTM. An online self-learning network is built as a data mapping layer in the M-B-LSTM model. The purpose of this network is to learn and equalize the traffic flow statistic distribution in order to mitigate the effects of distribution imbalance and the overfitting problem that arises during the process of network learning. In addition to this, the deep bidirectional Long Short-Term Memory network, also known as DBLSTM, is implemented in the stochasticity reducing layer in order to reduce the uncertainty problem through a forward and reverse contexts approximation process. After this, the long short-term memory network, also known as LSTM, is implemented in the forecasting layer in order to predict the subsequent state of the traffic flow. In addition, an adequate number of comparative experiments have been carried out, and the findings indicate that the proposed model is superior to the state-of-the-art methods in terms of its capacity to solve problems involving uncertainty and overfitting.