A Bidirectional Unidirectional Graph Convolutional Stacked LSTM Neural Network for Metro Ridership Prediction

PROJECT TITLE :

A Graph Convolutional Stacked Bidirectional Unidirectional-LSTM Neural Network for Metro Ridership Prediction

ABSTRACT:

Forecasting the number of people using the metro in a timely and accurate manner is helpful in revealing the real-time demand for traffic, which is an essential but challenging task in modern traffic management. Deep Learning algorithms have been widely applied because of their superior performance in capturing spatio-temporal features. This is due to the fact that riding behavior in a metro system is characterized by a complex spatial correlation as well as temporal variation. However, the current Deep Learning models make use of regular convolutional operations, which can hardly provide accurate results that are satisfactory due to either a lack of knowledge regarding the realistic topology of a traffic network or an inability to adequately capture representative spatiotemporal patterns. This study proposes a parallel-structured Deep Learning model that consists of a Graph Convolution Network and a stacked Bidirectional unidirectional Long short-term Memory network. The purpose of this model is to further improve the accuracy with which metro ridership predictions are made (GCN-SBULSTM). The GCN module views a metro network as a structured graph. In order to capture the dynamic spatial correlation among metro stations, a K-hop matrix is introduced. This matrix combines travel distance, population flow, and adjacency. The SBULSTM module takes into account both the backward and forward states of the ridership time series. Additionally, it is able to learn complex temporal features by stacking multiple recurrent layers. In order to demonstrate how well the proposed model works, experiments are run on three datasets that contain actual ridership information from metro systems. When compared to state-of-the-art prediction models, GCN-SBULSTM demonstrates superior performance across a variety of use cases and significantly improves the efficacy of training procedures.

Did you like this research project?

To get this research project Guidelines, Training and Code... Click Here