Passenger flow predictions is of great significance to bus scheduling and route optimiza-tion. In this paper, a novel algorithm, namely, Bi-directional Long Short-Term Memory with Attention Mechanism (Bi-LSTM-AT) are proposed to predict transit passenger flow. We utilize Bi-LSTM structure with attention mechanism to capture the spatiotemporal features, meanwhile, take into account external factors that affect passenger choices. We conducted a experiment using field data collected at Urumqi, china. The prediction results show an averaged absolute error (MAE) as low as 3.75, which demonstrated the feasibility of applying Bi-LSTM-AT in transit passenger flow forecasting.

CICTP