Passenger Flow Forecast of Rail Station Based on Multi-Source Data and Long Short Term Memory Network

Abstract

The existing rail station passenger flow prediction models are inefficient, due to that most of them use single-source data to predict. In this paper, a novel method is proposed based on multi-layer LSTM, which integrates multi-source traffic data and multi-techniques (including feature selection based on Spearman correlation and time feature clustering), to improve the performance of predicting passenger flow. The experimental results show that the multi-source data and the techniques integrated in the model are helpful, and the proposed method obtains a higher prediction accuracy which outperforms other methods (e.g. SARIMA, SVR and BP network) greatly.

Keywords

• Rail transit passenger flow
• prediction model
• long short term memory network
• multi-source data
• spearman correlation
• K-means