Short-Term Passenger Flow Prediction Model of Rail Transit Based on Inherit Long Short-term Memory Network

Abstract

Short-term passenger flow prediction is considered as an effective means of timely passenger flow evacuation. In this paper a novel inherit long short-term memory network (inherit-LSTM) model is proposed to predict short term passenger flow of transit. Traditional long short-term memory network models have the characteristics of long-term dependencies. This causes heavy weight on redundant information for a large number of historical data and leads current feature information to be ignored. In order to reduce excessive dependence on historical information, an inherit-LSTM model is proposed by using inheritance gate to retain more current input information. Comparing with the traditional model, this model takes full account of the effects for weather and large-scale emergencies. We compare our model against well-known existing machine/deep learning prediction models(Auto-Regressive with extra inputs and Long short-term memory network). The experimental results show that inherit-LSTM model is more accurate than the other two models in predicting short-term passenger flow.