01697769

Component

Short-term prediction of passenger flow in rail transit is an important foundation for metro transit management and crowd regulation. Due to the impact of different mass passenger flow events, traditional prediction models are unable to reach the accuracy requirements of metro operation and management. The paper proposes a new hybrid model which combines empirical mode decomposition and recurrent neural networks in order to predict the short-term demand and analyze the temporal and spatial characteristics of mass passenger flow. This approach consists of three stages: (i) Raw passengers flow data is classified based on the types of mass passenger flow events (e.g. extreme weather conditions, mass concerts). (ii) Empirical Mode Decomposition (EMD) is used to decompose each stations’ passenger flow series into several intrinsic mode function (IMF) components. (iii) An improved recurrent neural network (RNN) based on spatial correlation is established to predict passenger flow in a short period. Meanwhile, considering the relevance of passenger flow at different stations, a component called Long Short-Term Memory (LSTM) for measuring the influence of long-term and short-term passenger characteristics was built and added to the standard RNNs structure. This hybrid analysis approach is verified by classified passenger flow data collected by metro Automatic Fare Collection System in Shanghai, China. Experimental results indicate that the proposed hybrid approach performs well in terms of prediction accuracy and is suitable for predicting in 4 different types of scenarios.

This paper was sponsored by TRB committee AP065 Standing Committee on Rail Transit Systems.

19-04934

English

9p

2019

Transportation Research Board 98th Annual Meeting

Digital/other

Figures; References; Tables

Neural networks; Passenger counting; Rail transit; Ridership; Transit riders

Empirical Mode Decomposition

Shanghai (China)

Operations and Traffic Management; Passenger Transportation; Railroads

Transportation Research Board Annual Meeting 2019 Paper #19-04934

TRIS, TRB, ATRI

Dec 7 2018 9:36AM