Short-term Urban Rail Transit Passenger Flow Forecasting Based on Empirical Mode Decomposition and LSTM

This paper proposed a method to forecast the short-term passenger flow, which is a vital component of urban rail transit system. We used a hybrid EMD-LSTM prediction model which combines empirical mode decomposition (EMD) and long short-term memory (LSTM) to forecast the short-term passenger flow in urban rail transit system. EMD can extract the variation trend of passenger flow, then LSTM can make the prediction to prove the accuracy. The experimental results indicate that the EMD-LSTM model used in this paper has better prediction accuracy than the LSTM model alone. Besides, the amount of data used in this experiment is small, and there is no need to consider additional features except temporal factor. According to what we have learned, this is the first time to combine EMD and LSTM to make short-term prediction in the urban rail transit system. Introduction Short-term passenger flow forecasting is a vital component of urban rail transit system. The forecasting results is an important basis for urban rail transit feasibility study and design, and also the main basis of project construction. In the recent studies, linear forecasting method and non-linear forecasting method are used. Grey System Theory and ARIMA are the represent of linear forecasting methods. LSTM [1], deep learning [2] and spatio-temporal deep learning [3] are the represent of nonlinear forecasting methods. Urban rail transit passenger flow has the characteristics of non-linear, periodicity and random, and it is inapplicability for short-term passenger flow forecasting. Moreover, some factors, like emergency, which affect passenger flow, are hard to acquire or forecast. So as to solve this problem, hybrid EMD-LSTM prediction model is used. Firstly, the passenger flow data of Beijing subway Line 10 is used, considering only the time characteristics of the data, then the hybrid EMD-LSTM prediction model is used. The EMD is used to decompose the original passenger flow data, and statistical method is used to select each component, then LSTM is used to predict each component separately. Finally, the prediction results of each component are added to the final result. Methodology Empirical Mode Decomposition Empirical mode decomposition (EMD) [4] is a signal decomposition algorithm, which is suitable for non-liner and non-stationary signal. The original time series signal can be decomposed into a small number of oscillatory modes which can be expressed as some intrinsic modals functions (IMF) and a residue. The residue retains a non-periodic trend of the original signal, and any periodic fluctuation in original signal will be decomposed into IMFs. IMFs must satisfy the following two conditions [4]: 1. In the whole data set, the number of extrema and the number of zero crossings must either equal or differ at most by one. 2. At any point, the mean value of the envelope defined by the local maxima and the envelope defined by the local minima is zero. International Conference on Modeling, Analysis, Simulation Technologies and Applications (MASTA 2019) Copyright © 2019, the Authors. Published by Atlantis Press. This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/). Advances in Intelligent Systems Research, volume 168