Deep Learning Architecture for Short-Term Passenger Flow Forecasting in Urban Rail Transit

Abstract

Short-term passenger flow forecasting is an essential component in urban rail transit operation. Emerging deep-learning models provide good insight into improving prediction precision. Therefore, we propose a deep-learning architecture combined residual network (ResNet), graph convolutional network (GCN) and long short-term memory (LSTM) (called "ResLSTM") to forecast short-term passenger flow in urban rail transit on a network scale. First, improved methodologies of ResNet, GCN, and attention LSTM models are presented. Then, model architecture is proposed, wherein ResNet is used to capture deep abstract spatial correlations between subway stations, GCN is applied to extract network-topology information, and attention LSTM is used to extract temporal correlations. Model architecture includes four branches for inflow, outflow, graph-network topology, as well as weather conditions and air quality. To the best of our knowledge, this is the first time that air-quality indicators have been taken into account, and their influences on prediction precision have been quantified. Finally, ResLSTM is applied to Beijing subway. Three time granularities (10, 15, and 30 min) are chosen to conduct short-term passenger flow forecasting. Comparison of prediction performance of ResLSTM with those of many state-of-the-art models shows the advancement and robustness of ResLSTM. Moreover, comparison of prediction precisions obtained from time granularities of 10, 15, and 30 min indicates that prediction precision increases with increasing time granularity. The study can provide subway operators with insight into short-term passenger flow forecasting by leveraging deep-learning models.

A Deep Learning Approach for Multi-Depth Soil Water Content Prediction in Summer Maize Growth Period

“…The spatio-temporal feature extraction advantage of integrating CNN and LSTM to construct a high-precision prediction model is a research hotspot [27]. For field environmental data, the combination of SWC and meteorological data over multiple consecutive days can be viewed as a two-dimensional, single-channel grayscale image, so the fusion model can use both CNN to extract highdimensional features from the image [28] and LSTM to extract time-series features [29]. In other areas of research targeting regression prediction problems, Qin et al [30] and Li et al [31] constructed prediction models for PM2.5 by integrating CNN and LSTM structures, respectively, and obtained better performance compared to traditional machine learning models and independent CNN and LSTM models.…”

Multi‐graph convolutional network for short‐term passenger flow forecasting in urban rail transit

“…We used the average values in the last time step of three patterns to predict the value in the next time step [1]. ARIMA: We used the Expert Modeller in the Statistical‐Package‐for‐the‐Social‐Sciences (SPSS®IBM Corp., USA) to obtain the ARIMA results [44]. The Expert Modeller in SPSS can automatically give the best predicted results. LSTM: LSTM was first applied to traffic field in 2015 [8].…”

“…This model has the same parameter with the proposed Conv‐GCN while without the 3D CNN layer. ResLSTM: A deep‐learning architecture comprised GCN, ResNet, and attention LSTM. The parameters are the same with [44]. …”

Multi-Task Fusion Deep Learning Model for Short-Term Intersection Operation Performance Forecasting

“…It should be noted that the calculation of stacking multiple GCN layers is more complex, and the gradient is easier to disappear [35]. Furthermore, with the deeper GCNs arising, the over-smoothing will make the features of the same vertex indistinguishable and debase the forecast accuracy [36].…”