Predicting short-term passenger flow accurately is of great significance for daily management and for a timely emergency response of rail transit networks. In this paper, we propose an attention-based Graph–Temporal Fused Neural Network (GTFNN) that can make online predictions of origin–destination (OD) flows in a large-scale urban transit network. In order to solve the key issue of the passenger hysteresis in online flow forecasting, the proposed GTFNN takes finished OD flow and a series of features, which are known or observable, as the input and performs multi-step prediction. The model is constructed from capturing both spatial and temporal characteristics. For learning spatial characteristics, a multi-layer graph neural network is proposed based on hidden relationships in the rail transit network. Then, we embedded the graph convolution into a Gated Recurrent Unit to learn spatial–temporal features. For learning temporal characteristics, a sequence-to-sequence structure embedded with the attention mechanism is proposed to enhance its ability to capture both local and global dependencies. Experiments based on real-world data collected from Chongqing’s rail transit system show that the metrics of GTFNN are better than other methods, e.g., the SMAPE (Symmetric Mean Absolute Percentage Error) score is about 14.16%, with a range from 5% to 20% higher compared to other methods.
1 January 2024 in Journal of Advanced Transportation
Accurately predicting origin‐destination (OD) passenger flows serves as the basis for implementing efficient plans, including line planning and timetabling. However, due to the complexity and variety of OD passenger flows types, general prediction models have difficulty in capturing the features of different OD passenger flows, which in turn leads to poor prediction performance. To address this issue, we propose an integrated framework that combines clustering and prediction methods. First, an unsupervised deep learning model is devised to automatically cluster OD flow types by capturing shape characteristics. Second, three types of features are created to enhance training efficiency, including static features, time‐dependent observed features, and time‐dependent known features. Based on the clustering of OD passenger flow, a weighted adaptive passenger flow prediction model is developed. The study employs a temporal fusion transformers model to enable multitype OD passenger flow prediction. In the numerical experiments, the model was applied to the urban rail transit in South China, and the model clustered 15,168 OD pairs into 4 types for prediction. The findings show that this approach enhanced the prediction accuracy by 2.0%–9.6% compared to the LSTM model and by 1.6%–4.3% compared to the Graph WaveNet. Moreover, the model can accurately assess the various features for diverse types of OD flows.
A pick-up zone links a hub of inter-city transportation (e.g., airplane and rail, etc.) with a connected roadway belonging to the city road network. Passengers depart through the pick-up zone and the interactions between passengers and vehicles lead to delays and queues of vehicles on the connected roadway. A better understanding of the unique traffic characteristics of the connected roadway can help to reveal the collecting or dispersing capacity of the hub. This paper treats passenger boarding in a pick-up zone as a service process in a queue system and uses a fundamental diagram model derived from M/M/1 threshold queueing theory with two service phases to investigate the supply–demand relationship of the system. A calibration method is proposed to determine the parameters of the model. Numerical experiments showed that the model can capture a unique capacity drop in the pick-up zone. Moreover, analyses on the impacts of designed parameters on the model features, such as capacity and degree of capacity drop, were performed, and the comparisons showed that the model used performs better than other classical methods with a 0.69 mean squared error and a 0.90 sum of squares due to error). The results will be important supports for hub capacity management.