Abstract:The operation and management of the metro system in urban areas rely on accurate predictions of future passenger flow. While using all the available information can potentially improve on the accuracy of the flow prediction, there has been little attention to the hierarchical relationship between the type of tickets collected from the passengers entering/exiting a station and its resulting passenger flow. To this end, we propose a novel Integrative Prediction Framework with the Hierarchical Message-Passing Graph Neural Network (IPF-HMGNN). The proposed framework consists of three components: initial prediction, task judgment and hierarchical coordination modules. Using the Wuxi, China metro network as an example, we study two prediction approaches (i) traditional prediction approach where the model directly predicts passenger flow at the station, and (ii) hierarchical prediction approach where the prediction of ticket type and station passenger flow are performed simultaneously considering the hierarchical constraints (i.e., the sum of predicted passenger flow per ticket type equals the predicted station aggregated passenger flow). Experimental results indicate that in the traditional prediction approach, our IPF-HMGNN can significantly reduce the mean absolute error (MAE) and root mean square error (RMSE) of the GNN prediction model by 49.56% and 53.88%, respectively. In the hierarchical prediction approach, IPF-HMGNN can achieve a maximum reduction of 35.32% in MAE and 36.18% in RMSE, while satisfying the hierarchical constraint.

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve This paper aims to address the challenge of accurately predicting passenger flow in urban metro system operations and management. Although utilizing all available information can improve prediction accuracy, existing studies rarely focus on the hierarchical relationship between different ticket types used by passengers when entering and exiting stations and the final passenger flow. To fill this gap, the authors propose a new integrated prediction framework—Integrated Prediction Framework with Hierarchical Message Passing Graph Neural Network (IPF-HMGNN). This framework consists of three components: initial prediction, task judgment, and hierarchical coordination modules. Specifically, this paper proposes two prediction methods: 1. **Traditional Prediction Method**: The model directly predicts passenger flow at stations. 2. **Hierarchical Prediction Method**: It simultaneously predicts ticket types and station passenger flow, considering hierarchical constraints (i.e., the sum of passenger flows predicted for each ticket type equals the total predicted passenger flow at the station). Experimental results show that in the traditional prediction method, IPF-HMGNN can significantly reduce the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) of Graph Neural Network (GNN) prediction models by 49.56% and 53.88%, respectively. In the hierarchical prediction method, IPF-HMGNN can meet hierarchical constraints while reducing MAE and RMSE by up to 35.32% and 36.18%, respectively. ### Main Contributions 1. **New Input Data Segmentation Method**: Extracts local and global features by aggregating data blocks at 15-minute intervals. 2. **Establishing a General Hierarchical Message Passing Mechanism**: Considers the potential hierarchical relationship of predicted passenger flow in GNN models. 3. **Constructing a Deep Learning-Based Hierarchical Prediction Coordination Module**: Seamlessly integrates into existing deep learning prediction models through deep learning fusion or concatenation mechanisms. 4. **Proposing a New Integrated Prediction Framework**: Suitable for traditional and hierarchical prediction tasks with or without hierarchical constraints. 5. **Validating the Framework's Effectiveness with Real Data**: Conducted an empirical study on the Wuxi Metro network in China. ### Research Background Metro systems have become an important choice for urban travel worldwide due to their economic, punctual, and efficient characteristics. With the acceleration of urbanization, the rapid growth in passenger demand poses significant challenges to the operation and management of urban metro systems. Accurate real-time short-term passenger flow (STPF) prediction is crucial for operational management decisions in metro systems. Deep Learning (DL) models have been widely applied to metro STPF prediction due to their powerful nonlinear fitting capabilities. However, existing GNN prediction models have the following shortcomings: 1. They rely solely on historical passenger flow input features, without considering local or global trends. 2. They lack research on predicting passenger flow for different ticket types. 3. They do not consider hierarchical constraints, i.e., the sum of passenger flows predicted for each ticket type should equal the total predicted passenger flow at the station. This paper aims to address the above issues and improve the accuracy of metro passenger flow prediction by introducing new data processing methods and hierarchical message passing mechanisms.

IPF-HMGNN: A novel integrative prediction framework for metro passenger flow

Hybrid Approach Combining Modified Gravity Model and Deep Learning for Short-Term Forecasting of Metro Transit Passenger Flows

A Hybrid GLM Model for Predicting Citywide Spatio-Temporal Metro Passenger Flow

Learning spatial-temporal dynamics and interactivity for short-term passenger flow prediction in urban rail transit

A method for short-term passenger flow prediction in urban rail transit based on deep learning

A dynamic multigraph and multidimensional attention neural network model for metro passenger flow prediction

Metro Passenger Flow Prediction Via Dynamic Hypergraph Convolution Networks

MGSTCN: A Multi-Graph Spatio-Temporal Convolutional Network for Metro Passenger Flow Prediction

Optimizing Metro Passenger Flow Prediction: Integrating Machine Learning and Time-Series Analysis with Multimodal Data Fusion

A dynamic graph deep learning model with multivariate empirical mode decomposition for network‐wide metro passenger flow prediction

METcross: A framework for short-term forecasting of cross-city metro passenger flow

Parallel Multi-Graph Convolution Network For Metro Passenger Volume Prediction

Metro Flow Prediction with Hierarchical Hypergraph Attention Networks

Fine‐Grained Passenger Load Prediction inside Metro Network via Smart Card Data

Forecast of Short-Term Passenger Flow in Multi-Level Rail Transit Network Based on a Multi-Task Learning Model

Learning spatial–temporal pairwise and high-order relationships for short-term passenger flow prediction in urban rail transit

Correlation-based feature selection and parallel spatiotemporal networks for efficient passenger flow forecasting in metro systems

Multi-Spatio-Temporal Convolutional Neural Network for Short-Term Metro Passenger Flow Prediction

Multi-sequence spatio-temporal feature fusion network for peak-hour passenger flow prediction in urban rail transit

STP-TrellisNets: Spatial-Temporal Parallel TrellisNets for Metro Station Passenger Flow Prediction

Leveraging Intra-Period and Inter-Period Features for Enhanced Passenger Flow Prediction of Subway Stations