Abstract:This paper presents a novel framework for predicting metro passenger flow that is both interpretable and computationally efficient. The proposed method first uses a correlation-based spatiotemporal feature selection strategy (Cor-STFS) to identify the optimal input scheme for the prediction model, effectively reducing unnecessary interference. The framework then introduces a new multivariate passenger flow prediction architecture called STA-PTCN-BiGRU, which combines a spatiotemporal attention (STA) mechanism, parallel temporal convolutional networks (PTCN), and bidirectional gated recurrent units (BiGRU) to capture the dynamic internal patterns of passenger flow. By utilising parallel computing, this architecture significantly reduces resource consumption. The effectiveness of the proposed approach is evaluated using four datasets from the Shanghai Metro. Experimental results show that the new method outperforms baseline approaches in terms of root mean square error (RMSE), mean absolute error (MAE), and symmetric mean absolute percentage error (SMAPE), achieving average reductions of 9.98%, 8.08%, and 13.29% in these metrics, respectively.

What problem does this paper attempt to address?

This paper attempts to solve several key problems in passenger flow prediction in subway systems, specifically including: 1. **Selection and interpretability of input data**: Existing neural - network - based models face challenges in selecting appropriate and interpretable input data. Traditional methods usually directly use raw data as input, which will lead to the inclusion of irrelevant information and affect the prediction accuracy. In addition, in large - scale subway networks, including all available input data will result in a significant computational burden. Due to the black - box nature of neural networks, users cannot perform meaningful and interpretable analysis on the model's output based on the initial input data. 2. **Limitations of a single model**: It is difficult to achieve higher accuracy by relying solely on a single model. It is necessary to develop an effective combination framework that can effectively handle the spatio - temporal characteristics of passenger flow. However, existing research on combination models often faces problems of long computing time and excessive memory consumption. 3. **Lack of integration of domain - specific knowledge**: Current methods often fail to combine the specific domain knowledge of the predicted object when considering global external features. For example, the subway system has unique operational characteristics, such as trains strictly adhering to the timetable, and stations with similar characteristics may show similar passenger flow distributions in time and space. These characteristics have been ignored in previous studies. To solve the above problems, this paper proposes a new deep - learning - based framework for passenger flow prediction in subway systems. The main contributions include: - **Feature selection method based on maximum correlation (Cor - STFS)**: By retaining the most relevant spatio - temporal information, reducing the interference of irrelevant information, and improving the performance of the model. - **Network framework of parallel computing architecture (STA - PTCN - BiGRU)**: This framework combines the spatio - temporal attention mechanism (STA), parallel time - convolution network (PTCN), and bidirectional gated recurrent unit (BiGRU), effectively reducing the computational burden and achieving good results in terms of prediction accuracy. - **Train timetable features as global external features**: Utilizing the knowledge of the subway domain, extracting train timetable information as external features, further improving the prediction accuracy. - **Generalization performance of the model**: This model shows excellent generalization performance on different types of subway stations, and its feasibility and effectiveness are verified by the actual data of Shanghai Metro. Through these innovations, this paper aims to provide an efficient and interpretable method for subway passenger flow prediction.

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

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

Short-Term Passenger Flow Forecast of Rail Transit Station Based on MIC Feature Selection and ST-LightGBM considering Transfer Passenger Flow

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

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

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

Deep Learning for Metro Short-Term Origin-Destination Passenger Flow Forecasting Considering Section Capacity Utilization Ratio

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

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

Learning Spatial-Temporal Dynamics for Short-Term Passenger Flow Prediction in Urban Rail Transit

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

Transformer Based Spatial-Temporal Fusion Network for Metro Passenger Flow Forecasting.

Dynamic Relational Graph Convolutional Network for Metro Passenger Flow Forecasting

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

ADST: Forecasting Metro Flow Using Attention-Based Deep Spatial-Temporal Networks with Multi-Task Learning

Multi-STGCnet: A Graph Convolution Based Spatial-Temporal Framework for Subway Passenger Flow Forecasting

A Correlation Information-based Spatiotemporal Network for Traffic Flow Forecasting

Forecasting Short-Term Passenger Flow: An Empirical Study on Shenzhen Metro

Short-Term Passenger Flow Prediction in Urban Rail Transit Based on Points of Interest

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