Abstract:Accurate forecast of citywide crowd flows on flexible region partition benefits urban planning, traffic management, and public safety. Previous research either fails to capture the complex spatiotemporal dependencies of crowd flows or is restricted on grid region partition that loses semantic context. In this paper, we propose DeepFlowFlex, a graph-based model to jointly predict inflows and outflows for each region of arbitrary shape and size in a city. Analysis on cellular datasets covering 2.4 million users in China reveals dependencies and distinctive patterns of crowd flows in not only the conventional space and time domains, but also the speed domain, due to the diverse transportation modes in the mobility data. DeepFlowFlex explicitly groups crowd flows with respect to speed and time, and combines graph convolutional long short-term memory networks and graph convolutional neural networks to extract complex spatiotemporal dependencies, especially long-term and long-distance inter-region dependencies. Evaluations on two big cellular datasets and public GPS trace datasets show that DeepFlowFlex outperforms the state-of-the-art deep learning and big-data-based methods on both grid and non-grid city map partition.

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve This paper aims to address the issue of crowd flow prediction within urban areas. Specifically: 1. **Complex Spatiotemporal Dependencies**: Existing studies have failed to adequately capture the complex spatiotemporal dependencies of crowd flow, especially those over long distances and long time scales. 2. **Flexible Region Division**: Most current crowd flow prediction methods are limited to grid-based region divisions, resulting in the loss of semantic context information. For example, urban planning and traffic management are often based on irregularly shaped geographic or administrative regions, which cannot be effectively handled by existing methods. To address these issues, the authors propose the **DeepFlowFlex** model, a graph-based model capable of jointly predicting the inflow and outflow of regions of any shape and size within a city. This model utilizes Graph Convolutional LSTM (Long Short-Term Memory) and Graph Convolutional Neural Networks to extract complex spatiotemporal dependencies, particularly those over long distances and long time scales. Additionally, the model considers the velocity domain to distinguish flow patterns under different traffic modes. ### Main Contributions 1. **First Work**: To the best of the authors' knowledge, this is the first work to utilize deep graph learning to predict city-wide crowd flow under flexible region divisions. 2. **Velocity Domain Decomposition**: In addition to the traditional time and space domains, the velocity domain is introduced to distinguish different traffic modes. 3. **Performance Improvement**: Evaluations on large-scale cellular datasets show that DeepFlowFlex outperforms existing state-of-the-art models in both grid and non-grid region divisions.

Improving Urban Crowd Flow Prediction on Flexible Region Partition

Predicting Citywide Crowd Flows Using Deep Spatio-Temporal Residual Networks

Passenger Demand Prediction with Cellular Footprints

Crowd Flow Prediction by Deep Spatio-Temporal Transfer Learning.

Modeling Real-Time Human Mobility Based on Mobile Phone and Transportation Data Fusion

An urban crowd flow model integrating geographic characteristics

Deep Spatio-Temporal Residual Networks for Citywide Crowd Flows Prediction

Improving Intra-Urban Prediction of Regional Pedestrian Flow Using a Hybrid Deep Learning Approach

Urban Crowd Flow Forecasting Based on Cellular Network

A Foundation Model for Unified Urban Spatio-Temporal Flow Prediction

Online Spatio-Temporal Crowd Flow Distribution Prediction for Complex Metro System

Attentive Crowd Flow Machines

3DGCN: 3-Dimensional Dynamic Graph Convolutional Network for Citywide Crowd Flow Prediction

DeepSTN+: Context-Aware Spatial-Temporal Neural Network for Crowd Flow Prediction in Metropolis.

Crowd Flow Prediction for Irregular Regions with Semantic Graph Attention Network

GridFormer: Spatial-Temporal Transformer Network for Citywide Crowd Flow Prediction

Fuzzy Hybrid Framework with Dynamic Weights for Short‐term Traffic Flow Prediction by Mining Spatio‐temporal Correlations

Context-aware Spatial-Temporal Neural Network for Citywide Crowd Flow Prediction Via Modeling Long-range Spatial Dependency

Deep spatio-temporal neural network based on interactive attention for traffic flow prediction

Predicting Citywide Crowd Flows in Irregular Regions Using Multi-View Graph Convolutional Networks

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