Abstract:For traffic prediction in transportation services such as car-sharing and ride-hailing, mid-term road traffic prediction (within a few hours) is considered essential. However, the existing road-level traffic prediction has mainly studied how significantly micro traffic events propagate to the adjacent roads in terms of short-term prediction. On the other hand, recent attempts have been made to incorporate regional knowledge such as POIs, road characteristics, and real-time social events to help traffic prediction. However, these studies lack in understandings of different modalities of road-level and region-level spatio-temporal correlations and how to combine such knowledge. This paper proposes a novel method that embeds real-time region-level knowledge using POIs, satellite images, and real-time LTE access traces via a regional spatio-temporal module that consists of dynamic convolution and temporal attention, and conducts bipartite spatial transform attention to convert into road-level knowledge. Then the model ingests this embedded knowledge into a road-level attention-based prediction model. Experimental results on real-world road traffic prediction show that our model outperforms the baselines.

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve The paper aims to address the challenges in medium to long-term road traffic prediction (predictions within a few hours). Specifically, the paper proposes a new method to improve the accuracy of road traffic prediction by embedding real-time regional knowledge (such as Points of Interest (POIs), satellite images, and real-time LTE access traces). The main problems addressed include: 1. **Combining Regional and Road-Level Data**: - Existing studies often do not separately handle regional and road-level data, which may lead to performance degradation. For example, using Convolutional Neural Networks (CNN) to analyze regional data while using Graph Convolutional Networks (GCN) to process road data may yield better results. 2. **Dynamic Regional Knowledge Learning**: - While models can utilize real-time regional traffic data to find certain patterns, without information about the regional context, it is difficult for the model to interpret why these patterns occur. Therefore, a method is needed to analyze regional data from the built environment. 3. **Conversion of Regional Knowledge to Road-Level**: - When converting regional information to road information, determining how much regional information each road should use is a challenge. The traditional 500-meter range may not be suitable for all roads and needs to be adjusted based on road characteristics. ### Solution The paper proposes a new method to address the above problems through the following steps: 1. **Regional Spatiotemporal Module**: - Embedding real-time regional knowledge using dynamic convolution and temporal attention mechanisms. - Converting regional knowledge to road-level knowledge through a dual-part spatial transformation attention mechanism. 2. **Road-Level Attention Prediction Model**: - Inputting the embedded regional knowledge into an attention-based road-level prediction model. ### Experimental Results Experimental results show that the model outperforms baseline models in actual road traffic prediction tasks. Particularly in medium to long-term predictions (2 hours, 3 hours), attention-based models (such as GMAN and the proposed model) perform better. Additionally, the introduction of real-time regional knowledge significantly improves the model's prediction performance. ### Main Contributions 1. **First Use of Real-Time Regional Population Data for Road-Level Traffic Prediction**: - Proposing a new method to learn real-time regional knowledge through multimodal training. 2. **Dynamic Convolution and Dual-Part Masked Transformation Attention Mechanism**: - Dynamic convolution based on regional correlation and distance, and a dual-part masked transformation attention mechanism adding Gaussian masks for each road. 3. **Building a Real-World Dataset**: - Constructing a real-world dataset in Seoul, South Korea, and publicly sharing it to promote the development of the research community. ### Conclusion The paper significantly improves the accuracy of medium to long-term road traffic prediction by introducing real-time regional knowledge and improved attention mechanisms. This method performs excellently in complex urban environments, especially in high-density and large-population cities.

Spatio-Temporal Road Traffic Prediction using Real-time Regional Knowledge

Spatial–Temporal Deep Tensor Neural Networks for Large-Scale Urban Network Speed Prediction

Traffic Flow Prediction Based on Probe Vehicle GPS Traces Considering Temporal and Spatial Correlations

Network-scale traffic prediction via knowledge transfer and regional MFD analysis

Real-time road traffic prediction with spatio-temporal correlations

Understanding Dynamic Spatio-Temporal Contexts in Long Short-Term Memory for Road Traffic Speed Prediction

Enhancing Spatiotemporal Traffic Prediction through Urban Human Activity Analysis

Deep Spatio-Temporal Residual Networks for Connected Urban Vehicular Traffic Prediction.

Short-Term Traffic Flow Prediction Method Based on Spatio-Temporal Characteristics of Complex Road Network

Regional Correlation Aided Mobile Traffic Prediction with Spatiotemporal Deep Learning

A Deep Learning Framework Based on Spatio-Temporal Attention Mechanism for Traffic Prediction.

Short-Term Travel Time Prediction: A Spatiotemporal Deep Learning Approach.

Cellular Traffic Prediction: A Deep Learning Method Considering Dynamic Nonlocal Spatial Correlation, Self-Attention, and Correlation of Spatiotemporal Feature Fusion

Traffic Flow Prediction via Spatial Temporal Graph Neural Network

Urban Traffic Accident Risk Prediction Revisited: Regionality, Proximity, Similarity and Sparsity

Channel attention-based spatial-temporal graph neural networks for traffic prediction

A Hybrid Deep Learning Model for Short-Term Traffic Flow Pre-Diction Considering Spatiotemporal Features

Short-Term Urban Traffic Flow Prediction Using Deep Spatio-Temporal Residual Networks

Data-Driven Short-Term Forecasting for Urban Road Network Traffic Based on Data Processing and LSTM-RNN

Short-Term Traffic State Prediction Based On The Spatiotemporal Features Of Critical Road Sections

Traffic Congestion Prediction by Spatiotemporal Propagation Patterns