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

TL;DR

This work tackles mid-term road traffic prediction by integrating real-time regional knowledge with road-level signals. It introduces a regional spatio-temporal module that uses dynamic convolution and temporal attention, plus a bipartite spatial transform attention to translate region-level representations into road-level features, which are then processed by a road-focused attention-based predictor. The approach leverages LTE-based regional population, POIs, and satellite imagery, and demonstrates superior forecasting accuracy over strong baselines on a real-world Seoul dataset, with pronounced gains on POI-dense roads. The methodology advances multimodal traffic prediction by explicitly modeling modality separation and region-to-road information flow, offering practical benefits for urban traffic management and planning.

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.

Figures (5)

• Figure 1: Real-time road traffic speed and regional population (LTE access traces) visualization in Seoul, Korea.
• Figure 2: Proposed model architecture. Different arrow types represents different spatial- ($N_Z$: thin, $N_X$: thick), temporal-dimension ($P$: blue, $Q$: red), or spatio-temporal embedding ($\text{STE}_X$, $\text{STE}_Z$: dashed). For each transform attention block (Temporal TransAttn, Bipartite Spatial TransAttn), query and key are given from the left side (green and yellow bullets) and value is given from the bottom side (a blue bullet).
• Figure 3: Spatio-temporal embedding (STE) for (a) region-level ($\text{STE}_Z$) and (b) road-level ($\text{STE}_X$) for an attention module.
• Figure 4: Spatio-temporal attention block for (a) region-level (R-ST-Att Block) and (b) road(graph)-level (G-ST-Att Block) input data.
• Figure 5: Weekly analysis of performance on different roads of POI densities (average MAE, model improvement compared to GMAN+LTE).