6G conditioned spatiotemporal graph neural networks for real time traffic flow prediction

Area of Science:

• Intelligent Transportation Systems (ITS)
• Networked Systems
• Machine Learning

Background:

• Accurate, low-latency traffic forecasting is crucial for next-generation Intelligent Transportation Systems (ITS).
• Emerging 6G network context (slice-bandwidth, channel-quality) offers potential for enhancing spatio-temporal traffic prediction.
• Existing models often struggle to balance prediction accuracy with strict real-time constraints for freeway traffic.

Purpose of the Study:

• To investigate the fusion of 6G network context with spatio-temporal graph models for improved short-term freeway speed prediction.
• To evaluate the real-time performance and accuracy trade-offs of different graph neural network architectures, including a novel 6G-conditioned model.
• To identify limitations and potential areas for future advancements in network-aware traffic forecasting.

Main Methods:

• Developed a reproducible pipeline using the METR-LA benchmark, including data cleaning, temporal imputation, and 6G signal synthesis.
• Implemented and compared Spatio-Temporal Graph Convolutional Network (ST-GCN), Graph Attention (ST-GAT), Diffusion Convolutional Recurrent Neural Network (DCRNN), and a novel 6G-conditioned DCRNN (DCRNN6G).
• Conducted systematic evaluations across four feature regimes, hyperparameter sweeps, ablation studies, and latency profiling on commodity CPUs.

Main Results:

• Diffusion-recurrent modeling (DCRNN) achieved the optimal accuracy-latency trade-off, meeting real-time requirements with low RMSE and inference latency.
• Naïve integration of simulated 6G metrics yielded marginal accuracy gains for ST-GCN/ST-GAT and did not improve DCRNN, often increasing latency.
• Error diagnostics identified spatially localized 'hard' sensors and episodic time windows dominating prediction errors, suggesting targeted module improvements.

Conclusions:

• This work provides the first end-to-end benchmarking of 6G-conditioned spatio-temporal GNNs on METR-LA with real-time latency analysis.
• The proposed DCRNN6G model and network-aware fusion show potential but require further research for significant gains in ITS.
• Future directions include heterogeneous cross-graph fusion, dynamic adjacency learning, and probabilistic forecasting for co-optimized transportation and communication systems.