TL;DR
A novel traffic forecasting framework, FusionGraphSAGE with Neural Networks (FGSNN), which combines a predefined static adjacency matrix based on node distances with a dynamic adaptive graph to better capture evolving traffic relationships and achieves collaborative spatiotemporal feature learning and refined traffic flow prediction.
Traffic congestion is a growing challenge in urban environments, driven by increasing population and vehicle density, leading to significant economic and societal impacts. Accurate traffic forecasting is a key component of Cooperative Intelligent Transportation Systems (C-ITS), enabling proactive traffic management strategies such as adaptive signal control and dynamic routing. However, existing approaches often struggle to capture spatial dependencies in road networks and long-range temporal dynamics in traffic data. This paper proposes TETRA, a hybrid spatio-temporal traffic forecasting model that integrates Graph Convolutional Networks (GCNs) with Extended Long Short-Term Memory (xLSTM). By incorporating matrix-based memory and memory mixing, xLSTM enables the model to capture complex multi-timescale temporal patterns, including congestion propagation and delayed recovery dynamics, which are not well represented by conventional recurrent models. The proposed approach is evaluated on a real-world urban traffic dataset and the widely used METR-LA benchmark. Experimental results show that TETRA outperforms or matches established baseline models representative of the main spatio-temporal paradigms, with the most pronounced gains at medium- and long-term horizons, achieving up to 13.0% lower MAE, 20.0% lower RMSE, and 6.0% higher $$R^2$$ on a real-world dataset relative to the strongest investigated baseline at each horizon. Statistical analysis confirms that these improvements are robust across prediction horizons. Additional evaluations, including ablation and sensitivity analyses, demonstrate the effectiveness and scalability of the proposed model.
: With the progress of the system of intelligent transportation, traffic flow forecasting is essential to achieve efficient traffic management and control. It can not only optimize real-time traffic flow and improve travel efficiency, but also provide a basis for long-term road network planning and reduce carbon emissions. However, present models are unable to acquire the complicated spatio-temporal dependency of data, leading to low prediction accuracy. Therefore, this research uses a prediction model that incorporates multiple attention mechanisms with spatiotemporal graph convolutional networks (HASTGCN). This model combines channel attention, spatial attention, and temporal attention mechanisms, and designs a spatiotemporal map convolution module to collaboratively model the dynamic spatiotemporal connection of traffic flow collaboratively model. The trial results display the performance of HASTGCN on the PEMS04 dataset (Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) indicators) performed better than other models which proving its effectiveness in capturing spatiotemporal characteristics and improving prediction accuracy.
The ability to forecast traffic conditions in urban environments is essential for intelligent transport systems because it provides proactive congestion management, traffic control and informed urban mobility planning. However, due to the extreme spatial and temporal volatility of traffic flow patterns, conventional statistical, machine learning, and sequence-based deep learning approaches either fail to account for the spatial relationships between highway segments or fail to sufficiently model long-range temporal dynamics. To address these gaps, the present study uses a Spatial-Temporal Graph Neural Network (STGNN) framework that can learn a combination of spatial and temporal relationships in road networks and changing time-varying patterns in traffic flow. The model has a graph-based architecture where graph-convolutional layers are combined with gated recurrent units and transformer-based attention units, thus creating a hybrid architecture that is capable of multi-scale spatio-temporal dependencies. The METR-LA dataset was used in experiments, and it was observed that STGNN had smaller Mean Absolute Percentage Error(MAPE), Mean Absolute Error(MAE), and Root Mean Squared Error(RMSE) at 15, 30, and 60 minute horizons of prediction than the baseline and the implemented models: Sequence to Sequence (Seq2Seq) and Temporal Graph Convolutional Network (T-GCN). The model achieved the best results for the 15-minute interval with an MAE of 2.71, an RMSE of 5.17, and a MAPE of 7.08. The results imply that adaptive spatial learning, together with the temporal sequence modeling, can produce much better forecasting stability, which underlines the potential STGNN-based traffic prediction systems have to contribute to real-time traffic control.
An improved Transformer prediction model that integrates a graph convolutional network (GCN) and a self-attention mechanism is proposed for traffic flow prediction, combining temporal self-attention and learnable temporal encoding to capture both long-term traffic evolution patterns and sudden fluctuations.
Traffic flow prediction is a critical foundational problem in intelligent transportation systems. Although Large Language Model (LLM) has shown promising potential in time series modeling tasks in recent years, existing LLM-based methods generally overlook the inherent multi-scale characteristics of traffic flow data, which significantly limits their ability to capture complex spatio-temporal evolution patterns. To address this issue, this paper proposes a traffic flow forecasting framework named Multi-Scale Graph Convolution Enhanced Large Language Model (MSG-LLM). Firstly, the traffic flow series are decomposed based on frequency-domain analysis to identify periodic components, enabling the adaptive partitioning of the original series into multiple time scales. Subsequently, adaptive graph structures are constructed at different time scales, and graph convolution operations are introduced to fully characterize the correlation dependencies of traffic nodes during multi-scale spatio-temporal evolution. On this basis, a bidirectional multi-scale fusion module is designed to obtain comprehensive and consistent multi-scale representations through information fusion from fine-to-coarse and coarse-to-fine scales. Finally, the fused multi-scale spatio-temporal features are integrated into a partially frozen pre-trained large language model. By fine-tuning only task-specific parameters, this approach preserves the LLM’s general time series modeling capabilities while effectively reducing training costs and mitigating overfitting risks. Extensive experimental results on the PEMS04 and PEMS08 datasets demonstrate that the proposed method significantly outperforms existing mainstream models in both short-term and long-term traffic flow forecasting tasks, validating the effectiveness and strong generalization ability of MSG-LLM in modeling complex traffic systems.
This framework introduces an adaptive graph learning module that dynamically infers meaningful connectivity relationships among traffic sensors—not relying on fixed geographic or distance-based assumptions—but instead leveraging real-time traffic correlations and node-level embeddings, enabling effective modeling of both localized spatial interactions and multi-scale temporal dependencies across varying prediction horizons.