Jul 2026· Journal of King Saud University: Computer and Information Sciences· Vol 38· 0 citations· 30 references
TL;DR
A Tensor-Evolving Graph with Temporal Separation Network (TEG-TSNet) for traffic flow forecasting is proposed, which constructs a unified spatial prior via graph Laplacian spectral embedding and introduces a time-conditioned structure generation paradigm.
Abstract
Traffic flow forecasting requires joint modeling of temporal nonstationarity and diverse evolution patterns of traffic states under non-Euclidean road network constraints. Although recent spatiotemporal graph forecasting methods incorporate dynamic graph learning to alleviate the limitations of static topology, their spatial structure evolution is still largely driven by feature similarity or latent variables. An explicit temporal-prior-modulated mechanism for structure generation remains absent. Temporal modeling and spatial structure learning are often conducted independently at different levels, limiting the ability to capture periodically driven evolution of spatiotemporal dependencies. To address these limitations, a Tensor-Evolving Graph with Temporal Separation Network (TEG-TSNet) for traffic flow forecasting is proposed, which constructs a unified spatial prior via graph Laplacian spectral embedding. A node-level gating mechanism is then applied to enable differentiable temporal decoupling between trend and seasonal components. Next, a tensor-evolving graph encoder embeds time, source nodes, and target nodes into a unified multilinear tensor representation. This design dynamically generates structure-aware adjacency relations for different time slices and models time-varying spatial dependencies via diffusion graph convolution. During decoding, spatiotemporal attention fusion and a sparse expert routing mechanism are employed to strengthen representations of multimodal traffic states. Unlike prior studies that loosely combine temporal modeling and dynamic graph learning, the proposed framework introduces a time-conditioned structure generation paradigm, where temporal priors explicitly govern the evolution of spatial dependencies through a unified tensor formulation. This establishes a tightly coupled spatiotemporal modeling mechanism rather than a simple component-level integration. Extensive experiments on multiple real-world traffic datasets demonstrate stable and competitive forecasting performance and validate effectiveness and generalization. Related codes are available at https://github.com/Haku-zx/TEG-TSNet.
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.
Zhengxu Luan, Huan Wang, Miaobowen Wang et al.· Computers and artificial int...· 0 citations
Traffic flow prediction remains challenging due to the complex interaction between heterogeneous temporal frequencies and irregular spatial structures. Most existing graph neural network (GNN)-based methods fail to explicitly model spatiotemporal dependencies across multiple frequency components in traffic flow. To address this limitation, a novel framework, termed Spatiotemporal Trend-Event Decomposition Graph Convolutional Network (STEDGCN), is proposed. The framework introduces a temporal signal separator that decomposes raw traffic flow sequences into low-frequency trends and high-frequency events, thereby preserving frequency-specific temporal patterns. A dual-frequency spatiotemporal encoder is designed to model the temporal and spatial characteristics of the two components. It integrates multi-head attention and causal convolution to model temporal dynamics. It also employs trend-driven and event-driven graphs to capture inter-node dependencies and spatiotemporal correlations. A fusion-gated spatiotemporal decoder is introduced to reduce channel redundancy using a gating mechanism. It enables information interaction between the trend and event branches through a query-driven attention strategy. This improves the coherence of the final prediction. Experiments on four benchmark traffic flow datasets show that the proposed model consistently outperforms state-of-the-art methods across multiple metrics. These results confirm the effectiveness of frequency-aware decoupling and dual-path fusion in complex traffic flow modeling.
Linlong Chen· Journal of King Saud Univers...· 0 citations
A spatiotemporal graph Transformer framework that jointly models spatial interactions and temporal dependencies for traffic forecasting in edge computing and leverages Transformer-based self-attention to learn long-range temporal patterns from historical traffic observations is proposed.
Accurate traffic flow forecasting hinges on modeling coupled spatio-temporal dependencies rather than treating space and time in isolation. Many prior methods process spatial and temporal features separately — either in series or in parallel — and then fuse them with simple operators, which weakens their ability to capture intrinsic space–time interactions. We propose multi-graph transformer for traffic flow forecasting (MGTTP), a framework with an innovatively designed bidirectional spatio-temporal interaction mechanism: temporal signals guide multi-graph spatial fusion, while spatial context guides attention-based temporal aggregation. It addresses the limitations of static spatial fusion in existing multi-graph models and the serial spatio-temporal modeling paradigm in vanilla transformer baselines, achieving deep coupled modeling of spatio-temporal features. First, MGTTP builds three complementary graphs — adjacency, reachability, and similarity — and applies temporal feature-guided attention to dynamically fuse their multi-dimensional spatial representations. Subsequently, a transformer encoder captures long-term temporal dependencies, with spatial feature-guided attention to aggregate the time series. Finally, a gated fusion module realizes the ultimate fusion of spatio-temporal features for prediction. Extensive experiments on four public real-world traffic datasets demonstrate that MGTTP outperforms all compared mainstream baseline models across all evaluation metrics, with statistically significant performance gaps, validating the effectiveness of the proposed bidirectional spatio-temporal interaction mechanism.
Experimental results on the public PEMS04 and PEMS08 datasets demonstrate that the proposed ESDG-ALSTM model significantly improves forecasting accuracy, confirming that ESDG-ALSTM is more sensitive to abrupt events and multimodal evolution patterns and can effectively enhance prediction performance in complex traffic flow scenarios.
Guozheng Li, Baijing Wu, Ke Gao et al.· Frontiers of Computer Scienc...· 0 citations
The proposed position-aware spatio-temporal modeling strategy provides a practical reference for information fusion and dynamic state estimation in large-scale wireless sensing networks and electromagnetic signal-driven monitoring systems, supporting future intelligent perception and communication infrastructures.
J. Sun, Y. Liu, Y. Dou et al.· Advanced Electromagnetics· 0 citations