Wenzheng Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Wenzheng Liu has a total of 29 co-authors. Here are the top co-authors by number of shared publications:
- Hongtao Li
Lanzhou Jiaotong University
8 shared publications Shaolong Sun
7 shared publications- Haina Zhang
School of Traffic and Transportation, Lanzhou Jiaotong University, Lanzhou, 730070, China
6 shared publications Hongtao Li
6 shared publicationsHaina Zhang
4 shared publicationsZhipeng Huang
4 shared publications
Latest Publications
- 12 November 2026
- Published by Elsevier in Physica A: Statistical Mechanics and its Applications
- Vol. 704
PG-RDM: Physics-Guided Residual Diffusion Model for Multimodal Vehicle Trajectory Prediction
- 12 November 2026
- Published by Elsevier in Physica A: Statistical Mechanics and its Applications
- Vol. 704
PG-RDM: Physics-Guided Residual Diffusion Model for Multimodal Vehicle Trajectory Prediction
- 14 August 2026
- Published by Elsevier in Information Sciences
- Vol. 758
Continuous spatial-temporal traffic forecasting based on neural partial differential equations
- 14 August 2026
- Published by Elsevier in Information Sciences
- Vol. 758
Continuous spatial-temporal traffic forecasting based on neural partial differential equations
- Open Access
- 6 August 2026
- Published by Elsevier in SSRN Electronic Journal
SAFE: Scene-Adaptive Stochastic Flow-Energy Matching for Vehicle Trajectory Prediction
Trajectory prediction constitutes a foundational module for long-term behavioral decision-making and proactive safety-oriented path planning in autonomous driving systems. Nevertheless, existing generative forecasting paradigms suffer from critical bottlenecks when modeling trajectory modal diversity, stochastic fluctuations, and adherence to physical constraint laws. To address these challenges, this paper proposes the Scene-Adaptive Stochastic Flow-Energy Matching (SAFE) framework, formulated as a unified Flow-Energy Stochastic Differential Equation paradigm. At the global level, we develop a Stochastic Energy Matching (SEM) to preserve the multi-modal features of vehicle behaviors while suppressing physically non-compliant regions. Furthermore, by constructing a rigorous fluctuation-dissipation relation, we theoretically unify Stochastic Flow Matching (SFM) and SEM into a cohesive stochastic differential equation formulation. Crucially, to adaptively regulate the trade-off between exploration and exploitation in the solution space, our framework dynamically infers a scene uncertainty metric u from historical vehicle trajectories and map topologies, which finely modulates a set of physically differentiable energy constraint functions. Consequently, the proposed SAFE enables unconstrained exploration to achieve comprehensive multi-modal intent coverage in low-risk scenarios while strictly compressing trajectory particles into compliant, safe topological channels within highly interactive and conflict-prone regions. Extensive experiments on the nuScenes and INTERACTION datasets validate that SAFE achieves significant improvements in alleviating non-compliant long-tail failures, reducing the Miss Rate and maintaining high trajectory diversity.
- 6 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessSAFE: Scene-Adaptive Stochastic Flow-Energy Matching for Vehicle Trajectory Prediction
Trajectory prediction constitutes a foundational module for long-term behavioral decision-making and proactive safety-oriented path planning in autonomous driving systems. Nevertheless, existing generative forecasting paradigms suffer from critical bottlenecks when modeling trajectory modal diversity, stochastic fluctuations, and adherence to physical constraint laws. To address these challenges, this paper proposes the Scene-Adaptive Stochastic Flow-Energy Matching (SAFE) framework, formulated as a unified Flow-Energy Stochastic Differential Equation paradigm. At the global level, we develop a Stochastic Energy Matching (SEM) to preserve the multi-modal features of vehicle behaviors while suppressing physically non-compliant regions. Furthermore, by constructing a rigorous fluctuation-dissipation relation, we theoretically unify Stochastic Flow Matching (SFM) and SEM into a cohesive stochastic differential equation formulation. Crucially, to adaptively regulate the trade-off between exploration and exploitation in the solution space, our framework dynamically infers a scene uncertainty metric u from historical vehicle trajectories and map topologies, which finely modulates a set of physically differentiable energy constraint functions. Consequently, the proposed SAFE enables unconstrained exploration to achieve comprehensive multi-modal intent coverage in low-risk scenarios while strictly compressing trajectory particles into compliant, safe topological channels within highly interactive and conflict-prone regions. Extensive experiments on the nuScenes and INTERACTION datasets validate that SAFE achieves significant improvements in alleviating non-compliant long-tail failures, reducing the Miss Rate and maintaining high trajectory diversity.
- 2 August 2026
- Published by Springer Nature
- p. 19-28
Spatio-Temporal Attention Convolution Network for Short-Term Subway Passenger Flow Forecasting
- 2 August 2026
- Published by Springer Nature
- p. 19-28
Spatio-Temporal Attention Convolution Network for Short-Term Subway Passenger Flow Forecasting
- 12 May 2026
- Published by Elsevier in Information Sciences
- Vol. 753
Multi-spatio-temporal information fusion neural network for traffic flow forecasting
- 12 May 2026
- Published by Elsevier in Information Sciences
- Vol. 753
Multi-spatio-temporal information fusion neural network for traffic flow forecasting
- 10 May 2026
- Published by Elsevier in Neural Networks
- Vol. 202, 109086
C2F-LGNet: A coarse-to-fine framework with local-global differential modeling for traffic prediction
- 10 May 2026
- Published by Elsevier in Neural Networks
- Vol. 202, 109086
C2F-LGNet: A coarse-to-fine framework with local-global differential modeling for traffic prediction
- 5 March 2026
- Published by Elsevier in Information Sciences
- Vol. 743
Continuous spatio-temporal hybrid graph convolutional network for air pollutant prediction
- 5 March 2026
- Published by Elsevier in Information Sciences
- Vol. 743
Continuous spatio-temporal hybrid graph convolutional network for air pollutant prediction
- Open Access
- 23 January 2026
- Published by MDPI AG in Electronics
- Vol. 15 (3) , 495
Multi-Scale Graph-Decoupling Spatial–Temporal Network for Traffic Flow Forecasting in Complex Urban Environments
Accurate traffic flow forecasting is a fundamental component of Intelligent Transportation Systems and proactive urban mobility management. However, the inherent complexity of urban traffic flow, characterized by non-stationary dynamics and multi-scale temporal dependencies, poses significant modeling challenges. Existing spatio-temporal models often struggle to reconcile the discrepancy between static physical road constraints and highly dynamic, state-dependent spatial correlations, while their reliance on fixed temporal receptive fields limits the capacity to disentangle overlapping periodicities and stochastic fluctuations. To bridge these gaps, this study proposes a novel Multi-scale Graph-Decoupling Spatial–temporal Network (MS-GSTN). MS-GSTN leverages a Hierarchical Moving Average decomposition module to recursively partition raw traffic flow signals into constituent patterns across diverse temporal resolutions, ranging from systemic daily trends to high-frequency transients. Subsequently, a Tri-graph Spatio-temporal Fusion module synergistically models scale-specific dependencies by integrating an adaptive temporal graph, a static spatial graph, and a data-driven dynamic spatial graph within a unified architecture. Extensive experiments on four large-scale real-world benchmark datasets demonstrate that MS-GSTN consistently achieves superior forecasting accuracy compared to representative state-of-the-art models. Quantitatively, the proposed framework yields an overall reduction in Mean Absolute Error of up to 6.2% and maintains enhanced stability across multiple forecasting horizons. Visualization analysis further confirms that MS-GSTN effectively identifies scale-dependent spatial couplings, revealing that long-term traffic flow trends propagate through global network connectivity while short-term variations are governed by localized interactions.
- 23 January 2026
- Published by MDPI AG in Electronics
- Vol. 15 (3) , 495
Open AccessMulti-Scale Graph-Decoupling Spatial–Temporal Network for Traffic Flow Forecasting in Complex Urban Environments
Accurate traffic flow forecasting is a fundamental component of Intelligent Transportation Systems and proactive urban mobility management. However, the inherent complexity of urban traffic flow, characterized by non-stationary dynamics and multi-scale temporal dependencies, poses significant modeling challenges. Existing spatio-temporal models often struggle to reconcile the discrepancy between static physical road constraints and highly dynamic, state-dependent spatial correlations, while their reliance on fixed temporal receptive fields limits the capacity to disentangle overlapping periodicities and stochastic fluctuations. To bridge these gaps, this study proposes a novel Multi-scale Graph-Decoupling Spatial–temporal Network (MS-GSTN). MS-GSTN leverages a Hierarchical Moving Average decomposition module to recursively partition raw traffic flow signals into constituent patterns across diverse temporal resolutions, ranging from systemic daily trends to high-frequency transients. Subsequently, a Tri-graph Spatio-temporal Fusion module synergistically models scale-specific dependencies by integrating an adaptive temporal graph, a static spatial graph, and a data-driven dynamic spatial graph within a unified architecture. Extensive experiments on four large-scale real-world benchmark datasets demonstrate that MS-GSTN consistently achieves superior forecasting accuracy compared to representative state-of-the-art models. Quantitatively, the proposed framework yields an overall reduction in Mean Absolute Error of up to 6.2% and maintains enhanced stability across multiple forecasting horizons. Visualization analysis further confirms that MS-GSTN effectively identifies scale-dependent spatial couplings, revealing that long-term traffic flow trends propagate through global network connectivity while short-term variations are governed by localized interactions.