Xi Gao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Xi Gao has a total of 22 co-authors. Here are the top co-authors by number of shared publications:
- Sheng Jin
Institute of Intelligent Transportation Systems, College of Civil Engineering and Architecture, Zhejiang University, Hangzhou 310058, China
4 shared publications - Wenbin Yao
School of Civil Engineering and Architecture, Zhejiang Sci-Tech University, Hangzhou 310018, China
4 shared publications - Congcong Bai
Institute of Intelligent Transportation Systems, College of Civil Engineering and Architecture, Zhejiang University, Hangzhou, China
2 shared publications Wentong Guo
Institute of Intelligent Transportation Systems, College of Civil Engineering and Architecture, Zhejiang University, Hangzhou, China
2 shared publicationsChengcheng Yang
1 shared publication- Chengcheng Yang
Institute of Intelligent Transportation Systems, College of Civil Engineering and Architecture, Zhejiang University, Hangzhou, China
1 shared publication
Latest Publications
- Open Access
- 15 August 2026
- Published by Elsevier in SSRN Electronic Journal
Heterogeneous Evidential Reconstruction and Uncertainty-Aware Temporal Forecasting for Charging Station Cold-Start
Accurate forecasting of electric vehicle (EV) charging load is critical for distribution-network operation and charging-infrastructure planning, yet newly deployed or temporarily unobserved stations may have no historical load records, creating a challenging cold-start setting. Existing inductive forecasting methods can transfer information from observed locations to unobserved stations, but they generally provide deterministic reconstructions and do not explicitly account for how reconstruction reliability should influence subsequent temporal state evolution. To address these limitations, we propose the Heterogeneous Evidential Reconstruction Network (HERO-Net) for EV charging-load forecasting at cold-start stations with no historical load observations. HERO-Net first reconstructs the missing historical load representation of a target station by integrating geographic relationships, street-view imagery, and remote-sensing information within a multisource probabilistic graph framework. The reconstructed historical state is modeled as a latent Gaussian distribution, allowing the model to capture both the inferred load-related representation and its associated uncertainty. Based on this reconstruction, a routed uncertainty-modulated GRU combines the reconstructed evidence with an environment- and time-conditioned temporal prior and adaptively controls its contribution to hidden-state updating according to reconstruction uncertainty and prior-relative information, thereby improving subsequent charging-load forecasting. Extensive experiments across eight Chinese cities demonstrate that HERO-Net achieves state-of-the-art performance, reducing average MAE by 15.3% on unobserved target stations (63.80 kWh) and 35.3% on observed stations (58.62 kWh) compared to leading baselines. Latent representations achieve a strong reconstruction correlation (R2= 0.82), and estimated uncertainties demonstrate solid statistical calibration. Crucially for power grid operation, HERO-Net lowers peak load forecast error by 18.6% and ramp error by 17.8%, providing actionable, risk-aware guidance for smart grid dispatching and energy storage allocation.
- 15 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessHeterogeneous Evidential Reconstruction and Uncertainty-Aware Temporal Forecasting for Charging Station Cold-Start
Accurate forecasting of electric vehicle (EV) charging load is critical for distribution-network operation and charging-infrastructure planning, yet newly deployed or temporarily unobserved stations may have no historical load records, creating a challenging cold-start setting. Existing inductive forecasting methods can transfer information from observed locations to unobserved stations, but they generally provide deterministic reconstructions and do not explicitly account for how reconstruction reliability should influence subsequent temporal state evolution. To address these limitations, we propose the Heterogeneous Evidential Reconstruction Network (HERO-Net) for EV charging-load forecasting at cold-start stations with no historical load observations. HERO-Net first reconstructs the missing historical load representation of a target station by integrating geographic relationships, street-view imagery, and remote-sensing information within a multisource probabilistic graph framework. The reconstructed historical state is modeled as a latent Gaussian distribution, allowing the model to capture both the inferred load-related representation and its associated uncertainty. Based on this reconstruction, a routed uncertainty-modulated GRU combines the reconstructed evidence with an environment- and time-conditioned temporal prior and adaptively controls its contribution to hidden-state updating according to reconstruction uncertainty and prior-relative information, thereby improving subsequent charging-load forecasting. Extensive experiments across eight Chinese cities demonstrate that HERO-Net achieves state-of-the-art performance, reducing average MAE by 15.3% on unobserved target stations (63.80 kWh) and 35.3% on observed stations (58.62 kWh) compared to leading baselines. Latent representations achieve a strong reconstruction correlation (R2= 0.82), and estimated uncertainties demonstrate solid statistical calibration. Crucially for power grid operation, HERO-Net lowers peak load forecast error by 18.6% and ramp error by 17.8%, providing actionable, risk-aware guidance for smart grid dispatching and energy storage allocation.
- 27 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Systems, Man, and Cybernetics: Systems
- Vol. PP (21682216) , 1-14
Physically Informed Driving Style Recognition for Intelligent Vehicle Assistance
Recognizing driving styles is essential for enhancing advanced driver-assistance systems (ADASs), as it enables the provision of personalized driving experiences while improving road safety. Although trajectory-based driving style recognition has been widely studied, existing approaches often rely on behavioral statistical features, single-model parameters, or latent representations, which may compromise interpretability and cross-scenario generalizability. To bridge this gap, this article proposes a physically informed framework for driving style recognition that integrates multiple physical car-following models with data-driven clustering and classification techniques. The framework extracts driving preferences from vehicle trajectory data by calibrating parameters across multiple physical car-following models and identifying key style-discriminative features. This approach enhances the interpretability and robustness of trajectory-based representation while reducing reliance on sparse in-vehicle sensor data. Experimental results reveal three behaviorally distinct driving styles—cautious efficient, aggressive efficient, and conservative safe—each reflecting distinct tradeoffs among efficiency, safety, and stability. The extracted rules exhibit high classification accuracy and generalizability across datasets, forming a solid basis for style-aware ADAS strategies. The proposed framework offers an interpretable and transferable trajectory-based solution for scalable driving style analysis, offering a solid foundation for intelligent vehicle assistance.
- 27 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Systems, Man, and Cybernetics: Systems
- Vol. PP (21682216) , 1-14
Physically Informed Driving Style Recognition for Intelligent Vehicle Assistance
Recognizing driving styles is essential for enhancing advanced driver-assistance systems (ADASs), as it enables the provision of personalized driving experiences while improving road safety. Although trajectory-based driving style recognition has been widely studied, existing approaches often rely on behavioral statistical features, single-model parameters, or latent representations, which may compromise interpretability and cross-scenario generalizability. To bridge this gap, this article proposes a physically informed framework for driving style recognition that integrates multiple physical car-following models with data-driven clustering and classification techniques. The framework extracts driving preferences from vehicle trajectory data by calibrating parameters across multiple physical car-following models and identifying key style-discriminative features. This approach enhances the interpretability and robustness of trajectory-based representation while reducing reliance on sparse in-vehicle sensor data. Experimental results reveal three behaviorally distinct driving styles—cautious efficient, aggressive efficient, and conservative safe—each reflecting distinct tradeoffs among efficiency, safety, and stability. The extracted rules exhibit high classification accuracy and generalizability across datasets, forming a solid basis for style-aware ADAS strategies. The proposed framework offers an interpretable and transferable trajectory-based solution for scalable driving style analysis, offering a solid foundation for intelligent vehicle assistance.
- 1 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 36 (4) , 4106-4122
Deep Adaptive Fusion of Multimodal Satellite-Street View Imagery for Fine-Grained Intersection Risk Mapping
The preponderance of traffic accidents takes place within urban intersection areas. It is crucial to extrapolate accident risk maps specifically for these locations to proactively mitigate and prevent future occurrences of traffic accidents. Nevertheless, inferring fine-grained intersection risk remains a challenging task, primarily due to the intricate structure of the road network, variability in scene information, and the stringent requirements for high-quality data. In this work, we propose an end-to-end Adaptive Risk-Feature Aware Fusion Network (ARFAF-Net) based on multimodal data to achieve fine-grained inference of traffic risk maps in intersection areas. Specifically, we introduce a Heterogenous Feature Adaptive Fusion Module to extract complementary features of risk from satellite imagery and streetscape data. The Dynamic Correlation Analysis Module is used to capture large-scale changes in risk in the region to improve multi-scale information perception. In addition, the Macro-Aware Guided Fusion Module is used to introduce macro-satellite image data features to enhance the accuracy of perceiving risks at the periphery of intersections. Ultimately, pixel-level extrapolation maps of intersection crash risk are generated, thereby offering more cost-effective and rational guidance for crash prevention. Both quantitative evaluation and qualitative analysis on real-world datasets demonstrate that the proposed ARFAF-Net achieves superior performance. Finally, the codes and models used in this study for intersection crash risk map inference are available at https://github.com/gwt-ZJU/ARFAF-Net.
- 1 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 36 (4) , 4106-4122
Deep Adaptive Fusion of Multimodal Satellite-Street View Imagery for Fine-Grained Intersection Risk Mapping
The preponderance of traffic accidents takes place within urban intersection areas. It is crucial to extrapolate accident risk maps specifically for these locations to proactively mitigate and prevent future occurrences of traffic accidents. Nevertheless, inferring fine-grained intersection risk remains a challenging task, primarily due to the intricate structure of the road network, variability in scene information, and the stringent requirements for high-quality data. In this work, we propose an end-to-end Adaptive Risk-Feature Aware Fusion Network (ARFAF-Net) based on multimodal data to achieve fine-grained inference of traffic risk maps in intersection areas. Specifically, we introduce a Heterogenous Feature Adaptive Fusion Module to extract complementary features of risk from satellite imagery and streetscape data. The Dynamic Correlation Analysis Module is used to capture large-scale changes in risk in the region to improve multi-scale information perception. In addition, the Macro-Aware Guided Fusion Module is used to introduce macro-satellite image data features to enhance the accuracy of perceiving risks at the periphery of intersections. Ultimately, pixel-level extrapolation maps of intersection crash risk are generated, thereby offering more cost-effective and rational guidance for crash prevention. Both quantitative evaluation and qualitative analysis on real-world datasets demonstrate that the proposed ARFAF-Net achieves superior performance. Finally, the codes and models used in this study for intersection crash risk map inference are available at https://github.com/gwt-ZJU/ARFAF-Net.
- 14 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 26 (11) , 19845-19861
A Robust Method for Bus Scheduling and Passenger Flow Coordination Considering Arterial Signal Coordination Under Connected Environment
Urban public transportation is a complex and open system integral to urban mobility. Its operation is often disrupted by various random factors, necessitating robust scheduling solutions. This study develops a bus robust scheduling model based on mixed-integer linear programming to enhance system resilience. First, an arterial signal coordination model is proposed for mixed traffic environments, enabling autonomous public transport vehicles to traverse intersections without stopping. Second, a demand-deterministic bus scheduling model is constructed, integrating timetables, trajectories, and origin-destination transfer schemes to balance passenger waiting time fairness and efficiency. Third, to address stochastic passenger demand during actual operations, a robust bus scheduling model is developed by incorporating robust constraints. Numerical experiments demonstrate that the demand-deterministic model generates optimal scheduling schemes when passenger demand remains within bus capacity. However, when passenger demand exceeds capacity, the demand-deterministic model becomes infeasible. In such scenarios, the robust scheduling model produces feasible schemes, albeit with reduced optimization, and its robustness can be tuned by adjusting model parameters. Additionally, practical management insights are provided for real-world applications.
- 14 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 26 (11) , 19845-19861
A Robust Method for Bus Scheduling and Passenger Flow Coordination Considering Arterial Signal Coordination Under Connected Environment
Urban public transportation is a complex and open system integral to urban mobility. Its operation is often disrupted by various random factors, necessitating robust scheduling solutions. This study develops a bus robust scheduling model based on mixed-integer linear programming to enhance system resilience. First, an arterial signal coordination model is proposed for mixed traffic environments, enabling autonomous public transport vehicles to traverse intersections without stopping. Second, a demand-deterministic bus scheduling model is constructed, integrating timetables, trajectories, and origin-destination transfer schemes to balance passenger waiting time fairness and efficiency. Third, to address stochastic passenger demand during actual operations, a robust bus scheduling model is developed by incorporating robust constraints. Numerical experiments demonstrate that the demand-deterministic model generates optimal scheduling schemes when passenger demand remains within bus capacity. However, when passenger demand exceeds capacity, the demand-deterministic model becomes infeasible. In such scenarios, the robust scheduling model produces feasible schemes, albeit with reduced optimization, and its robustness can be tuned by adjusting model parameters. Additionally, practical management insights are provided for real-world applications.