Minghao Zhu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Minghao Zhu has a total of 37 co-authors. Here are the top co-authors by number of shared publications:
Bo Gao
School of Computer Science and Technology, Beijing Jiaotong University, Beijing, China
7 shared publications- Tao Jing
School of Electronic and Information Engineering, Beijing Jiaotong University, Beijing, China
6 shared publications - Xuehan Li
School of Automation and Intelligence, Beijing Jiaotong University, Beijing, China
5 shared publications - Fei Richard Yu
Shenzhen University
4 shared publications - Boyang Zhang
School of Electronics and Information Engineering, Beijing Jiaotong University, Beijing, China
3 shared publications - Hongwei Wang
Beijing Jiaotong University
3 shared publications
Latest Publications
- 19 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. PP  (00189545) , 1-15
RAMA-STGCN: Risk-Aware Multi-Adjacency Spatio-Temporal Graph Convolutional Network for Traffic Flow Prediction
Traffic flow prediction faces challenges in spatial relationship modeling and risk-aware external factor integration. Current graph-based methods typically rely on single adjacency matrices that have limited ability to capture diverse spatial relationships, while external factors like weather and accidents are handled through simple concatenation strategies, ignoring their spatial propagation characteristics. This paper introduces Risk-Aware Multi-Adjacency Spatio-Temporal Graph Convolutional Network (RAMA-STGCN), a framework that systematically transforms risk-aware external factors into structured spatial knowledge. The approach integrates five complementary adjacency matrices: spatial distance (SP), multi-hop topology (TO), statistical correlation (SC), accident correlation (AC), and weather correlation (WC) matrices. An attention-based fusion mechanism automatically learns the relative importance of different spatial relationships, with AC and WC matrices capturing spatial risk correlation patterns derived from historical accident and weather statistics. Through comprehensive ablation studies across 15 architectural configurations and 13 adjacency matrix combinations, we establish evidence-based design principles. Extensive experiments on four benchmark datasets demonstrate substantial performance improvements: 6.3% MAE reduction on PEMS04, 3.6% on PEMS07, 5.1% on PEMS08, and 8.1% on PEMSBAY compared to state-of-the-art baselines. The framework achieves a lightweight design with 25-40% fewer parameters than complex baseline models while maintaining computational efficiency with inference times of 1.5-2.5ms per batch, making it suitable for real-time deployment in connected vehicle networks.
- 19 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. PP  (00189545) , 1-15
RAMA-STGCN: Risk-Aware Multi-Adjacency Spatio-Temporal Graph Convolutional Network for Traffic Flow Prediction
Traffic flow prediction faces challenges in spatial relationship modeling and risk-aware external factor integration. Current graph-based methods typically rely on single adjacency matrices that have limited ability to capture diverse spatial relationships, while external factors like weather and accidents are handled through simple concatenation strategies, ignoring their spatial propagation characteristics. This paper introduces Risk-Aware Multi-Adjacency Spatio-Temporal Graph Convolutional Network (RAMA-STGCN), a framework that systematically transforms risk-aware external factors into structured spatial knowledge. The approach integrates five complementary adjacency matrices: spatial distance (SP), multi-hop topology (TO), statistical correlation (SC), accident correlation (AC), and weather correlation (WC) matrices. An attention-based fusion mechanism automatically learns the relative importance of different spatial relationships, with AC and WC matrices capturing spatial risk correlation patterns derived from historical accident and weather statistics. Through comprehensive ablation studies across 15 architectural configurations and 13 adjacency matrix combinations, we establish evidence-based design principles. Extensive experiments on four benchmark datasets demonstrate substantial performance improvements: 6.3% MAE reduction on PEMS04, 3.6% on PEMS07, 5.1% on PEMS08, and 8.1% on PEMSBAY compared to state-of-the-art baselines. The framework achieves a lightweight design with 25-40% fewer parameters than complex baseline models while maintaining computational efficiency with inference times of 1.5-2.5ms per batch, making it suitable for real-time deployment in connected vehicle networks.
- 10 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. PP  (00189545) , 1-16
Dependable Task Offloading for CAVs in Web3: An Intelligence-Based Active Inference Approach
Connected and autonomous vehicles (CAVs) enhance traffic efficiency and safety via massive data-driven computation and decision-making. The computational demands of massive data challenge centralized cloud networks, leading to a novel CAV paradigm supported by mobile edge computing (MEC) and built on Web3. CAVs in Web3 can efficiently and securely offload compute-intensive tasks to edge devices in a decentralized and self-controlled manner, necessitating dependable task offloading schemes. However, existing deep reinforcement learning (DRL)-based offloading schemes face two challenges: overlooking security risks like privacy exposure in dependability definitions, while being constrained by reward function formulation, resulting in poor generalization. In this paper, we propose a dependable offloading scheme based on intelligence and active inference for CAVs in Web3. First, we introduce a dependable offloading framework utilizing double-layer blockchain and decentralized identifiers to ensure offloading source dependability. Then, by introducing a security-measuring dependability metric called cost from energy consumption, delay, and privacy exposure risk (cEDP), we formulate the dependable offloading optimization problem from an intelligence and active inference perspective, enabling higher-level environmental cognition without rewards. The problem is solved by the proposed intelligence-based active inference (INAI) algorithm. Experimental results demonstrate that reward-free INAI outperforms mainstream DRL and heuristic approaches in convergence, efficiency, and generalization capabilities.
- 10 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. PP  (00189545) , 1-16
Dependable Task Offloading for CAVs in Web3: An Intelligence-Based Active Inference Approach
Connected and autonomous vehicles (CAVs) enhance traffic efficiency and safety via massive data-driven computation and decision-making. The computational demands of massive data challenge centralized cloud networks, leading to a novel CAV paradigm supported by mobile edge computing (MEC) and built on Web3. CAVs in Web3 can efficiently and securely offload compute-intensive tasks to edge devices in a decentralized and self-controlled manner, necessitating dependable task offloading schemes. However, existing deep reinforcement learning (DRL)-based offloading schemes face two challenges: overlooking security risks like privacy exposure in dependability definitions, while being constrained by reward function formulation, resulting in poor generalization. In this paper, we propose a dependable offloading scheme based on intelligence and active inference for CAVs in Web3. First, we introduce a dependable offloading framework utilizing double-layer blockchain and decentralized identifiers to ensure offloading source dependability. Then, by introducing a security-measuring dependability metric called cost from energy consumption, delay, and privacy exposure risk (cEDP), we formulate the dependable offloading optimization problem from an intelligence and active inference perspective, enabling higher-level environmental cognition without rewards. The problem is solved by the proposed intelligence-based active inference (INAI) algorithm. Experimental results demonstrate that reward-free INAI outperforms mainstream DRL and heuristic approaches in convergence, efficiency, and generalization capabilities.
- Top Cited PapersOpen Access
- 27 January 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16  (1)
YOLO-DC for vehicle detection using deformable convolutional networks and cross-channel coordinate attention
Vehicle detection is crucial for intelligent decision support in transportation systems. However, real-time detection of vehicles is challenging due to geometric variations of vehicles and complex environmental factors such as light conditions and weather. To address these issues, the paper introduces the You-Only-Look-Once with Deformable Convolution and Cross-channel Coordinate Attention (YOLO-DC) framework that improves the performance and reliability of vehicle detection. First, YOLO-DC incorporates Cross-channel Coordinate Attention, which combines channel attention and coordinate attention, to more accurately cover target sampling positions and enhance feature extraction from vehicles of various shapes. Second, to better handle vehicles of different sizes, we employ Multi-scale Grouped Convolution to enable multi-scale awareness and streamline parameter sharing. Additionally, we incorporate channel prior convolutional attention so that the model can concentrate on areas of vehicles that are critical for detection. We also optimize feature fusion by leveraging a highly efficient fusion of C2f (CSP Bottleneck with 2 Convolutions) and FasterNet to reduce the model size. Experimental results demonstrate that YOLO-DC performs better than the state-of-the-art YOLOv8n method in detecting small, medium, and large-sized vehicles, and in detecting vehicles in adverse weather conditions. In addition to its superior performance, YOLO-DC also features fast detection speed, making it appropriate for real-time detection on devices with limited computational power.
- 27 January 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16  (1)
Top Cited PapersOpen AccessYOLO-DC for vehicle detection using deformable convolutional networks and cross-channel coordinate attention
Vehicle detection is crucial for intelligent decision support in transportation systems. However, real-time detection of vehicles is challenging due to geometric variations of vehicles and complex environmental factors such as light conditions and weather. To address these issues, the paper introduces the You-Only-Look-Once with Deformable Convolution and Cross-channel Coordinate Attention (YOLO-DC) framework that improves the performance and reliability of vehicle detection. First, YOLO-DC incorporates Cross-channel Coordinate Attention, which combines channel attention and coordinate attention, to more accurately cover target sampling positions and enhance feature extraction from vehicles of various shapes. Second, to better handle vehicles of different sizes, we employ Multi-scale Grouped Convolution to enable multi-scale awareness and streamline parameter sharing. Additionally, we incorporate channel prior convolutional attention so that the model can concentrate on areas of vehicles that are critical for detection. We also optimize feature fusion by leveraging a highly efficient fusion of C2f (CSP Bottleneck with 2 Convolutions) and FasterNet to reduce the model size. Experimental results demonstrate that YOLO-DC performs better than the state-of-the-art YOLOv8n method in detecting small, medium, and large-sized vehicles, and in detecting vehicles in adverse weather conditions. In addition to its superior performance, YOLO-DC also features fast detection speed, making it appropriate for real-time detection on devices with limited computational power.
- Open Access
- 1 January 2026
- Published by Tech Science Press in Computers, Materials & Continua
- Vol. 87  (2) , 1-10
Cloud-Edge-End Collaborative SC3 System in Smart Manufacturing: A Survey
With the deep integration of cloud computing, edge computing and the Internet of Things (IoT) technologies, smart manufacturing systems are undergoing profound changes. Over the past ten years, an extensive body of research on cloud-edge-end systems has been generated. However, challenges such as heterogeneous data fusion, real-time processing and system optimization still exist, and there is a lack of systematic review studies. In this paper, we review a cloud-edge-end collaborative sensing-communication-computing-control (SC3) system. This system integrates four layers of sensing, communication, computing and control to address the complex challenges of real-time decision making, resource scheduling and system optimization. The paper combs through the key implementation methods of intelligent sensing, data preprocessing, task offloading and resource allocation in this system, and analyzes their advantages and disadvantages. On this basis, feasible methods for overall system optimization are further explored. Finally, the paper summarizes the main challenges facing the deep integration of cloud-edge-end and proposes prospective research directions, providing a structured knowledge base and development framework for subsequent research. The paper aims to stimulate further exploration of multilevel collaborative mechanisms for smart manufacturing systems to enhance the real-time decision-making and overall performance of the smart manufacturing system.
- 1 January 2026
- Published by Tech Science Press in Computers, Materials & Continua
- Vol. 87  (2) , 1-10
Open AccessCloud-Edge-End Collaborative SC3 System in Smart Manufacturing: A Survey
With the deep integration of cloud computing, edge computing and the Internet of Things (IoT) technologies, smart manufacturing systems are undergoing profound changes. Over the past ten years, an extensive body of research on cloud-edge-end systems has been generated. However, challenges such as heterogeneous data fusion, real-time processing and system optimization still exist, and there is a lack of systematic review studies. In this paper, we review a cloud-edge-end collaborative sensing-communication-computing-control (SC3) system. This system integrates four layers of sensing, communication, computing and control to address the complex challenges of real-time decision making, resource scheduling and system optimization. The paper combs through the key implementation methods of intelligent sensing, data preprocessing, task offloading and resource allocation in this system, and analyzes their advantages and disadvantages. On this basis, feasible methods for overall system optimization are further explored. Finally, the paper summarizes the main challenges facing the deep integration of cloud-edge-end and proposes prospective research directions, providing a structured knowledge base and development framework for subsequent research. The paper aims to stimulate further exploration of multilevel collaborative mechanisms for smart manufacturing systems to enhance the real-time decision-making and overall performance of the smart manufacturing system.
- Open Access
- 1 January 2026
- Published by KEAI Publishing in High-speed Railway
Development of next-generation high-speed railways: Current status, trends, and suggestions
- 1 January 2026
- Published by KEAI Publishing in High-speed Railway
Open AccessDevelopment of next-generation high-speed railways: Current status, trends, and suggestions
- Open Access
- 1 January 2026
- Published by Tech Science Press in Computers, Materials & Continua
- Vol. 86  (2) , 1-17
Model Construction for Complex and Heterogeneous Data of Urban Road Traffic Congestion
Urban traffic generates massive and diverse data, yet most systems remain fragmented. Current approaches to congestion management suffer from weak data consistency and poor scalability. This study addresses this gap by proposing the Urban Traffic Congestion Unified Metadata Model (UTC-UMM). The goal is to provide a standardized and extensible framework for describing, extracting, and storing multisource traffic data in smart cities. The model defines a two-tier specification that organizes nine core traffic resource classes. It employs an eXtensible Markup Language (XML) Schema that connects general elements with resource-specific elements. This design ensures both syntactic and semantic interoperability across siloed datasets. Extension principles allow new elements or constraints to be introduced without breaking backward compatibility. A distributed pipeline is implemented using Hadoop Distributed File System (HDFS) and HBase. It integrates computer vision for video and natural language processing for text to automate metadata extraction. Optimized row-key designs enable low-latency queries. Performance is tested with the Yahoo! Cloud Serving Benchmark (YCSB), which shows linear scalability and high throughput. The results demonstrate that UTC-UMM can unify heterogeneous traffic data while supporting real-time analytics. The discussion highlights its potential to improve data reuse, portability, and scalability in urban congestion studies. Future research will explore integration with association rule mining and advanced knowledge representation to capture richer spatiotemporal traffic patterns.
- 1 January 2026
- Published by Tech Science Press in Computers, Materials & Continua
- Vol. 86  (2) , 1-17
Open AccessModel Construction for Complex and Heterogeneous Data of Urban Road Traffic Congestion
Urban traffic generates massive and diverse data, yet most systems remain fragmented. Current approaches to congestion management suffer from weak data consistency and poor scalability. This study addresses this gap by proposing the Urban Traffic Congestion Unified Metadata Model (UTC-UMM). The goal is to provide a standardized and extensible framework for describing, extracting, and storing multisource traffic data in smart cities. The model defines a two-tier specification that organizes nine core traffic resource classes. It employs an eXtensible Markup Language (XML) Schema that connects general elements with resource-specific elements. This design ensures both syntactic and semantic interoperability across siloed datasets. Extension principles allow new elements or constraints to be introduced without breaking backward compatibility. A distributed pipeline is implemented using Hadoop Distributed File System (HDFS) and HBase. It integrates computer vision for video and natural language processing for text to automate metadata extraction. Optimized row-key designs enable low-latency queries. Performance is tested with the Yahoo! Cloud Serving Benchmark (YCSB), which shows linear scalability and high throughput. The results demonstrate that UTC-UMM can unify heterogeneous traffic data while supporting real-time analytics. The discussion highlights its potential to improve data reuse, portability, and scalability in urban congestion studies. Future research will explore integration with association rule mining and advanced knowledge representation to capture richer spatiotemporal traffic patterns.
- 15 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Informatics
- Vol. 21  (12) , 9491-9502
D3QN-Enabled Diversified-Task Co-Offloading for Synthetic-Expense Minimization in Industrial Internet of Things (IIoT)
In recent years, during the neo-industrial revolution, mobile edge computing (MEC) offloading and peer offloading have received adequate attention in the Industrial Internet of Things (IIoT). In this article, we delve into task co-offloading on a single production line, which involves device-to-device (D2D) enabled peer offloading and 5G-enabled MEC offloading. The line conforms to industrial reality, with the focus on correlation, discrepancy of tasks, and industrial channel characteristics. However, the rapidly increasing transmission expense poses an obstacle to wireless offloading in IIoT. Confronting this thorny problem, we first construct a novel synthetic-expense (SE) function and accordingly deliberate the optimization problem as minimizing the sum of SE of diversified tasks. The SE minimization diversified-task co-offloading (SEIDEL) scheme is thoroughly designed afterward. Then, we conduct mathematical analyses of the corresponding offloading strategy, targeting different task categories, and introduce dueling double deep Q network (DQN) enabled SEIDEL (D3QN-SEIDEL) algorithm as the solution for the original nondeterministic polynomial hard problem. Eventually, the experiments based on a self-set production line demonstrate the excellent convergence of D3QN-SEIDEL and the superior performance of SEIDEL in handling D2D network fluctuations. Furthermore, the experiments reveal that SEIDEL offers a wide adjustable range for expense and latency, indicating significant potential for personalized applications. The experiments based on a practical line prove that SEIDEL exhibits higher potential ability in minimizing expense and delay in genuine IIoT scenarios.
- 15 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Informatics
- Vol. 21  (12) , 9491-9502
D3QN-Enabled Diversified-Task Co-Offloading for Synthetic-Expense Minimization in Industrial Internet of Things (IIoT)
In recent years, during the neo-industrial revolution, mobile edge computing (MEC) offloading and peer offloading have received adequate attention in the Industrial Internet of Things (IIoT). In this article, we delve into task co-offloading on a single production line, which involves device-to-device (D2D) enabled peer offloading and 5G-enabled MEC offloading. The line conforms to industrial reality, with the focus on correlation, discrepancy of tasks, and industrial channel characteristics. However, the rapidly increasing transmission expense poses an obstacle to wireless offloading in IIoT. Confronting this thorny problem, we first construct a novel synthetic-expense (SE) function and accordingly deliberate the optimization problem as minimizing the sum of SE of diversified tasks. The SE minimization diversified-task co-offloading (SEIDEL) scheme is thoroughly designed afterward. Then, we conduct mathematical analyses of the corresponding offloading strategy, targeting different task categories, and introduce dueling double deep Q network (DQN) enabled SEIDEL (D3QN-SEIDEL) algorithm as the solution for the original nondeterministic polynomial hard problem. Eventually, the experiments based on a self-set production line demonstrate the excellent convergence of D3QN-SEIDEL and the superior performance of SEIDEL in handling D2D network fluctuations. Furthermore, the experiments reveal that SEIDEL offers a wide adjustable range for expense and latency, indicating significant potential for personalized applications. The experiments based on a practical line prove that SEIDEL exhibits higher potential ability in minimizing expense and delay in genuine IIoT scenarios.
- 22 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12  (17) , 35048-35064
Double-Layer Blockchain and MEC Deployment Enabled Secure and Efficient Entity Interaction Framework for the Industrial IoT
The Industrial Internet of Things (IIoT), a core driver of Industrial 4.0, is considered as one of the most promising revolutionary technologies propelling the evolution of smart manufacturing toward Specialization, Reinforcement, Distinctiveness, and Innovation. The security and efficiency of smart manufacturing depend on the secure and efficient interaction of massive production data among entities. Yet, as a crucial measure of securing entity interactions, current authentication mechanisms overlook the single-point-of-failure issue and lightweight design. Moreover, interaction efficiency is rarely optimized and enhanced from the perspective of communication-supporting nodes. Paramountly, the assurance and optimization of entity interaction security and efficiency are strongly coupled, which is not considered in existing interaction frameworks. This article designs a three-layer entity interaction framework based on mobile edge computing (MEC) and blockchain technology. Specifically, the double-layer blockchain and MEC-cluster assisted lightweight authentication (BCLA) mechanism is proposed under the three-layer framework to achieve lightweight entity authentication in a weakly centralized manner. To optimize the entity interaction efficiency from joint authentication and transmission, this article further proposes an industrial edge server (IES) deployment optimization scheme and the proximity policy optimization-based IES deployment (PAID) algorithm. The security features and efficiency of the three-layer framework are demonstrated by carrying out security analysis and performance evaluation, which is based on the Hyperledger Fabric platform.
- 22 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12  (17) , 35048-35064
Double-Layer Blockchain and MEC Deployment Enabled Secure and Efficient Entity Interaction Framework for the Industrial IoT
The Industrial Internet of Things (IIoT), a core driver of Industrial 4.0, is considered as one of the most promising revolutionary technologies propelling the evolution of smart manufacturing toward Specialization, Reinforcement, Distinctiveness, and Innovation. The security and efficiency of smart manufacturing depend on the secure and efficient interaction of massive production data among entities. Yet, as a crucial measure of securing entity interactions, current authentication mechanisms overlook the single-point-of-failure issue and lightweight design. Moreover, interaction efficiency is rarely optimized and enhanced from the perspective of communication-supporting nodes. Paramountly, the assurance and optimization of entity interaction security and efficiency are strongly coupled, which is not considered in existing interaction frameworks. This article designs a three-layer entity interaction framework based on mobile edge computing (MEC) and blockchain technology. Specifically, the double-layer blockchain and MEC-cluster assisted lightweight authentication (BCLA) mechanism is proposed under the three-layer framework to achieve lightweight entity authentication in a weakly centralized manner. To optimize the entity interaction efficiency from joint authentication and transmission, this article further proposes an industrial edge server (IES) deployment optimization scheme and the proximity policy optimization-based IES deployment (PAID) algorithm. The security features and efficiency of the three-layer framework are demonstrated by carrying out security analysis and performance evaluation, which is based on the Hyperledger Fabric platform.
- 21 June 2025
- Published by Springer Nature
- p. 255-267
Diverse Delay-Sensitive Task Offloading and Resource Allocation for IIoT: An SoI Enhanced Approach
- 21 June 2025
- Published by Springer Nature
- p. 255-267
Diverse Delay-Sensitive Task Offloading and Resource Allocation for IIoT: An SoI Enhanced Approach