Xiaolin Liang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Xiaolin Liang has a total of 32 co-authors. Here are the top co-authors by number of shared publications:
- Wangbin Cao
North China Electric Power University
7 shared publications - Jianpeng Xu
College of Electronic and Information Engineering, HeBei University, BaoDing, China
3 shared publications - Xiongwen Zhao
School of Electrical and Electronic Engineering, North China Electric Power University, Beijing, China
3 shared publications Guangxing Wang
2 shared publications- Shuaiqi Liu
College of Electronic and Information Engineering, Hebei University, Machine Vision Engineering Research Center of Hebei Province, Baoding, China
2 shared publications Wangbin Cao
Hebei Key Laboratory of Power Internet of Things Technology, North China Electric Power University, Baoding, Hebei, China
2 shared publications
Latest Publications
- 1 December 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (23) , 60220-60234
Joint Optimization of Task Offloading and Digital Twin Synchronization in SAGIVNs: A Personalized Federated Deep Reinforcement Learning Approach
- 1 December 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (23) , 60220-60234
Joint Optimization of Task Offloading and Digital Twin Synchronization in SAGIVNs: A Personalized Federated Deep Reinforcement Learning Approach
- 17 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. PP (00189545) , 1-6
When Movable Antenna Meets URLLC Systems: A Deep Reinforcement Learning Approach
Movable antenna (MA) represents a promising technology for boosting wireless communication performance by enabling the repositioning of antennas within the given region. In this paper, we investigate one MA enabled uplink ultra reliable and low-latency communication (URLLC) system. Firstly, we characterize an optimization problem that seeks to acquire the maximization of the system's weighted sum rate via jointly optimizing the MA positions as well as transmit power at the users. Subsequently, we employ the twin-delayed deep deterministic policy gradient (TD3), a deep reinforcement learning algorithm, to address the weighted sum rate maximization problem. At last, simulation results validate that the proposed TD3-based algorithm exhibits notable advantages in enhancing the weighted sum rate compared to other benchmark algorithms including the fixed-position antenna algorithm.
- 17 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. PP (00189545) , 1-6
When Movable Antenna Meets URLLC Systems: A Deep Reinforcement Learning Approach
Movable antenna (MA) represents a promising technology for boosting wireless communication performance by enabling the repositioning of antennas within the given region. In this paper, we investigate one MA enabled uplink ultra reliable and low-latency communication (URLLC) system. Firstly, we characterize an optimization problem that seeks to acquire the maximization of the system's weighted sum rate via jointly optimizing the MA positions as well as transmit power at the users. Subsequently, we employ the twin-delayed deep deterministic policy gradient (TD3), a deep reinforcement learning algorithm, to address the weighted sum rate maximization problem. At last, simulation results validate that the proposed TD3-based algorithm exhibits notable advantages in enhancing the weighted sum rate compared to other benchmark algorithms including the fixed-position antenna algorithm.
- Open Access
- 23 February 2026
- Published by Elsevier in SSRN Electronic Journal
Deep Learning Enhancement for Wind Power Prediction Using Pattern Memory Bank and Multi-scale Feature Perceptive Modulation
Accurate wind power prediction is essential for secure and economical grid operation. However, conventional deep learning models for wind power prediction struggle to capture long-term dependencies and adapt to the rapid shifts in feature importance caused by varying weather conditions. To addresses these limitations, this paper proposes an enhanced wind power prediction model by constructing the Pattern Memory Bank (PMB) and Multi-scale Feature Perceptive Modulation (MFPM) and integrating them into conventional deep learning models as a closed-loop architecture. The PMB consists of a context encoder, a memory matrix, a triple importance tracking system and a gated read/write mechanism. As a differentiable external memory architecture, the PMB stores and updates key climatic patterns, enabling the model to capture extended temporal dependencies. The MFPM incorporates gating mechanisms and dilated convolutions to dynamically adjust feature weights across multiple time scales, thereby enhancing the model ability to represent complex relationships. The closed-loop optimization architecture is further to facilitate bidirectional interaction between PMB and MFPM, which ensures that feature selection reinforces memory retrieval, while memory optimization guides feature enhancement, ultimately improving the prediction performance. Experiments are conducted based on real-world wind farm datasets and the results are compared with the widely used conventional deep learning models. It shows that the proposed model achieves reductions in mean squared error and mean absolute error of 51.59% and 35.90%, respectively. These results demonstrate that the proposed enhanced wind power prediction model using PMB and MFPM significantly improves the accuracy and reliability of wind power prediction.
- 23 February 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessDeep Learning Enhancement for Wind Power Prediction Using Pattern Memory Bank and Multi-scale Feature Perceptive Modulation
Accurate wind power prediction is essential for secure and economical grid operation. However, conventional deep learning models for wind power prediction struggle to capture long-term dependencies and adapt to the rapid shifts in feature importance caused by varying weather conditions. To addresses these limitations, this paper proposes an enhanced wind power prediction model by constructing the Pattern Memory Bank (PMB) and Multi-scale Feature Perceptive Modulation (MFPM) and integrating them into conventional deep learning models as a closed-loop architecture. The PMB consists of a context encoder, a memory matrix, a triple importance tracking system and a gated read/write mechanism. As a differentiable external memory architecture, the PMB stores and updates key climatic patterns, enabling the model to capture extended temporal dependencies. The MFPM incorporates gating mechanisms and dilated convolutions to dynamically adjust feature weights across multiple time scales, thereby enhancing the model ability to represent complex relationships. The closed-loop optimization architecture is further to facilitate bidirectional interaction between PMB and MFPM, which ensures that feature selection reinforces memory retrieval, while memory optimization guides feature enhancement, ultimately improving the prediction performance. Experiments are conducted based on real-world wind farm datasets and the results are compared with the widely used conventional deep learning models. It shows that the proposed model achieves reductions in mean squared error and mean absolute error of 51.59% and 35.90%, respectively. These results demonstrate that the proposed enhanced wind power prediction model using PMB and MFPM significantly improves the accuracy and reliability of wind power prediction.
- 17 December 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 304
A VE-CONV2D model for wind power prediction based on frequency feature reconstruction and time efficiency optimization
- 17 December 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 304
A VE-CONV2D model for wind power prediction based on frequency feature reconstruction and time efficiency optimization
- 25 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (14) , 27450-27460
AoI Minimization for RIS-Assisted V2V Relay System With Deep Reinforcement Learning
By leveraging the inherent ability of reconfigurable intelligence surface (RIS) to enhance wireless communication channels, the integration of RIS into simultaneous wireless information and power transfer (SWIPT)-enabled vehicle-to-vehicle (V2V) systems presents a promising solution to jointly enhance communication performance and energy harvesting efficiency. Building on this potential, an RIS-assisted V2V dual-hop relay system is constructed, which deploys RIS on the gate of relay vehicle user equipment (VUE), enabling efficient signal refraction from source to relay VUEs. To address the critical challenge of information freshness in such RIS-assisted systems, Age of Information (AoI) is adopted as the key metric. And the AoI optimization problem is formulated that jointly considers energy/data buffer capacity limitations, relay sustainability, and real-time packet freshness. To effectively resolve this optimization problem under dynamic vehicular conditions, a prioritized experience replay-dueling double deep Q network (PER-D3QN) scheme based on deep reinforcement learning (DRL) is proposed to make the optimal relay decision for AoI minimization. Numerical results demonstrate that the average AoI using the proposed PER-D3QN scheme is reduced by 20% compared with the existing schemes.
- 25 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (14) , 27450-27460
AoI Minimization for RIS-Assisted V2V Relay System With Deep Reinforcement Learning
By leveraging the inherent ability of reconfigurable intelligence surface (RIS) to enhance wireless communication channels, the integration of RIS into simultaneous wireless information and power transfer (SWIPT)-enabled vehicle-to-vehicle (V2V) systems presents a promising solution to jointly enhance communication performance and energy harvesting efficiency. Building on this potential, an RIS-assisted V2V dual-hop relay system is constructed, which deploys RIS on the gate of relay vehicle user equipment (VUE), enabling efficient signal refraction from source to relay VUEs. To address the critical challenge of information freshness in such RIS-assisted systems, Age of Information (AoI) is adopted as the key metric. And the AoI optimization problem is formulated that jointly considers energy/data buffer capacity limitations, relay sustainability, and real-time packet freshness. To effectively resolve this optimization problem under dynamic vehicular conditions, a prioritized experience replay-dueling double deep Q network (PER-D3QN) scheme based on deep reinforcement learning (DRL) is proposed to make the optimal relay decision for AoI minimization. Numerical results demonstrate that the average AoI using the proposed PER-D3QN scheme is reduced by 20% compared with the existing schemes.
- 3 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (15) , 29291-29301
Adaptive CR-NOMA/OMA Scheduling Scheme and AoI Minimization for Vehicular Communication Systems
The Age of Information (AoI) as a new metric in vehicular communication system has begun to attract public attention. This article considers a muti-user vehicular communication network, where the base station (BS) transmits with vehicle users by adaptively selecting orthogonal multiple access (OMA) or non-OMA (NOMA) schemes. The cognitive radio (CR) technology has been performed in NOMA system to improve the spectrum resources utilization. The adaptive CR-NOMA/OMA scheduling scheme is proposed based on user matching algorithm to reduce the user interference of intergroups and intragroup. And the scheduling schemes for primary user and secondary user of CR-NOMA system using the generate-at-will (GAW) and the generate-at-request (GAR) data generation models are proposed. Then the nonlinear average AoI of CR-NOMA and OMA systems are formulized. Based on GAW and GAR models, the minimizing average AoI are derived. The Quality of Service (QoS) as to average AoI and achievable rate with different vehicle speeds, signal to interference plus noise ratio (SINR), transmit power and outage probability are analyzed. The simulation results are compared with validated analysis and better performances of vehicular systems can be achieved.
- 3 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (15) , 29291-29301
Adaptive CR-NOMA/OMA Scheduling Scheme and AoI Minimization for Vehicular Communication Systems
The Age of Information (AoI) as a new metric in vehicular communication system has begun to attract public attention. This article considers a muti-user vehicular communication network, where the base station (BS) transmits with vehicle users by adaptively selecting orthogonal multiple access (OMA) or non-OMA (NOMA) schemes. The cognitive radio (CR) technology has been performed in NOMA system to improve the spectrum resources utilization. The adaptive CR-NOMA/OMA scheduling scheme is proposed based on user matching algorithm to reduce the user interference of intergroups and intragroup. And the scheduling schemes for primary user and secondary user of CR-NOMA system using the generate-at-will (GAW) and the generate-at-request (GAR) data generation models are proposed. Then the nonlinear average AoI of CR-NOMA and OMA systems are formulized. Based on GAW and GAR models, the minimizing average AoI are derived. The Quality of Service (QoS) as to average AoI and achievable rate with different vehicle speeds, signal to interference plus noise ratio (SINR), transmit power and outage probability are analyzed. The simulation results are compared with validated analysis and better performances of vehicular systems can be achieved.
- 12 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 11 (24) , 41021-41032
Intelligent Reflecting Surface Assisted High-Mobile Edge Computing System: Joint Optimization of Computation and Communication
Mobile edge computing (MEC) can effectively cope with compute-intensive applications, but its full play is limited by the communication environment. Intelligent reflecting surface (IRS), as one of the key technologies of 6G, has the ability to improve the transmission environment and effectively improve the performance of the communication system. In this article, the IRS-assisted MEC system with high-mobile users (H-MEC) is considered. Based on the system model, the minimizing total energy consumption is investigated. Partial offload mode is adopted to jointly optimize computation and communication for H-MEC system. The minimizing total energy consumption is a complex nonconvex problem and the variables are coupled to each other, which is decomposed into five subproblems by the method of block coordinate descent (BCD) and the corresponding algorithms are proposed. Specifically, the quadratic transformation is used to deal with the fractional term, and the Lagrange dual transformation is used to deal with the logarithmic term. A low-complexity iterative algorithm is then used to alternately optimize the computation and communication settings. Finally, the effects of system parameters, such as the number of users and antennas, the computing task size, and the bandwidth of the system on the total energy consumption, are investigated. The results show that the proposed algorithms for IRS assisted H-MEC system can effectively improve the system performance as to energy consumption.
- 12 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 11 (24) , 41021-41032
Intelligent Reflecting Surface Assisted High-Mobile Edge Computing System: Joint Optimization of Computation and Communication
Mobile edge computing (MEC) can effectively cope with compute-intensive applications, but its full play is limited by the communication environment. Intelligent reflecting surface (IRS), as one of the key technologies of 6G, has the ability to improve the transmission environment and effectively improve the performance of the communication system. In this article, the IRS-assisted MEC system with high-mobile users (H-MEC) is considered. Based on the system model, the minimizing total energy consumption is investigated. Partial offload mode is adopted to jointly optimize computation and communication for H-MEC system. The minimizing total energy consumption is a complex nonconvex problem and the variables are coupled to each other, which is decomposed into five subproblems by the method of block coordinate descent (BCD) and the corresponding algorithms are proposed. Specifically, the quadratic transformation is used to deal with the fractional term, and the Lagrange dual transformation is used to deal with the logarithmic term. A low-complexity iterative algorithm is then used to alternately optimize the computation and communication settings. Finally, the effects of system parameters, such as the number of users and antennas, the computing task size, and the bandwidth of the system on the total energy consumption, are investigated. The results show that the proposed algorithms for IRS assisted H-MEC system can effectively improve the system performance as to energy consumption.
- 26 April 2024
- Published by Springer Nature
- p. 382-391
Power Allocation of Relay-Aided NOMA V2V System with Imperfect SIC
- 26 April 2024
- Published by Springer Nature
- p. 382-391
Power Allocation of Relay-Aided NOMA V2V System with Imperfect SIC
- 18 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Informatics
- Vol. 20 (2) , 2455-2464
Deep Reinforcement Learning for RIS-Aided Secure Mobile Edge Computing in Industrial Internet of Things
Mobile edge computing (MEC) has been regarded as a promising paradigm to support the compute-intensive and delay-sensitive industrial Internet of things (IIoT) applications. However, the nature of broadcasting in wireless communications may cause that the task offloading security is easy to be threatened from eavesdroppers. Aiming at improving the task offloading security, this article studies the benefit of deploying the emerging reconfigurable intelligent surface (RIS) in MEC-enabled IIoT networks with eavesdroppers, and forms the RIS-aided secure MEC system with time-division multiple access. In addition, we formulate a joint RIS phase shift, power control, local computation rate, and time-slot allocation optimization problem to maximize the weighted sum secrecy computation efficiency (WSSCE) among IIoT devices. To address this intractable problem, we propose a deep reinforcement learning (DRL)-based algorithm, where a deep deterministic policy gradient (DDPG) agent is adopted. Numerical results demonstrate that 1) deploying the RIS can improve the WSSCE performance; 2) the proposed DDPG-based algorithm can obtain higher WSSCE than other baseline methods.
- 18 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Informatics
- Vol. 20 (2) , 2455-2464
Deep Reinforcement Learning for RIS-Aided Secure Mobile Edge Computing in Industrial Internet of Things
Mobile edge computing (MEC) has been regarded as a promising paradigm to support the compute-intensive and delay-sensitive industrial Internet of things (IIoT) applications. However, the nature of broadcasting in wireless communications may cause that the task offloading security is easy to be threatened from eavesdroppers. Aiming at improving the task offloading security, this article studies the benefit of deploying the emerging reconfigurable intelligent surface (RIS) in MEC-enabled IIoT networks with eavesdroppers, and forms the RIS-aided secure MEC system with time-division multiple access. In addition, we formulate a joint RIS phase shift, power control, local computation rate, and time-slot allocation optimization problem to maximize the weighted sum secrecy computation efficiency (WSSCE) among IIoT devices. To address this intractable problem, we propose a deep reinforcement learning (DRL)-based algorithm, where a deep deterministic policy gradient (DDPG) agent is adopted. Numerical results demonstrate that 1) deploying the RIS can improve the WSSCE performance; 2) the proposed DDPG-based algorithm can obtain higher WSSCE than other baseline methods.