Zhengkun He
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Zhengkun He has a total of 13 co-authors. Here are the top co-authors by number of shared publications:
- Bin Wang
School of Computer Science and engineering, Central South University, Changsha, China
2 shared publications - Jinfang Sheng
School of Computer Science and engineering, Central South University, Changsha, China
2 shared publications - Jun Wang
School of Information Engineering, Henan University of Science and Technology, Luoyang 471003, China
2 shared publications - Xunyang Wang
Department of Applied Mathematics Lanzhou University of Technology Lanzhou China
2 shared publications - Yu Chen
School of Computer Science and Engineering, Central South University, Changsha 410083, China
2 shared publications - Zhuangzhuang Du
National Innovation Center for Digital Fishery China Agricultural University Beijing China
2 shared publications
Latest Publications
- 13 April 2023
- Published by SAGE Publications in Proceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile Engineering
- Vol. 238 (9) , 2880-2888
Multi-agent deep reinforcement learning with actor-attention-critic for traffic light control
In recent years, with the increase of urbanization and car ownership, urban traffic congestion have become increasingly prominent. Traffic light control can effectively reduce urban traffic congestion. In the research of controlling traffic lights of multiple intersections, most methods introduced theories related to deep reinforcement learning, but few methods considered the information interaction between intersections or the way of information interaction is unreasonable. Inspired by this, this paper proposes a multi-agent deep reinforcement learning with actor-attention-critic network for traffic light control (MAAC-TLC) algorithm. In MAAC-TLC, each agent introduces the attention mechanism in the process of learning, so that it will not pay attention to all the information of other agents indiscriminately, but only focus on the important information of the agents that plays an important role in it, so as to ensure that all intersections can learn the optimal policy. Finally, the traffic lights at each intersection in the MAAC-TLC algorithm are controlled according to its own policy, thereby improving the traffic efficiency of the traffic network. The experimental results on light and heavy traffic flow scenarios have demonstrated that MAAC-TLC can improve traffic congestion at multiple intersections effectively.
- 13 April 2023
- Published by SAGE Publications in Proceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile Engineering
- Vol. 238 (9) , 2880-2888
Multi-agent deep reinforcement learning with actor-attention-critic for traffic light control
In recent years, with the increase of urbanization and car ownership, urban traffic congestion have become increasingly prominent. Traffic light control can effectively reduce urban traffic congestion. In the research of controlling traffic lights of multiple intersections, most methods introduced theories related to deep reinforcement learning, but few methods considered the information interaction between intersections or the way of information interaction is unreasonable. Inspired by this, this paper proposes a multi-agent deep reinforcement learning with actor-attention-critic network for traffic light control (MAAC-TLC) algorithm. In MAAC-TLC, each agent introduces the attention mechanism in the process of learning, so that it will not pay attention to all the information of other agents indiscriminately, but only focus on the important information of the agents that plays an important role in it, so as to ensure that all intersections can learn the optimal policy. Finally, the traffic lights at each intersection in the MAAC-TLC algorithm are controlled according to its own policy, thereby improving the traffic efficiency of the traffic network. The experimental results on light and heavy traffic flow scenarios have demonstrated that MAAC-TLC can improve traffic congestion at multiple intersections effectively.
- Open Access
- 1 December 2022
- Published by MDPI AG in Mathematics
- Vol. 10 (23) , 4563
Attributed Graph Embedding Based on Attention with Cluster
Graph embedding is of great significance for the research and analysis of graphs. Graph embedding aims to map nodes in the network to low-dimensional vectors while preserving information in the original graph of nodes. In recent years, the appearance of graph neural networks has significantly improved the accuracy of graph embedding. However, the influence of clusters was not considered in existing graph neural network (GNN)-based methods, so this paper proposes a new method to incorporate the influence of clusters into the generation of graph embedding. We use the attention mechanism to pass the message of the cluster pooled result and integrate the whole process into the graph autoencoder as the third layer of the encoder. The experimental results show that our model has made great improvement over the baseline methods in the node clustering and link prediction tasks, demonstrating that the embeddings generated by our model have excellent expressiveness.
- 1 December 2022
- Published by MDPI AG in Mathematics
- Vol. 10 (23) , 4563
Open AccessAttributed Graph Embedding Based on Attention with Cluster
Graph embedding is of great significance for the research and analysis of graphs. Graph embedding aims to map nodes in the network to low-dimensional vectors while preserving information in the original graph of nodes. In recent years, the appearance of graph neural networks has significantly improved the accuracy of graph embedding. However, the influence of clusters was not considered in existing graph neural network (GNN)-based methods, so this paper proposes a new method to incorporate the influence of clusters into the generation of graph embedding. We use the attention mechanism to pass the message of the cluster pooled result and integrate the whole process into the graph autoencoder as the third layer of the encoder. The experimental results show that our model has made great improvement over the baseline methods in the node clustering and link prediction tasks, demonstrating that the embeddings generated by our model have excellent expressiveness.
- Open Access
Deep Reinforcement Learning for Traffic Light Timing Optimization
Existing inflexible and ineffective traffic light control at a key intersection can often lead to traffic congestion due to the complexity of traffic dynamics, how to find the optimal traffic light timing strategy is a significant challenge. This paper proposes a traffic light timing optimization method based on double dueling deep Q-network, MaxPressure, and Self-organizing traffic lights (SOTL), namely EP-D3QN, which controls traffic flows by dynamically adjusting the duration of traffic lights in a cycle, whether the phase is switched based on the rules we set in advance and the pressure of the lane. In EP-D3QN, each intersection corresponds to an agent, and the road entering the intersection is divided into grids, each grid stores the speed and position of a car, thus forming the vehicle information matrix, and as the state of the agent. The action of the agent is a set of traffic light phase in a signal cycle, which has four values. The effective duration of the traffic lights is 0–60 s, and the traffic light phases switching depends on its press and the rules we set. The reward of the agent is the difference between the sum of the accumulated waiting time of all vehicles in two consecutive signal cycles. The SUMO is used to simulate two traffic scenarios. We selected two types of evaluation indicators and compared four methods to verify the effectiveness of EP-D3QN. The experimental results show that EP-D3QN has superior performance in light and heavy traffic flow scenarios, which can reduce the waiting time and travel time of vehicles, and improve the traffic efficiency of an intersection.
Open AccessDeep Reinforcement Learning for Traffic Light Timing Optimization
Existing inflexible and ineffective traffic light control at a key intersection can often lead to traffic congestion due to the complexity of traffic dynamics, how to find the optimal traffic light timing strategy is a significant challenge. This paper proposes a traffic light timing optimization method based on double dueling deep Q-network, MaxPressure, and Self-organizing traffic lights (SOTL), namely EP-D3QN, which controls traffic flows by dynamically adjusting the duration of traffic lights in a cycle, whether the phase is switched based on the rules we set in advance and the pressure of the lane. In EP-D3QN, each intersection corresponds to an agent, and the road entering the intersection is divided into grids, each grid stores the speed and position of a car, thus forming the vehicle information matrix, and as the state of the agent. The action of the agent is a set of traffic light phase in a signal cycle, which has four values. The effective duration of the traffic lights is 0–60 s, and the traffic light phases switching depends on its press and the rules we set. The reward of the agent is the difference between the sum of the accumulated waiting time of all vehicles in two consecutive signal cycles. The SUMO is used to simulate two traffic scenarios. We selected two types of evaluation indicators and compared four methods to verify the effectiveness of EP-D3QN. The experimental results show that EP-D3QN has superior performance in light and heavy traffic flow scenarios, which can reduce the waiting time and travel time of vehicles, and improve the traffic efficiency of an intersection.
- Open Access
- 27 September 2022
- Published by Springer Nature in BioData Mining
- Vol. 15 (1) , 1-19
Interpretable recurrent neural network models for dynamic prediction of the extubation failure risk in patients with invasive mechanical ventilation in the intensive care unit
Background: Clinical decision of extubation is a challenge in the treatment of patient with invasive mechanical ventilation (IMV), since existing extubation protocols are not capable of precisely predicting extubation failure (EF). This study aims to develop and validate interpretable recurrent neural network (RNN) models for dynamically predicting EF risk. Methods: A retrospective cohort study was conducted on IMV patients from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. Time series with a 4-h resolution were built for all included patients. Two types of RNN models, the long short-term memory (LSTM) and the gated recurrent unit (GRU), were developed. A stepwise logistic regression model was used to select key features for developing light-version RNN models. The RNN models were compared to other five non-temporal machine learning models. The Shapley additive explanations (SHAP) value was applied to explain the influence of the features on model prediction. Results: Of 8,599 included patients, 2,609 had EF (30.3%). The area under receiver operating characteristic curve (AUROC) of LSTM and GRU showed no statistical difference on the test set (0.828 vs. 0.829). The light-version RNN models based on the 26 features selected out of a total of 89 features showed comparable performance as their corresponding full-version models. Among the non-temporal models, only the random forest (RF) (AUROC: 0.820) and the extreme gradient boosting (XGB) model (AUROC: 0.823) were comparable to the RNN models, but their calibration was deviated. Conclusions: The RNN models have excellent predictive performance for predicting EF risk and have potential to become real-time assistant decision-making systems for extubation.
- 27 September 2022
- Published by Springer Nature in BioData Mining
- Vol. 15 (1) , 1-19
Open AccessInterpretable recurrent neural network models for dynamic prediction of the extubation failure risk in patients with invasive mechanical ventilation in the intensive care unit
Background: Clinical decision of extubation is a challenge in the treatment of patient with invasive mechanical ventilation (IMV), since existing extubation protocols are not capable of precisely predicting extubation failure (EF). This study aims to develop and validate interpretable recurrent neural network (RNN) models for dynamically predicting EF risk. Methods: A retrospective cohort study was conducted on IMV patients from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. Time series with a 4-h resolution were built for all included patients. Two types of RNN models, the long short-term memory (LSTM) and the gated recurrent unit (GRU), were developed. A stepwise logistic regression model was used to select key features for developing light-version RNN models. The RNN models were compared to other five non-temporal machine learning models. The Shapley additive explanations (SHAP) value was applied to explain the influence of the features on model prediction. Results: Of 8,599 included patients, 2,609 had EF (30.3%). The area under receiver operating characteristic curve (AUROC) of LSTM and GRU showed no statistical difference on the test set (0.828 vs. 0.829). The light-version RNN models based on the 26 features selected out of a total of 89 features showed comparable performance as their corresponding full-version models. Among the non-temporal models, only the random forest (RF) (AUROC: 0.820) and the extreme gradient boosting (XGB) model (AUROC: 0.823) were comparable to the RNN models, but their calibration was deviated. Conclusions: The RNN models have excellent predictive performance for predicting EF risk and have potential to become real-time assistant decision-making systems for extubation.
- Open Access
- 21 December 2020
- Published by Wiley in Complexity
- Vol. 2020, 1-13
A Cluster-Head Rotating Election Routing Protocol for Energy Consumption Optimization in Wireless Sensor Networks
Balancing energy consumption using the clustering routing algorithms is one of the most practical solutions for prolonging the lifetime of resource-limited wireless sensor networks (WSNs). However, existing protocols cannot adequately minimize and balance the total network energy dissipation due to the additional tasks of data acquisition and transmission of cluster heads. In this paper, a cluster-head rotating election routing protocol is proposed to alleviate the problem. We discovered that the regular hierarchical clustering method and the scheme of cluster-head election area division had positive effects on reducing the energy consumption of cluster head election and intracluster communication. The election criterion composed of location and residual energy factor was proved to lower the probability of premature death of cluster heads. The chain multihop path of intercluster communication was performed to save the energy of data aggregation to the base station. The simulation results showed that the network lifetime can be efficiently extended by regulating the adjustment parameters of the protocol. Compared with LEACH, I-LEACH, EEUC, and DDEEC, the algorithm demonstrated significant performance advantages by using the number of active nodes and residual energy of nodes as the evaluation indicators. On the basis of these results, the proposed routing protocols can be utilized to increase the capability of WSNs against energy constraints.
- 21 December 2020
- Published by Wiley in Complexity
- Vol. 2020, 1-13
Open AccessA Cluster-Head Rotating Election Routing Protocol for Energy Consumption Optimization in Wireless Sensor Networks
Balancing energy consumption using the clustering routing algorithms is one of the most practical solutions for prolonging the lifetime of resource-limited wireless sensor networks (WSNs). However, existing protocols cannot adequately minimize and balance the total network energy dissipation due to the additional tasks of data acquisition and transmission of cluster heads. In this paper, a cluster-head rotating election routing protocol is proposed to alleviate the problem. We discovered that the regular hierarchical clustering method and the scheme of cluster-head election area division had positive effects on reducing the energy consumption of cluster head election and intracluster communication. The election criterion composed of location and residual energy factor was proved to lower the probability of premature death of cluster heads. The chain multihop path of intercluster communication was performed to save the energy of data aggregation to the base station. The simulation results showed that the network lifetime can be efficiently extended by regulating the adjustment parameters of the protocol. Compared with LEACH, I-LEACH, EEUC, and DDEEC, the algorithm demonstrated significant performance advantages by using the number of active nodes and residual energy of nodes as the evaluation indicators. On the basis of these results, the proposed routing protocols can be utilized to increase the capability of WSNs against energy constraints.
- Open Access
- 4 November 2020
- Published by Wiley in Complexity
- Vol. 2020, 1-17
Bioinspired Mitigation Scheme for Cascading Failures in Farmland Wireless Sensor Networks
Existing mitigation strategies on wireless sensor networks (WSNs) against cascading failures cannot appropriately adapt the particular characteristics of farmland WSNs. Spider web provides a new reference for improving network invulnerability. In this study, a bionic network scheme is built based on symmetric analysis of a series of spider-web vibration transmission trials, which include networking methodology, communication rules, and load capacity model. The basic idea of this scheme is to apply the cascading-failure coping mechanism inspired by spider web into the construction and operation of farmland WSNs. We found that the link backup contributed by a topological structure and communication rules had positive effects on suppressing the spread of cascading failures. The study showed that the damages of cascading failures can be efficiently lowered by regulating the adjustment coefficient of the load capacity model. The difference between the inner-layer node failures and outer-layer node failures for network invulnerability was verified under deliberate attack circumstances. Based on these results, the proposed network scheme can be utilized to enhance the invulnerability performance of farmland WSNs.
- 4 November 2020
- Published by Wiley in Complexity
- Vol. 2020, 1-17
Open AccessBioinspired Mitigation Scheme for Cascading Failures in Farmland Wireless Sensor Networks
Existing mitigation strategies on wireless sensor networks (WSNs) against cascading failures cannot appropriately adapt the particular characteristics of farmland WSNs. Spider web provides a new reference for improving network invulnerability. In this study, a bionic network scheme is built based on symmetric analysis of a series of spider-web vibration transmission trials, which include networking methodology, communication rules, and load capacity model. The basic idea of this scheme is to apply the cascading-failure coping mechanism inspired by spider web into the construction and operation of farmland WSNs. We found that the link backup contributed by a topological structure and communication rules had positive effects on suppressing the spread of cascading failures. The study showed that the damages of cascading failures can be efficiently lowered by regulating the adjustment coefficient of the load capacity model. The difference between the inner-layer node failures and outer-layer node failures for network invulnerability was verified under deliberate attack circumstances. Based on these results, the proposed network scheme can be utilized to enhance the invulnerability performance of farmland WSNs.