A Round-robin Scheduling Algorithm of Vehicle Gateway Node Based on Self-adaptive Weighted Learning

Xiaoyong Zhang, Jun Peng, Shuo LiProceedings of the 2nd International Conference on Computer Science and Electronics Engineering (ICCSEE 2013) 2013 · DOI

Abstract

How to decrease the gateway queue delay in the Train Communication Network (TCN) with heavy communication load is a key problem. This paper models the queue scheduling as a reinforcement learning process and presents a self-adaptive scheduling algorithm of vehicle gateway node based on self-adaptive weighted learning. The proposed algorithm can schedule the queues dynamically. We then compare the simulation results respectively under two circumstances: with enough band resource or with limited band resource. The proposed algorithm's superiority is proved by simulation.

Introduction

In a Train Communication Network (TCN), the gateway node acts as a critical bottleneck where diverse data types converge—ranging from high-priority braking and signaling commands to lower-priority passenger information. Traditional scheduling methods often struggle with the "delay vs. fairness" trade-off:

  • First Come First Service (FCFS): Fails to prioritize urgent safety data.
  • Priority Queuing: Can lead to "starvation," where low-priority data is never transmitted.
  • Weighted Round Robin (WRR): Uses static weights that cannot adapt to the bursty nature of train traffic.

To address these issues, this paper proposes a self-adaptive scheduling algorithm that uses Reinforcement Learning (RL) to dynamically adjust queue weights in real-time based on network congestion.

Methodology

The scheduling problem is modeled as a Reinforcement Learning task where the gateway node acts as an agent. The goal is to learn an optimal scheduling policy that minimizes delay while maintaining fairness.

Reinforcement Learning Framework

  • State Representation ($S$): A binary vector indicating whether each queue has exceeded a specific length threshold ($R_i$). This simplified state space ensures rapid convergence.
  • Reward Function ($r$): The agent receives rewards based on how well it maintains queue lengths near their thresholds. A critical component is the Penalty Factor ($r_H$), which penalizes the agent for serving low-priority traffic when high-priority resources are strained.
  • Q-Learning: The agent employs the standard Q-learning update rule to estimate the future value of scheduling actions, allowing it to converge on an optimal strategy within approximately 2 seconds of simulation time.

Simulation Setup

The algorithm was evaluated using a Matlab simulation of a typical TCN topology. The setup includes three service sending ends (S1, S2, S3) and three receiving ends (D1, D2, D3) connected via two vehicle gateways.

The simulation topology used to evaluate the self-adaptive scheduling algorithm across two vehicle gateways.

Key Results and Discussion

The simulation compared the proposed self-adaptive algorithm against standard Weighted Round Robin (WRR) under varying bandwidth conditions (10 Mb/s and 2 Mb/s).

Threshold Maintenance and Stability

As shown in the simulation results, the proposed algorithm successfully kept queue lengths near their pre-defined targets (200, 400, and 600 bytes). Compared to WRR, the RL-based approach demonstrated significantly better stability. While WRR weights are static and can lead to queue growth during traffic shifts, the self-adaptive learner adjusted its weights dynamically to prevent any single queue from exceeding its threshold.

Comparison of queue lengths over time: the proposed adaptive algorithm maintains queue lengths closer to their thresholds with less variance than standard WRR.

Performance Under Scarcity

Even when bandwidth was restricted to 2 Mb/s, the algorithm maintained its effectiveness. The inclusion of the penalty factor ensured that critical control data (Queue 1) was prioritized, yet the round-robin nature of the underlying mechanism prevented the total starvation of low-priority data (Queue 3), which is a common failure point in strict priority-based systems.

Conclusion

The self-adaptive weighted learning algorithm provides a robust solution for TCN gateway scheduling. By treating the gateway as an intelligent agent, the system can balance the strict real-time requirements of train control with the fairness required for general data. Future work will focus on testing this algorithm on physical gateway hardware to assess processing overhead and exploring its scalability in longer train configurations.

Summary updated

Abstract

How to decrease the gateway queue delay in the Train Communication Network (TCN) with heavy communication load is a key problem. This paper models the queue scheduling as a reinforcement learning process and presents a self-adaptive scheduling algorithm of vehicle gateway node based on self-adaptive weighted learning. The proposed algorithm can schedule the queues dynamically. We then compare the simulation results respectively under two circumstance: with enough band resource or with limited band resource. The proposed algorithm's superiority is proved by simulation.
BibTeX
@inproceedings{zhang2013round,
  author = {Xiaoyong Zhang and Jun Peng and Shuo Li},
  title = {{A Round-robin Scheduling Algorithm of Vehicle Gateway Node Based on Self-adaptive Weighted Learning}},
  booktitle = {Proceedings of the 2nd International Conference on Computer Science and Electronics Engineering (ICCSEE 2013)},
  year = {2013},
  doi = {10.2991/iccsee.2013.436},
}
Figure