Fusion-based congestion control method for the power internet of things combining data-driven and rule-engine models

Abstract

The integration of the Internet of Things (IoT) into smart grids is placing an ever-growing emphasis on achieving high transmission stability. With the massive integration of heterogeneous devices into the power IoT, the volume of transmitted data has surged, and the dynamic complexity of networks has increased, leading to prolonged data congestion due to delayed transmission. Existing congestion control algorithms are typically divided into rule-based and learning-based approaches. However, relying exclusively on a single type of algorithm can result in challenges like suboptimal bandwidth utilization or slow convergence. To enhance data transmission stability, this paper introduces a congestion control method that integrates rule-based models with data-driven models. The method integrates network temporal features extracted by the Transformer algorithm with rule-based features from Google's Congestion Control (GCC) algorithm, and employs Proximal Policy Optimization (PPO) reinforcement learning to generate real-time transmission rate control strategies. This approach improves both the efficiency and robustness of deep reinforcement learning algorithms for congestion control. The performance evaluation metrics used for comparison are Quality of Service (QoS), mainly focusing on throughput, latency, and average packet loss rate. Compared to other algorithms, this model achieves the best performance with a score of 88.13. The hybrid algorithm offers a promising direction for future network congestion control.