AI 中文总结
研究物联网中CoAP拥塞控制问题，提出prCoAP方法，用线性支持向量回归集成预测RTO，结合校准随机森林丢弃分类器，经实验验证该方法在分组交付率和系统效率上表现良好。
AI 中文摘要
物联网网络需要轻量级应用层消息传递，CoAP因支持受限设备上基于UDP的REST式交互而成为一种选择。然而，CoAP拥塞控制仍依赖固定启发式方法，无法很好适应动态有损无线链路。本文提出prCoAP，一种轻量级数据驱动方法，用每次尝试的线性支持向量回归集成取代启发式重传超时（RTO）选择，直接根据节点可观测特征预测RTO。该模型在低端微控制器上运行，框架还包括校准随机森林丢弃分类器。通过离散事件模拟器和FIT IoT-LAB测试床评估，实验证实线性SVR实现97.25%的分组交付率，优于标准CoAP，内核SVR虽回归拟合更好，但线性SVR系统级效率更高。
英文摘要
Internet of Things (IoT) networks require lightweight application layer messaging, and CoAP is an option because it supports REST-style interactions over UDP on constrained devices. However, CoAP congestion control still depends on fixed heuristics, including binary exponential backoff (BEB) and RTT-based mechanisms such as CoCoA and CoCoA+, which do not adapt well to dynamic and lossy wireless links. This paper proposes prCoAP, a lightweight data-driven approach that replaces heuristic Retransmission Timeout (RTO) selection with a per-attempt linear Support Vector Regression (SVR) ensemble for direct RTO prediction from node-observable features. The model runs on-device on low-end microcontrollers and operates within strict memory and energy budgets. The framework also includes a calibrated Random Forest drop classifier that identifies likely-to-fail transactions in later retransmission attempts and terminates them early to reduce channel occupancy. We evaluate the approach using a discrete-event simulator implementing IEEE 802.15.4 and RFC 7252 and validate it against the FIT IoT-LAB testbed. Our experiments confirm that the proposed linear SVR achieves 97.25% PDR, outperforming standard CoAP under the evaluated conditions. We also evaluate a kernel SVR variant; while it improves regression fit (R2 0.84 vs. 0.63), the linear SVR provides better system-level efficiency, achieving comparable PDR with lower energy overhead.
CommentsSubmitted to WFIoT 2026 regular track
