Abstract：To address frequent topology changes, high-dimensional burst data and physical consistency issues in distribution line online monitoring, a physics-constrained decision tree (PhyDT) is proposed. A two-layer star-cluster hybrid topology is established to integrate edge star subnets, 5G/power line carrier backhaul and cloud-side Kalman filtering. Kirchhoff residual gating and sparse random projection are employed to compress features and reject outliers. An incremental topology-aware re-splitting algorithm updated only affects sub-trees when the network changes. The experimental results indicate that PhyDT improves accuracy by 1.7%~3.3% and macro-F1 by 1.8%~3.3% compared with light gradient boosting machine (LightGBM), deep forest, physics-informed neural networks (PINN) and gated recurrent unit-fully connected hybrid network (GRU-FC), while cutting incremental update time by 44.5%~59.7% and keeping inference latency at 5.1 ms. This study provides a new approach for real-time status assessment of distribution lines that balances accuracy, real-time performance, and topology adaptability, and has engineering promotion value.
姚明坤, 付丽伟, 田野, 薛明志, 李正日. 基于物联网技术与决策树的配电线路在线监测方法[J]. 电气技术, 2026, 27(3): 43-47.
YAO Mingkun, FU Liwei, TIAN Ye, XUE Mingzhi, LI Zhengri. Online monitoring method for distribution lines based on IoT technology and decision tree. Electrical Engineering, 2026, 27(3): 43-47.
2026, Vol. 27 
