Answers are grounded in this paper and cite their sources.
In constructing the new type of power system, the installed capacity of various electrical equipment such as transformers, motors, distribution terminals, and switchgears keeps expanding. Challenges including variable operating conditions of equipment, scarce early-fault samples, and heterogeneous mixed multi-source monitoring data have become increasingly prominent. Traditional manual inspection and fixed-threshold early-warning modes can hardly realize advance identification and accurate localization of minor faults. Boasting advantages such as powerful nonlinear fitting, autonomous feature extraction, and small-sample adaptability, machine learning has become a core technical approach for intelligent monitoring and early fault warning of electrical equipment. Based on 17 relevant literatures, this paper systematically sorts out existing research achievements from five dimensions: data preprocessing, feature engineering, diagnosis and early-warning models, lightweight deployment, and engineering application. It compares applicable scenarios and performance of Random Forest, Support Vector Machine, LSTM, CNN, Dynamic Graph Convolutional Neural Network (DGCNN), lightweight MobileNet, and intelligent optimization ensemble algorithms. This paper summarizes key bottlenecks: insufficient generalization ability under small-sample conditions, low fusion degree of multimodal data, edge-end deployment constraints and poor model interpretability. Combined with digital twin and edge computing, this paper prospects future research directions, providing theoretical and technical support for constructing a full-lifecycle intelligent operation-maintenance system of electrical equipment.
No takes yet. Share an insight, caveat, or question.
Yuchen Zhu (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: