Analysis and Research of Unbalanced Transformer Insulation Oil Monitoring Data Using Machine Learning Methods

• Published in: Automatic control and computer sciences Vol. 59; no. 2; pp. 230 - 243
• Main Authors: Shanghu Zhou, Mo, Bingyu, He, Yanjiao, Han, Menglong, Xie, Pengsheng, Li, Peixuan
• Format: Journal Article
• Language: English
• Published: Moscow Pleiades Publishing 01.04.2025; Springer Nature B.V
• Subjects: Algorithms; Big Data; Clustering; Computer Science; Control Structures and Microprogramming; Data analysis; Data collection; Datasets; Effectiveness; Insulation; Machine learning; Monitoring; Numerical methods; Preventive maintenance; transformers; SMOTE; machine learning; dissolved gas analysis
• ISSN: 0146-4116; 1558-108X
• DOI: 10.3103/S0146411625700191

As intelligent transformers continue to advance—a comprehensive preventive maintenance system has been gradually established for transformers. However, the main impediment to effective analysis is the imbalanced distribution of character data and positive anomaly data in the monitoring data of transformer oil, which adversely affects the intelligent evaluation of transformer status. Therefore, in this paper, we proposed analysis and research of unbalanced transformer insulation oil monitoring data using machine learning methods. First, we collected data pertaining to the status of insulation oil in smart transformers. Subsequently, it designed a numerical method based on word vector clustering tailored to the characteristics of insulation oil status data. Furthermore, a novel algorithm named KASMOTE (k nearest neighbor average smote) was introduced to process imbalanced insulation oil data. Finally, the paper validates the efficacy of the implemented dataset by employing seven machine learning algorithms. The experimental results demonstrate that the insulation monitoring dataset, incorporating word vector clustering and the KASMOTE algorithm, is both efficient and challenging, thus enhancing the feasibility of big data analysis.