DGA Ground Truth Datasets

The integration of artificial intelligence (AI) techniques in fault diagnosis of oil-filled transformers offers a promising approach to enhance the reliability and efficiency of power distribution networks. This paper presents a methodology for developing machine learning (ML) models using a no-code approach in Azure Machine Learning to diagnose transformer faults based on dissolved gas analysis (DGA). By leveraging multiclass classification algorithms, the study aims to accurately identify fault types such as partial discharges, thermal faults, low and high-energy discharges. The proposed method addresses challenges associated with conventional DGA interpretation methods, including incomplete ratio ranges and the absence of a fault-free area. The research includes a comprehensive evaluation of model performance through various preprocessing techniques, cross-validation and field trials, demonstrating the model's robustness and practical applicability.将人工智能（AI）技术应用于充油变压器的故障诊断，是提升配电网络可靠性与运行效率的极具前景的技术路径。本文提出一种依托Azure机器学习（Azure Machine Learning）平台、采用无代码开发方式构建机器学习（ML）模型的方法，基于溶解气体分析（DGA）实现变压器故障诊断。本研究借助多分类算法，旨在精准识别局部放电、热故障、低能与高能放电等故障类型。所提方法解决了传统溶解气体分析解读方法存在的比例区间不全、未覆盖无故障工况等技术痛点。研究通过多种预处理技术、交叉验证与现场试验对模型性能开展全面评估，验证了该模型的鲁棒性与实际应用价值。