摘要：

变压器是大型电力系统中的关键重要设备, 其机理复杂且影响面广, 对变压器的状态检测与故障诊断是传统电力系统中的关键问题, 也是智能化时代下智能算法应用的重要方向。为解决现有的智能化故障诊断研究受限于故障样本稀缺、诊断结果可解释性差、模型更新困难等问题, 提出一种基于时序流数据动态分析的变压器故障诊断模型。首先通过人工辅助标注和数据增强方法, 构建具有序列特征的高置信故障数据样本库; 然后使用由融合时序特征分析器和多分类器构成的神经网络模型作为训练及分析的模型基础, 构造基于相似案例的推理方式, 通过距离相似性、模式相似性、形状相似性等多维度距离度量方法, 对实时检测到的流数据进行故障诊断及分类预警, 以指导运维人员结合历史经验及智能技术开展故障分析。实验结果表明, 所提方法在故障诊断的准确性与可解释性上显著提升, 可应用于变压器故障在线诊断真实场景中。

关键词: 变压器, 流数据分析, 故障诊断模型, 演进机制, 案例推理

Abstract:

Transformers, which possess complex mechanisms and exert an extensive influence, are important equipment in large power systems. The state detection and fault diagnosis of transformers are hence a key challenge in traditional power systems; they also constitute an important field for the application of intelligent algorithms in the era of intelligence. Existing intelligent fault diagnosis research is limited by the scarcity of fault samples, poor interpretability of diagnostic results, and difficulties in model updating. This paper proposes a transformer fault diagnosis model based on the dynamic analysis of time series flow data. First, a high-confidence fault data sample library with sequence features is built using manually assisted annotation and data augmentation methods, and a neural network model constituting a fusion temporal feature analyzer and multiple classifiers is constructed as the foundational model for training and analysis. A reasoning method based on similar cases is thus realized, using multidimensional distance measurement methods such as the distance similarity, pattern similarity, and shape similarity. This helps diagnose and classify real-time detection flow data for fault diagnosis and early warning, thereby helping operation and maintenance personnel conduct fault analysis based on historical experience and intelligent technology. Experimental verification confirms that the proposed method significantly improves the accuracy and interpretability of the fault diagnosis. The proposed method is applied to real scenarios of online transformer fault diagnosis.

Key words: transformer, stream data analysis, fault diagnosis model, evolutionary mechanism, case reasoning