Transformer is important to the electric power systems, and its accurate fault diagnosis is still hard. In the paper, a novel transformer fault diagnosis method using an Internet of Things (IoT) based monitoring system and an ensemble machine learning (EML) is presented. The monitoring system based on IoT technology consists of two parts: a data measurement subsystem and a data reception subsystem. Firstly, transformer vibration signals are measured by using the data measurement subsystem, and they are sent to the remote server by using the data reception subsystem. Then, an EML composed of deep belief networks (DBNs) and stacked denoising autoencoders (SDAs) with different activation functions, and relevance vector machines (RVMs) is proposed. DBNs and SDAs are respectively used to extract features from the signals, and RVMs are respectively employed as classifier. In order to ensure efficient of the EML, a novel combination strategy is proposed. A transformer fault diagnosis experiment is performed, and the diagnosis results confirm that the designed monitoring system can collect vibration signals effectively and long-distance, and the proposed EML can usefully remedy the inadequate information of features extracted by individual deep learning method, and its RVM classifier is obviously better than other commonly used classifiers.

变压器对电力系统很重要，其准确的故障诊断仍然很困难。本文提出了一种基于物联网（IoT）的监控系统和集成机器学习（EML）的变压器故障诊断方法。基于物联网技术的监控系统由两部分组成：数据测量子系统和数据接收子系统。首先，使用数据测量子系统测量变压器的振动信号，然后使用数据接收子系统将其发送到远程服务器。然后，提出了一种由深度信念网络（DBN）和具有不同激活功能的堆叠式降噪自动编码器（SDA）以及相关矢量机（RVM）组成的EML。DBN和SDA分别用于从信号中提取特征，RVM和RVM分别用作分类器。为了保证EML的有效性，提出了一种新颖的组合策略。进行了变压器故障诊断实验，诊断结果证实所设计的监控系统能够有效，远距离地采集振动信号，并且所提出的EML方法可以有效地弥补个体深度学习方法提取的特征信息不足及其RVM的不足。分类器明显优于其他常用分类器。