中文摘要：

针对电力变压器故障诊断中的信息和知识具有随机性和不确定性的特点，提出了一种利用AdaBoostM1算法构建组合贝叶斯网络进行变压器故障诊断的方法。AdaBoostM1算法能够提高分类器的性能。为此．将若干个不同结构的TAN看作一系列基分类器，进行boosting迭代。即依次在训练集上训练每个基分类器。第1个基分类器用原始的训练集训练．其他基分类器的训练决定于在其之前产生的分类器的表现．被已有分类器错误判断的实例将以较大的概率出现在新分类器的训练集中，最后，这些分类器组合成为一个贝叶斯网络组合分类器。由于贝叶斯网络是用来表示变量间连接概率的图形模式．它提供了一种自然的表示因果信息的方法．用来发现数据间的潜在信息．因此应用中显示了该方法对于变压器故障诊断的适用性。在讨论变压器故障空间的基础上．针对已积累的故障变压器的大量油中溶解气体等数据，利用boosting迭代，并在此基础上构造出组合贝叶斯网络诊断模型，实现了变压器故障诊断，有利于提高诊断的准确性。此外．通过与其他组合诊断的方法进行比较进一步表明了该模型的有效性。

英文摘要：

Due to the randomness and uncertainty of power transformer fault diagnosis data, a transformer fault diagnostic method based on combinatorial Bayes network with AdaBoostM1 is proposed. As AdaBoostM1 algorithm can improve the performance of classifier, several different TAN（Tree Augmented Naive Bayes） are taken as a series of basic classifiers carrying out the boosting iteration. That is,every basic classifier is trained using training set. The first basic classifier is trained using the original training set and the others are trained according to the behavior of former one. Those objects that were wrong diagnosed are presented in the new training set with bigger probability. All these classifiers are composed to the combinatorial Bayes network. Bayes network is a graphic mode presenting the connection probability between variables,which is used to provide a way to represent the consequence information and to find the potential information ,very suitable for the transformer fault diagnosis. Based on the discussion of transformer fault space,the combinatorial Bayes network diagnosis model is constructed by the boosting iteration with the dissolved gas data of faulty transformers,which is used to realize the power transformer fault diagnosis with higher diagnostic correctness. Comparison with other combinatorial diagnostic methods shows its effectiveness.