对故障空间的划分以及组合神经网络的构造方式 ,是利用组合神经网络进行变压器故障识别的关键。在讨论变压器故障空间划分方法及其存在问题的基础上 ,针对已积累的故障变压器的大量溶解气体数据 ,考察了各类故障的气体特征及聚类分析结果 ,并在此基础上构造了组合神经网络分层结构模型 ,实现了对变压器故障由粗到细的逐级划分 ,以提高诊断的准确性 ,为制定维修策略提供了依据。最后 ,结果显示了该模型的有效性 The division of the fault space and the construction of the combined neural network are the key points to identify the transformer fault using the combined neural network. On the basis of discussing the method of dividing the transformer fault space and its existing problems, aiming at the accumulated dissolved gas data of the fault transformer, the gas characteristics and cluster analysis results of various faults were investigated. Based on this, Network hierarchical model to achieve a breakdown of the transformer from coarse to fine step by step, in order to improve the diagnostic accuracy, to provide a basis for the development of maintenance strategy. Finally, the results show the validity of the model