Abstract: The traction transformer faults diagnosis method is the combination of artificial intelligence algorithm and Dissolved Gas Analysis（DGA）. However, due to the reasons of regeneration, sampling and chromatographic analysis, the data from DGA exist many uncertainties. In view of these, a new method combining the electric parameters and a new wavelet neural network model is proposed to diagnosis the traction transformer faults. The electric parameters work as input signal of the new network model. The hidden layer uses orthogonal Daubechies function as basis function. Learning and optimization algorithm adopts a kind of hybrid particle swarm optimization algorithm that introduces the concepts of quantum computation and immune algorithm. The test results show that the proposed intelligent faults diagnosis algorithm owns the faster diagnosis speed and higher accuracy.

Key words: on-board traction transformer, short-circuit fault, wavelet neural network, hybrid particle swarm optimization

摘要： 车载牵引变压器故障诊断的方法是将人工智能算法和油中气体分析法（DGA）相结合，但溶解气体由于再生、取样、色谱分析的原因，其数据存在许多的不确定性。提出将电气量与一种新的小波神经网络模型相结合的新方法来诊断牵引变压器故障。将电气量信号作为新网络模型的输入，网络的隐藏层采用具有正交性的Daubechies函数作为激函数，学习优化算法则采用引入量子计算和免疫算法的混合粒子群算法。试验结果证明，提出的智能故障诊断算法拥有更快的诊断速度和更高的准确率。

关键词: 车载牵引变压器, 短路故障, 小波神经网络, 混合粒子群算法