Raman spectroscopy, with its specific ability to generate a unique fingerprint-like spectrum of certain substances, has attracted much attention in diagnosing the ageing degree of oil–paper insulation. In this study, the feature extraction and ageing diagnosis methods of oil–paper insulation Raman spectroscopy data are further studied. Based on the non-linear analysis of Raman spectra of different ageing samples, kernel principal component analysis was applied to extract the spectral features, and the back-propagation neural network was used to build a diagnosis model with high diagnostic accuracy. The results show that Raman spectroscopy combined with kernel principal component analysis and the back-propagation neural network can diagnose the ageing state of oil–paper insulation, with a diagnostic accuracy of 91.43% (64/70). The proposed method provides an effective and feasible method for the ageing assessment of oil-immersed electrical equipment.

拉曼光谱法具有产生某些物质独特的指纹状光谱的特殊能力，在诊断油纸绝缘层的老化程度时引起了极大的关注。本研究进一步研究了油纸绝缘拉曼光谱数据的特征提取和老化诊断方法。在对不同老化样品的拉曼光谱进行非线性分析的基础上，采用核主成分分析提取光谱特征，并利用反向传播神经网络建立具有较高诊断精度的诊断模型。结果表明，拉曼光谱结合核主成分分析和反向传播神经网络可以诊断油纸绝缘的老化状态，诊断准确率为91.43％（64/70）。