中文摘要：

为了研究自适应神经模糊推理系统（ANFIS）在油纸绝缘局部放电模式识别中的应用，针对大型电力变压器内常见的油纸绝缘局部放电缺陷，建立了3种基本局部放电缺陷模型，并在实验室中测取了各种缺陷类型的局部放电数据。通过对局部放电谱图的统计学计算以及分析，选取可有效表征局部放电类型的几个特征量．建立了ANFIS模型作为分类器，采用减法聚类生成规则，并利用梯度下降法和最小平方估计法相结合的混合学习算法进行训练，最后对该模型有效性进行测试。测试结果表明，不同类型局部放电的识别效果略有差异．但总体识别率达90％以上。研究结果表明，采用ANFIS进行局部放电模式识别，具有识别效果好、收敛速度快、稳定性高的优点。

英文摘要：

In order to study the pattern recognition techniques using adaptive neuro-fuzzy system （ANFIS） as the classifier, three PD defect models are established according to the common PD defects in oil/paper insulation of large electrical transformers. PD signals of the models are collected under different experiment conditions and are then statistically processed to obtain features for the classifier. Finally an ANFIS model is established adopting hybrid learning algorithm and subtractive clustering. The recognition results of the ANFIS model show that the accuracy rate is different between different models and the overall accuracy rate reaches above 90% with a satisfactory convergence rate and stability, which proves the feasibility of ANFIS in PD pattern recognition.