基于PRPD图谱多特征融合的局部放电类型识别研究

局部放电(partial discharge, PD)信号的检测能够为电力系统提供绝缘缺陷诊断和运行状态评估。现有的局部放电类型识别算法难以有效识别相似度较高的绝缘缺陷，限制了其应用范围。为此，提出一种基于PRPD(phase resolved partial discharge)图谱多特征融合的局部放电类型识别算法。该算法利用卷积神经网络(convolutional neural network, CNN)提取局部放电PRPD图谱图像特征，将图像特征与PD信号统计特征进行有效融合，利用融合特征识别局部放电类型。在实验室环境下建立了4种局部放电模型，并进行了模拟对比实验。实验结果表明，相比传统的支持向量机(support vector machine, SVM)和反向传播神经网络(back propagation neural network, BPNN)算法，所提出方法的正确识别率分别提高了12.82%和19.70%，对相似度较高的缺陷类型也能进行有效识别，算法具有较好的鲁棒性。

Research on partial discharge type recognition based on multi-feature fusion of PRPD map

The detection of partial discharge signals can provide insulation defect diagnosis and operating state evaluation for power systems. However, the existing partial discharge type recognition algorithm is difficult to effectively identify the insulation defects with high similarity, which limits its application range. Therefore, this paper proposes a partial discharge type recognition algorithm based on multi-feature fusion of Phase-Resolved Partial Discharge (PRPD) map. The algorithm firstly extracts PRPD map image features by using convolutional neural network, then effectively fuses the image features with the statistical features of the PD signal, and then uses the fusion feature to identify the type of partial discharge. Four partial discharge models were established in the laboratory environment, and the simulation and comparison experiments were carried out. The experimental results show that compared with the traditional support vector machine and back propagation neural network algorithms, the correct recognition rate of the proposed method is improved by 12.82% and 19.70%, respectively, and it can effectively identify the defect types with high similarity, and the algorithm has better robustness.