中文摘要：

为实现对用电系统低压端串联电弧故障的准确诊断，根据交流系统中低压串联电弧故障的奇异性、能量特性及不确定性，通过自主搭建的电弧故障模拟实验平台及不同负载下的串联电弧故障模拟实验，提出一种基于多特征融合的串联电弧故障诊断方法。该方法根据信号不同特性，结合小波变换理论对经降噪预处理后的采样信号进行主成分分析，提取各频段特性对信号的贡献率，并以信号3种特性中最大贡献率所在频段的空间位置关系作为特征向量构成1×3阶信号特性分布矩阵；将此矩阵作为网络的输入向量，利用改进多层前馈神经网络构建特征向量与电弧故障之间的映射关系。测试结果表明，该方法可减小电弧燃烧对诊断结果的影响，实现对串联电弧故障的诊断分类。

英文摘要：

To diagnose series arc faults occurred in low-voltage side at power utilization system accurately, according to three features of low-voltage series arc fault signals, namely the singularity, the uncertainty and the energy features, in AC power system and based on simulation experiment results of series arc faults under different loads carried out by the self-constructed experimental platform, a multi-feature fusion based series arc fault diagnosis method is proposed. According to the three features of signal and combining with wavelet transform theory, the proposed method performs principal component analysis （PCA） for sampled signals, whose noise is preprocessed and extracts contribution rates of different frequency bands to signal, then taking the spatial relations among frequency bands where the maximum contribution rates of the three features of the signal locate as the characteristic vectors, a 1＆#215;3 order signal features distribution matrix is constituted;taking this matrix as the input vector of improved multi-level BP neural network, using this neural network the mapping relation between characteristic vector and series arc fault is established. Test results show that using the proposed method the impact of burning arc on diagnosis result can be reduced and the diagnosis classification of series arc faults can be realized.