基于改进经验模态分解的直流串联电弧故障检测

Direct Current Series Arc Fault Detection Based on Improved Empirical Mode Decomposition

针对直流系统中存在强噪声干扰时串联电弧故障检测准确度较低的问题,提出一种基于改进自适应噪声完备集合经验模态分解和模糊k均值聚类相结合的直流串联电弧故障检测方法;首先运用改进自适应噪声完备集合经验模态分解方法分解回路电流信号,得到多个本征模态函数;然后计算各本征模态函数的Hurst指数值以区分噪声分量和有用分量,将Hurst指数值大于0.5的有用分量进行重构;最后计算重构信号的峰峰值特征量和模糊熵特征量以构建特征向量作为模糊k均值聚类的输入,通过聚类中心的不同位置识别正常与故障状态。仿真与试验结果表明,所提出的方法区分系统正常与故障状态准确度为100%,区分系统干扰与故障状态准确度为93%,能有效识别直流串联电弧故障。

Aiming at the problem that accuracy of series arc fault detection was low when there was strong noise inter-ference in a direct current system,a direct current series arc fault detection method based on improved complete ensemble empirical mode decomposition with adaptive noise and fuzzy k-means clustering was proposed.Firstly,the improved complete ensemble empirical mode decomposition with adaptive noise method was used to decompose the loop current signal,and several intrinsic modal functions were obtained.Then,Hurst exponent value of each intrinsic modal function was calculated to distinguish the noise component from the useful component,and the useful components with Hurst exponent value greater than 0.5 were reconstructed.Finally,peak to peak value and fuzzy entropy of the reconstructed signal were calculated to construct the feature vector as the input of the fuzzy k-means clustering.Normal and fault states were recognized by using different locations of cluster centers.The simulation and test results show that the accuracy of the proposed method to distinguish the normal and fault states is 100%,and the accuracy to distinguish the interference and fault states is 93%,which can effectively identify direct current series arc faults.