Novel Approach for Arc Fault Identification With Transient and Steady State Based Time-Frequency Analysis

The occurrence of electric arcs poses a huge threat to personal and equipment safety. As one of the effective ways to actively protect personnel and equipment away from serious arcing incidents, the signature recognition based arcing identification method has drawn much attention. However, since the strong nonlinear dynamics of the arcs, merely based on a specific time-domain or frequency-domain feature to develop the identification criteria may not be applicable in practice. To overcome the limitations, this article proposes a novel arc fault identification approach, which evaluated both the transient and steady dynamic states of arcing faults by time-frequency analysis. The complete ensemble empirical mode decomposition with adaptive noise, incorporated with Hilbert transform, has been designed to realize the rapid and reliable signatures extraction from arc fault profiles. Moreover, for dimension reduction purposes, correlation coefficient and partial least square regression based time-series dominant features selection method was developed. For ensuring the accuracy and robustness of the identification algorithm, a multiscenario based long short-term memory was also proposed. With the series of actual arc fault cases under different configurations, the effectiveness of the proposed method has.