AI-Based Models for Distance and Region Estimation of High Impedance Faults Under Variate Scenarios

High-impedance faults (HIFs) occur when an energized distribution system (DS) conductor comes into contact with a high-impedance surface. Their erratic behavior and low fault current make them difficult to detect with conventional protection schemes, and HIF location, in particular, remains a scarcely explored and challenging problem. Recent advances in artificial intelligence (AI), however, offer promising opportunities to design more robust solutions. Hence, this paper proposes and evaluates an AI-based HIF location method, addressing both fault distance and region estimation. The algorithm is evaluated under variations in distributed generation (DG) output, system loading, and background noise levels, as these scenarios changes occur in real DSs. To account for these variations, two distinct strategies for data partition are introduced, enabling the method’s assessment in scenarios not considered during training. In addition, different algorithmic configurations, such as input window size, signal sampling frequency, and measurements devices are analyzed to ensure practical applicability. The results show that leveraging measurements from multiple distributed meters significantly improves accuracy. Moreover, training algorithms exclusively on extreme scenarios regarding loading and DG power in DSs can be sufficient to ensure robust performance across intermediate conditions, thereby enhancing the practical viability of AI-based HIF location methods.