An Artificial Neural Network-Based Method To Improve Selectivity In Fault Detection Algorithms For Wind Farm Collector Lines

This paper presents an advanced methodology for fault detection and selectivity improvement in Wind Farm Collector Networks (WFCNs). Usual detection algorithms naturally detect faults in WFCNs, but are not able to indicate in which Collector Bus (CB) the fault occurred. In this approach, a Current Slope-Based (CSB) fault detection algorithm is enhanced by utilizing the Discrete Stockwell Transform (DST) for feature extraction, followed by Artificial Neural Networks (ANNs) for the intelligent classification of fault regions. This combination allows applying only voltage and current measurements, and extensive simulation results confirm the approach’s robust performance across more than 55,000 diverse fault scenarios, covering all fault types, surpassing methods reported in the literature. The methodology consistently achieves high accuracy, exceeding 98% for noise-free signals and maintaining above 90% for noisy signals. These results demonstrate clear improvements in fault selectivity, supporting more secure protection for modern wind farms.