Microgrid Fault Localization Using Voltage-Current Classification

The patent transforms the protection scheme from using fixed pre-defined thresholds to using dynamic, learned parameters through machine learning classification models. The system extracts multiple features (positive sequence, negative sequence, zero sequence components) and uses learned decision boundaries instead of static thresholds, allowing the parameters to adapt to varying system conditions with distributed energy resources.

The patent replaces the traditional mechanical/electrical threshold-based protection mechanism with an intelligent software-based classification system. Instead of relying on hardware-defined fixed thresholds, the system uses machine learning models (such as support vector machines, neural networks) to perform classification and decision-making, substituting the mechanical threshold comparison with computational intelligence.

The patent segments the complex protection problem into distinct feature extraction stages and classification stages. The system separately extracts positive sequence, negative sequence, and zero sequence features, then feeds these segmented features into classification models. This segmentation allows the complex task to be divided into manageable, modular components that can be implemented and maintained more easily.

The patent introduces feature extraction as an intermediary layer between raw measurements and final protection decisions. Instead of directly comparing raw signals to thresholds, the system first transforms measurements into meaningful features (sequence components, harmonics), then uses these intermediate representations for classification. This intermediary processing simplifies the overall system by creating a structured bridge between sensing and decision-making.