Invention Reference Number
This technology introduces an unsupervised machine learning framework that automatically identifies and classifies power system faults without requiring labeled data. By leveraging unlabeled electrical event signals, the approach enables detection of a broader range of grid disturbances than traditional protection systems. This innovation enhances situational awareness and reliability for increasingly complex, power-electronics-based grids.
Description
Modern electrical grids experience diverse disturbances caused by weather, vegetation, and equipment failures, many of which go undetected by standard protective devices. This technology applies an advanced data-driven method based on a convolutional variational autoencoder (CVAE) architecture to address this limitation. The CVAE model learns underlying patterns in unlabeled electrical signal data, automatically clustering and categorizing distinct fault types. The system bridges the gap between simulated and real-world data, allowing utilities to analyze and classify fault events more comprehensively. The method operates on a reduced data representation that naturally groups similar disturbances together, enabling efficient, automated classification. This solution can be adapted to various grid configurations and sensor types, improving monitoring capabilities and accelerating operational response without extensive retraining or costly manual labeling.
Benefits
- Enables automatic detection and categorization of grid disturbances without labeled data
- Improves grid reliability, resilience, and operational efficiency
- Reduces costs and time associated with data labeling and system retraining
- Supports rapid response and predictive maintenance strategies
Applications and Industries
- Electric utilities and grid operators
- Critical infrastructure and energy security agencies
- Smart grid analytics and monitoring system developers
Contact
To learn more about this technology, email partnerships@ornl.gov or call 865-574-1051.
