Method for Power Grid Digital Operation Data Integration Based on K-Medoids Clustering with Support for Real-Time...

Area of Science:

• Electrical Engineering
• Data Science
• Computer Science

Background:

• Power grid digital operations face challenges with multisource heterogeneous data, leading to inefficient integration and slow anomaly detection.
• Existing methods struggle to effectively process and analyze diverse data streams from various sensors.

Purpose of the Study:

• To propose an efficient data integration method for power grid digital operations using K-medoids clustering.
• To enhance the speed and accuracy of anomaly detection in intelligent power grid environments.

Main Methods:

• Utilized a Field Programmable Gate Array (FPGA) parallel architecture for millisecond-level synchronous acquisition and preprocessing of multisource data (vibration, partial discharge, temperature).
• Implemented a K-medoids clustering algorithm with a density-weighted Euclidean distance metric and adaptive centroid selection in the application layer.
• Developed a cloud service layer for data filtering, analysis, and access, ensuring seamless data flow from basic to application layers.

Main Results:

• Achieved data throughput exceeding 110 MB/s from various power grid data sources.
• Attained a silhouette coefficient greater than 0.91 for integrated datasets, indicating high clustering performance and data reliability.
• Demonstrated effective acquisition and integration of multisource heterogeneous power grid data.

Conclusions:

• The proposed method significantly enhances the clustering performance of multisource data, enabling rapid anomaly detection.
• The architecture supports real-time processing and can be extended to cross-modal scenarios, improving power grid operation and maintenance management.
• This approach lays a foundation for timely decision-making in intelligent power grid operations.