Signal-piloted processing and machine learning based efficient power quality disturbances recognition.

Area of Science:

• Electrical Engineering
• Computer Science
• Data Science

Background:

• Power Quality Disturbances (PQDs) cause significant financial losses for smart grid stakeholders.
• Accurate and timely recognition and mitigation of PQDs are crucial for grid stability.
• Current time-invariant PQD identification methods lead to excessive data processing, power consumption, and latency.

Purpose of the Study:

• To develop an efficient machine learning-assisted system for PQD management.
• To reduce data storage, processing, and transmission requirements for PQD classification.
• To decrease computational cost and latency in identifying voltage and transient disturbances.

Main Methods:

• A novel approach combining signal-piloted acquisition, adaptive-rate segmentation, and time-domain feature extraction.
• Real-time data compression through signal-piloted acquisition and processing.
• Classification using robust machine learning algorithms including k-Nearest Neighbor, Naïve Bayes, Artificial Neural Network, and Support Vector Machine.

Main Results:

• Achieved a 6.75-fold reduction in collected information and processing load.
• Attained 98.05% classification accuracy for major voltage and transient disturbances.
• Demonstrated reduced computational cost and latency in the classification process.

Conclusions:

• The proposed signal-piloted acquisition and processing method significantly enhances PQD management efficiency.
• The machine learning-based approach offers a robust and accurate solution for automated disturbance recognition.
• This system effectively minimizes data handling burdens and improves smart grid operational performance.