Hybrid AI and semiconductor approaches for power quality improvement

Area of Science:

• Electrical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Electric power quality is crucial for grid stability and reliability.
• Challenges include voltage sags, swells, harmonics, and transient disturbances.
• Traditional control methods struggle with complex, dynamic power quality issues.

Purpose of the Study:

• To develop and evaluate a novel approach for enhancing electric power quality using Machine Learning (ML) and Deep Learning (DL).
• To compare the performance of various ML and DL algorithms in identifying and forecasting power quality disturbances.
• To investigate the effectiveness of hybrid systems integrating traditional and data-driven control strategies.

Main Methods:

• Implementation of a data-driven framework combining traditional control with adaptive ML/DL models.
• Testing of algorithms including Support Vector Machines (SVM), Random Forests, Neural Networks, Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM).
• Utilizing real-time data for simulations and real-world experiments.

Main Results:

• Deep learning models, especially LSTM, demonstrated superior accuracy and dependability in power quality issue detection and forecasting.
• CNN achieved 91.8% precision, while LSTM reached 100% accuracy and 94.5% recall.
• Traditional ML models faced challenges with imbalanced datasets, leading to lower precision and recall compared to DL models.

Conclusions:

• Hybrid systems integrating traditional and data-driven control strategies show promise for adaptive and dependable power quality management.
• Deep learning models offer enhanced accuracy for complex power quality scenarios, essential for future smart grid applications.
• Practical deployment necessitates balancing computational demands and addressing data imbalance challenges.