Enhancing PI control in microgrids using machine-learning techniques.

Area of Science:

• Electrical Engineering
• Computer Science
• Control Systems

Background:

• Renewable energy integration into power systems necessitates advanced control strategies for stability.
• Traditional Proportional-Integral (PI) controllers struggle with optimal parameter tuning for renewable energy sources (RES).
• Suboptimal PI gains can lead to microgrid instability and reduced performance.

Purpose of the Study:

• To develop and evaluate a Machine Learning (ML)-enhanced framework for microgrid control.
• To combine Artificial Neural Networks (ANNs) and Reinforcement Learning (RL) with PI controllers for improved performance.
• To address the parameter tuning challenges of PI controllers in microgrids with distributed energy resources (DERs).

Main Methods:

• Simulation of a microgrid with DERs using three control strategies: traditional PI, ANN-based PI, and RL-based PI.
• Dynamic adjustment of PI controller gains based on real-time operational data and historical performance.
• Evaluation of voltage Total Harmonic Distortion (THD), settling time, and frequency stability.

Main Results:

• The RL-based PI controller reduced voltage THD to 0.43% (vs. 16.99% for traditional PI).
• The ANN-based PI controller achieved 0.58% THD, a 96.6% improvement over conventional methods.
• ML-enhanced controllers improved settling time by 75% and frequency stability by 93%, exceeding IEEE 1547 standards.

Conclusions:

• ML and deep learning techniques significantly enhance microgrid stability and reliability.
• The proposed ML-enhanced framework offers practical solutions for advanced RES management.
• Dynamic gain adjustment using ANNs and RL overcomes limitations of traditional PI controllers.